Documente Academic
Documente Profesional
Documente Cultură
ABC
Editors
Prof. Dr.-Ing. Berend Denkena Dr.-Ing. Ferdinand Hollmann
Leibniz Universität Hannover German Research Foundation
Coordinator of Priority Program 1180 Program Director Engineering Science
Garbsen Bonn
Germany Germany
This book consists of the four parts “Basics”, “Grinding”, “Cutting” and “Form-
ing”. Part I “Basics” gives an overview of the applied and developed methods in
Measuring Technology, Modeling and Simulation and Mathematical Methods. The
following 3 parts “Grinding”, “Cutting” and “Forming” contain the main scientific
results and modeling approaches of all 20 research projects, covering a wide range
of topics, e. g. tool grinding, milling and deep drawing. Despite the large number
of different manufacturing methods investigated by the projects, some topics, such
as dynamic self-excitation of machine tools, known as “chatter vibration”, the static
deflection of machine tool parts due to the acting process forces or the influence of
thermal effects, were addressed by nearly all of the research projects.
Part I: Basics
Measurement and Test Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
E. Abele, J.C. Aurich, B.-A. Behrens, D. Biermann, C. Brecher,
E. Brinksmeier, M. Czora, B. Denkena, U. Engel, K. Großmann, U. Heisel,
D. Heinisch, R. Hermes, B. Kirsch, F. Klocke, A. Krause, T. Kroiß,
R. Laurischkat, M. Löser, F. Mahr, H. Meier, M. Pischan, P. Rasper,
A.V. Scheidler, M. Storchak, E. Uhlmann, M. Weiß
Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
C. Brecher, A. Bouabid, M. Deichmueller, B. Denkena, K. Großmann,
A. Hardtmann, D. Hömberg, R. Hermes, F. Klocke, M. Löser, O. Rott,
P. Steinmann, M. Weiß
Adaptive Finite Elements and Mathematical Optimization Methods . . . . 53
M. Andres, H. Blum, C. Brandt, C. Carstensen, P. Maaß, J. Niebsch,
A. Rademacher, R. Ramlau, A. Schröder, E.-P. Stephan, S. Wiedemann
Interaction Effects between Strip and Work Roll during Flat Rolling
Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
S. Puchhala, M. Franzke, G. Hirt
Increase of the Dimensional Accuracy of Sheet Metal Parts Utilizing a
Model-Based Path Planning for Robot-Based Incremental Forming . . . . 459
H. Meier, S. Reese, Y. Kiliclar, R. Laurischkat
Gear Rolling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
R. Neugebauer, U. Hellfritzsch, M. Lahl, M. Milbrandt, S. Schiller,
T. Druwe
Investigation of the Complex Interactions during Impulse Forming
of Tubular Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
Fr.-W. Bach, M. Kleiner, A.E. Tekkaya
Abele, E.
Technische Universität Darmstadt, Institute of Production Management,
Technology and Machine Tools
Andres, M.
Leibniz Universität Hannover, Institute for Applied Mathematics
Aurich, J. C.
University of Kaiserslautern, Institute for Manufacturing Technology
and Production Systems
Bach, F.-W.
Leibniz Universität Hannover, Institute of Material Science
Bauer, J.
Technische Universität Darmstadt, Institute of Production Management,
Technology and Machine Tools
Behrens, B.-A.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
Biermann, D.
Technische Universität Dortmund, Faculty of Mechanical Engineering,
Institute of Machining Technology
Blum, H.
Technische Universität Dortmund, Faculty of Mathematics (LS X)
Bouabid, A.
Universität Erlangen-Nürnberg, Chair of Applied Mechanics
Bouguecha, A.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
Brandt, C.
University of Bremen, Center for Industrial Mathematics
XII List of Authors
Brecher, C.
RWTH Aachen University, Laboratory of Machine Tools and Production
Engineering, Chair of Machine Tools
Brinksmeier, E.
University of Bremen, Foundation Institute of Materials Science, Laboratory of
Precision Machining
Britz, R.
Technische Universität München, Institute of Applied Mechanics,
Department of Mechanical Engineering
Carstensen, C.
Humboldt University of Berlin, Department of Mathematics
Czora, M.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
De Payrebrune, K.M.
Technische Universität Bergakademie Freiberg, Institute for Machine Elements,
Design and Manufacturing
Deichmueller, M.
Leibniz Universität Hannover, Institute of Production Engineering and Machine
Tools
Denkena, B.
Leibniz Universität Hannover, Institute of Production Engineering and Machine
Tools
Druwe, T.
Fraunhofer Institute for Machine Tools and Forming Technology
Eberhard, P.
Universität Stuttgart, Institute of Engineering and Computational Mechanics
Engel, U.
Universität Erlangen-Nürnberg, Chair of Manufacturing Technology
Esser, M.
RWTH Aachen University, Laboratory of Machine Tools and Production
Engineering
Franzke, M.
RWTH Aachen University, Institute of Metal Forming
Friedmann, M.
Technische Universität Darmstadt, Department of Computer Science, Simulation,
Systems Optimization and Robotics Group
Gaugele, T.
Universität Stuttgart, Institute of Engineering and Computational Mechanics
List of Authors XIII
Großmann, A.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Hardtmann, A.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Heinisch, D.
Leibniz Universität Hannover, Institute of Production Engineering and Machine
Tools
Heisel, U.
Universität Stuttgart, Institute for Machine Tools
Hellfritzsch, U.
Fraunhofer Institute for Machine Tools and Forming Technology
Hemker, T.
Technische Universität Darmstadt, Department of Computer Science, Simulation,
Systems Optimization and Robotics Group
Hermes, R.
RWTH Aachen University, Laboratory of Machine Tools and Production
Engineering, Chair of Machine Tools
Hirt, G.
RWTH Aachen University, Institute of Metal Forming
Hömberg, D.
Weierstrass Institute, Berlin
Kiliclar, Y.
RWTH Aachen University, Institute of Applied Mechanics
Kirsch, B.
University of Kaiserslautern, Institute for Manufacturing Technology and
Production Systems
Kleiner, M.
Technische Universität Dortmund, Institute of Forming Technology and Lightweight
Construction
Klocke, F.
RWTH Aachen University, Laboratory of Machine Tools and Production
Engineering, Chair of Manufacturing Technology
XIV List of Authors
Krause, A.
University of Bremen, Laboratory for Precision Machining
Kriechenbauer, S.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Krimm, R.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
Kröger, M.
Technische Universität Bergakademie Freiberg, Institute for Machine Elements,
Design and Manufacturing
Kroiß, T.
Universität Erlangen-Nürnberg, Chair of Manufacturing Technology
Lahl, M.
Fraunhofer Institute for Machine Tools and Forming Technology
Laurischkat, R.
Ruhr-University Bochum, Chair of Production Systems
Löser, M.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Maaß, P.
University of Bremen, Center for Industrial Mathematics
Mahr, F.
Technische Universität Berlin, Institute of Machine Tools and Factory Management
Maier, T.
Technische Universität München, Institute for Machine Tools and Industrial
Management, Department of Mechanical Engineering
Maischak, M.
Leibniz Universität Hannover, Institute for Applied Mathematics
Matthias, T.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
Meier, H.
Ruhr-University Bochum, Chair of Production Systems
Milbrandt, M.
Fraunhofer Institute for Machine Tools and Forming Technology
Neugebauer, R.
Fraunhofer Institute for Machine Tools and Forming Technology
List of Authors XV
Niebsch, J.
Johann Radon Institute for Computational and Applied Mathematics,
Austrian Academy of Sciences
Penter, L.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Pischan, M.
Technische Universität Darmstadt, Institute of Production Management,
Technology and Machine Tools
Puchhala, S.
RWTH Aachen University, Institute of Metal Forming
Rademacher, A.
Technische Universität Dortmund, Faculty of Mathematics (LS X)
Ramlau, R.
Johannes Kepler University Linz, Industrial Mathematics Institute
Rasper, P.
Technische Universität Berlin, Institute of Machine Tools and Factory Management
Reese, S.
RWTH Aachen University, Institute of Applied Mechanics
Reinl, C.
Technische Universität Darmstadt, Department of Computer Science, Simulation,
Systems Optimization and Robotics Group
Rott, O.
Weierstrass Institute, Berlin
Scheidler, A. V.
Technische Universität Dortmund, Faculty of Mechanical Engineering,
Institute of Machining Technology
Schiller, S.
Fraunhofer Institute for Machine Tools and Forming Technology
Schröder, A.
Humboldt University of Berlin, Department of Mathematics
Schwarz, F.
Technische Universität München, Institute for Machine Tools and Industrial
Management, Department of Mechanical Engineering
Shi, Y.
Technische Universität Berlin, Departments of Mechanics, Chair of Mechatronics
and Machine Dynamics
XVI List of Authors
Steinmann, P.
Universität Erlangen-Nürnberg, Chair of Applied Mechanics
Stephan, E. P.
Leibniz Universität Hannover, Institute for Applied Mathematics
Storchak, M.
Universität Stuttgart, Institute for Machine Tools
Tekkaya, A. E.
Technische Universität Dortmund, Institute of Forming Technology and Lightweight
Construction
Uhlmann, E.
Technische Universität Berlin, Institute of Machine Tools and Factory Management
Ulbrich, H.
Technische Universität München, Institute of Applied Mechanics,
Department of Mechanical Engineering
Vehmeyer, J.
University of Bremen, Center for Industrial Mathematics
von Stryk, O.
Technische Universität Darmstadt, Department of Computer Science, Simulation,
Systems Optimization and Robotics Group
von Wagner, U.
Technische Universität Berlin, Departments of Mechanics, Chair of Mechatronics
and Machine Dynamics
Wegner, H.
RWTH Aachen University, Laboratory for Machine Tools and Production
Engineering
Weinert, K.
Technische Universität Dortmund, Faculty of Mechanical Engineering,
Institute of Machining Technology
Weiß, M.
RWTH Aachen University, Laboratory for Machine Tools and Production
Engineering
Wiedemann, S.
Humboldt University of Berlin, Department of Mathematics
Wiemer, H.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Zaeh, M.
Technische Universität München, Institute for Machine Tools and Industrial
Management, Department of Mechanical Engineering
Part I
Basics
Chapter 1
Measurement and Test Techniques
Abstract. Nowadays, different measurement and test techniques are used to inves-
tigate the interaction between processes and machine tool structures. Machine and
workpiece properties are determined after analyzing the individual factors of
process metrology, which have an effect on the process. This chapter explains the
measurement methods for the structural analysis of the machine tool as well as for
manufacturing processes and for the workpiece analysis. In addition, an overview
of different measurement and test techniques based on selected examples related
to the priority program 1180 is given.
1.1 Introduction
This chapter analyzes the metrological possibilities in order to determine the inte-
raction between process and machine structure. The metrological analysis is sub-
divided into three main sections: Section 1.2 refers to the structural analysis of the
machine tool, section 1.3 to the process analysis and section 1.4 to the analysis of
the workpiece. In section 1.2 different measurement and test techniques of the
static and the dynamic machine tool behavior, the kinematics and the temperature
of machine tools are described. Section 1.3 describes different measurements and
test techniques for force, acoustic emission and temperature, whereas section 1.4
describes refuse to surface and geometry assessment. Here, a small imported out-
line can be obtained and its influence must be analyzed to understand the interac-
tion between process and machine tool structure for specific examples.
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 3–27.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
4 E. Abele et al.
to significantly increased process loads the demands concerning the working accu-
racy of the machine tool have increased as well. Thus, apart from performance
values like spindle power and speed or feed rate the structural and mechanical
properties of the machine tool have to be known in order to assess the accuracy
and productivity [1].
To analyze and improve the machine tool behavior it is essential to describe the
machine tool characteristics by defined parameters. Despite the good progress in
calculating machine tool parameters, the experimental determination of the struc-
tural properties is still essential for the parameterization and evaluation of machine
tool models. The accuracy of machine tools is primarily affected by deviations at
the tool-workpiece interface. Depending on the transmission behavior of the
machine tool thermal, static or dynamic loads result in kinematic and geometric
deviations from the desired working motions. Therefore, the thermal and dynamic
behavior limits the theoretical performance of the machine tool. Due to dynamic
instabilities and displacements caused by thermal dilatation the full potential of
machines cannot be exploited, which results in a negative impact on the productiv-
ity. To ensure the efficiency of machine tools the interactions between process and
structure have to be determined. In the following sections, different measurement
methods for the analysis of the static, dynamic and thermal structure, their beha-
vior as well as the kinematics are described.
δz
Ram Δz
3 ΔFz
4 5 6
2
1
Fz
Non-
Linear
Bolster plate linear
z
y
Force
x
Pressure pad transducer
Distance 1 -
Compensation device 6 Displacement transducers
plates
(Cardan calotte)
Fig. 1.2 Measurement setup according to the DIN 55189 and resulting displacement
development
6 E. Abele et al.
Fig. 1.3 Experimental setup (left) and typical results (right) of an articulated robot
The setup consists of a force measurement rod to apply and detect tensile and
compressive forces and laser distance sensors to measure the displacement of the
robot due to the force. The difference in the tensile and compressive cycles
indicates hysteresis. The backlash amounts to approximately 0.2 mm at the mea-
surement point x = 1,900 mm, y = 0 mm and z = 600 mm in the base coordinate
system. Furthermore the measured curves show a slight s-shape due to a non-
linear structural behavior. Using the least-squares method a linear slope can be
fitted into the measured curve. The gradient of the measured curve then indicates
the stiffness.
1 Measurement and Test Techniques 7
the machine tool structure and the exciter. The third subsystem consists of the
measurement technique for detecting the vibration response of the machine tool
structure. In addition to systems for direct measurement of the displacement, e.g.
inductive transducers or optical measurement techniques piezoelectric accelero-
meters are used as well.
The frequency response functions (FRF) represent the dynamic compliance beha-
vior of a machine tool structure in frequency domain. Also, process stability and
the occurrence of forced vibrations can be assessed on the basis of these mea-
surement data.
For the determination of the frequency response function Fast-Fourier-
Transformation-analyzers (FFT) are used. Therefore, the analog force and deflec-
tion/acceleration signals are sampled and digitalized. The sample rate determines
the frequency range and the number of samples defines the frequency resolution of
the analysis. To suppress high frequency disturbances analog force and deflection
signals must be filtered before the digitalization.
The sampled time signals can be weighted by a so-called window function to
avoid errors, which may occur when the signals are transformed into frequency
domain. The type of window function depends on the signal that has to be ana-
lyzed. Commonly used functions are transient window (impact force), exponential
window (response to an impact) and Hanning window. The transformation of the
weighted signals into frequency domain is carried out by a discrete Fourier Trans-
formation. The result of the transformation is a complex frequency spectrum,
which can be depicted either as real part and imaginary part, magnitude and phase
or as a Nyquist plot (Fig. 1.4). Separately from the depiction FRF describes the
frequency-dependent deflection answer of a mechanical system regarding the dy-
namic load acting on it.
Fig. 1.5 Geometry model with measurement locations of a horizontal milling machine
10 E. Abele et al.
modal model of the examined system the proper selection of excitation and re-
sponse locations is of particular importance. The mathematical models obtained
from modal analysis can also be used for the prediction of structure responses due
to exciting forces or for substructure coupling when a simple dynamic representa-
tion is more suitable than a complex finite element model.
Figure 1.5 shows a geometric model of a horizontal milling machine. Each
geometry point represents a measurement location in the machine tool. The struc-
ture was excited with an impulse hammer at several locations and the vibration
response was measured using tri-axial acceleration sensors. The obtained modal
model can be used, for example, to predict chatter vibrations or to identify struc-
tural instabilities.
The circularity test allows the determination of the accuracy of a circular path, in-
terpolated by a computer control unit. The deviations and vibrations can be traced
back to the control unit, the drives and the machine kinematics. The test can be
conducted using a double-ball-bar or a grid encoder. In the case of a grid encoder
a photoelectric sensor moves over a plate, which contains a very precise measur-
ing grid without contact. Using a double-ball-bar the circularity of the machine
tool is determined by a position-measuring system. The system is integrated in the
gauge. Performing the test with large radii gives information about the machine
geometry, whereas small radii are used for the evaluation of the feed drive dynam-
ics [1]. Afterwards, the measured path can be compared with the desired path.
Figure 1.6 presents the measuring setup with a grid encoder (left) and a double-
ball-bar (right).
1 Measurement and Test Techniques 11
Fig. 1.6 Measuring setup for the ball-bar test with a grid encoder (left) and a double-ball-
bar (right) [1]
Fig. 1.7 Procedure of the back step method (left); characteristic diagram and determined
parameters (right)
12 E. Abele et al.
An approach for the determination of the kinematics of a machine tool is the mea-
surement of the procedure movement of the machine axes. The measurement of
the procedure axis is briefly described on the basis of a face grinding machine
(Geibel & Hotz FS 635-Z CNC). The temporal response of the machine control
was examined for the input of a correcting variable. The velocity and the accelera-
tion of the machine table were measured. A laser interferometer system with a
scanning rate of fa = 20 Hz was used in order to avoid the influence of machine
vibrations on the velocity and acceleration measurements. The change of move-
ment between the interferometer and the retro reflector was measured with evalua-
tion software. [8]
For the investigations the strokes were measured by several sequential starting
points with constant, well-defined point distances at different workpiece veloci-
ties. The dependency of the workpiece velocity on the selected step size is shown
in Figure 1.8.
Fig. 1.8 Dependency of the workpiece velocity on the selected step size [8]
Thermocouples are available for different applications and then classified into
different types. For example, thermocouples of the type T have an accuracy of
± 0.5 °C for a measurement range between approx. - 200 °C and 300 °C. Further-
more, this type is capable of measuring temperatures in fluids such as in the lubri-
cating oil system [13]. In Figure 1.9, some types of temperature sensors are
shown.
For measuring the temperature of the main eccentric shaft of a press electrically
insulated thermocouples with screw threads are often applied as close as possible
to the shaft. This can be done by fixing the sensor directly to the bearing of the ec-
centric shaft within the press frame or the connecting rod. For measurements
which do not allow a screwing fixation of the thermocouples the sensors can be
fixed with a thermal conductance paste and adhesive tape [9]. The signals of the
thermocouples can be recorded with a PC including a measuring board. The mea-
suring board should be equipped with an internal cold-junction compensation,
which is required for thermocouples. Within this compensation the reference tem-
perature is simulated by means of an integrated transistor. The difference in tem-
perature between the junction and the measurement point induces an electrical
voltage. For the measurement of the oil temperature in larger containers electrical-
ly-shielded resistance thermometers can be applied [13].
14 E. Abele et al.
Fig. 1.9 Different types of temperature sensors for measuring the temperature of machine
tool components
This kind of load cell consists of a steel body – the sensing device – which acts as
a spring. On this body, strain gauges are applied as measuring devices. Thus, the
forces are converted into elastic deformation. A calibration permits the correlation
between the elastic strains and the applied force [14].
The basic effect of a strain gauge is the change of resistance in an electrical
conductor due to the effect of mechanical stress, discovered by Wheatstone and
Thomson [15]. This change of resistance in a single wire is very small. For that
purpose, metal strain gauges with “wound wires”, which form a kind of grid, have
been developed. For an efficient production the grids are manufactured by etched
foil technology nowadays. Apart from metal strain gauges there are other types of
electrical resistive strain gauges, e. g. semi-conductor and vapor-deposited strain
gauges [16].
Strain gauge-based force transducers can be used for static and dynamic mea-
surements and are available in a variety of scales with nominal forces from about
10 N up to 5 MN. This range can be necessary, for example, for the measurement
of forming forces in cold forging. However, with larger nominal loads the height
of the transducers increases noticeably up to about 180 mm at 5 MN. Via a mea-
suring amplifier and an analog digital converter the output signal can be recorded
and processed electronically.
measuring devices are applied on. In this case, the piezoelectric element itself is
exposed to the force and also emits the electric output signal. Piezoelectric measur-
ing devices possess a high stiffness and can be manufactured in a very compact de-
sign. Since they cover a large range of force magnitudes, which can be measured at
high sampling rates they are predestinated for the use in processes with interrupted
cutting conditions. The high resolution is also advantageous for measuring very
small cutting forces as they occur, e.g. in ultra precision cutting. Compared to force
transducers based on strain gauges, quartz load elements at a comparable measur-
ing range for forming processes have a more compact design and a higher stiffness.
However, piezoelectric force transducers are more expensive.
-2 -0,5
0 50 100 150 s 250 0 50 100 150 s 250
Time t Time t
Fig. 1.11 Measured forces before (left) and after post-processing (offset and drift removed,
low-pass filter applied)
18 E. Abele et al.
1.3.1.4 Force Measurement in Micro and High Speed Cutting (HSC) Milling
For the force measurement of processes with interrupted cutting conditions at high
spindle speeds some particularities have to be taken into consideration. Due to the
periodical tooth engagement the machine tool structure is excited by high dynami-
cal loads. Depending on spindle speed and number of cutting teeth the tooth pass-
ing frequency may achieve values of up to several kHz. Since the cutting force
does not match an ideal sine wave the force signal also contains the harmonics of
the tooth passing frequency. Thus, if the actual cutting force has to be measured, a
very high sample frequency must be chosen. Another problem occurs when the
tooth passing frequency exceeds the measurement device’s eigenfrequency.
Therefore, it is also very challenging to obtain time signals of the actual process
forces acting on the cutter (e. g. to compare them with simulated force signals).
However, in a lot of cases it is much easier and less error-prone to evaluate low-
pass filtered or averaged signals (e. g. to identify cutting force coefficients).
Dynamometers based on piezoelectric elements are the most suitable solution
in terms of the highest possible eigenfrequencies and of achievable measuring res-
olution. The measuring device is usually located between the machine table and
the workpiece. Solutions for force sensors integrated into the tool holder are also
available.
Measuring the forces in cold extrusion processes requires the consideration of the
following specific process characteristics: The high flow stress in cold forging due
to forming at room temperature leads to high forces even if the die dimensions are
small. In the full forward extrusion process investigated in the project “Optimiza-
tion of Tool and Process Design for the Cold Forging of Net-Shape Parts by Simu-
lation” (chapter 19), forces of about 200 kN occur with a die diameter of only
12 mm above and 6 mm below the die shoulder. Thus, force transducers with a
high load capacity are necessary. Because of the high forces the presses and tool-
ing systems in use have to be stiff and compact. As a consequence, only limited
space is available for the force measurement. In contrast, the quasi-static characte-
ristic of extrusion processes with a smooth force increase does not imply high de-
mands on the dynamic behavior of the load cells.
Figure 1.12 shows the measurement setup in a tooling system for lateral cold
extrusion. The use of quartz load washers is required due to the limited height in
the press. The closed-die tooling system will be used on a single-acting stroke-
controlled press. The press has to apply the closing force, via disc springs, and the
forming force at the same time. To be able to measure both forces two quartz load
washers 9091A from Kistler Instrumente GmbH (nominal force 1200 kN) with a
height of only 28 mm are used. The washer (1) on the top is for measuring the en-
tire press force, the one (2) between the top tool plate and the disc springs for the
closing force. Subtracting these forces from each other leads to the punch force.
With the presented tooling system interactions between process, tool and machine
in closed-die forging will be investigated in the above-mentioned project (see
chapter 19).
1 Measurement and Test Techniques 19
Fig. 1.12 Tooling system for closed-die lateral cold extrusion with two quartz load washers
Kistler 9091A (1 and 2) for measuring press force and closing force
In grinding, the major part of energy input is dissipated as heat. The heat flows in-
to the chips, the coolant, the environment, the workpiece and the grinding wheel.
The degree of thermal damage of the workpiece surface depends on the machining
parameters. Diffusion, chemical reaction and soft annealing appear in the contact
area as a result of thermal stresses. These effects depend exponentially on the heat
development [24]. The productivity is limited by the heat development at increas-
ing process parameters [25]. The quantification of these thermal effects in
grinding can be applied to specify the workpiece properties which are subsequent-
ly required in the application [26]. Based on this knowledge, simulations can be
implemented to determine adequate parameters for the NC-machine in advance.
One method for determining the transformation of the microstructure by heat input
is to examine micro-sections [27].
1 Measurement and Test Techniques 21
Fig. 1.13 a) Scheme of the setup for temperature measurement and b) workpiece with
welded thermocouples
22 E. Abele et al.
In this example, C45 was used as test material and the standard carbide plates
P20 of the company Walter AG were used as inserts. A value of 5° was selected
for the rake angle and a value of 8° for the clearance angle. Characteristic temper-
atures in the primary cutting zone and in the chip are shown in Figure 1.14 [36].
Regarding the starting point of the cutting process, the temperature signal for the
sensor position can be identified. Thus, the temperature in the cutting zones and in
the basic material can be determined. The signal shape and the amplitude during
cutting correspond to the position of the temperature sensor or the constantan leg
respectively in the different layers of the material.
Fig. 1.14 Characteristic signal profile in the primary cutting zone and in the chip
In practice, this method can only be applied for comparably large cutting depths to
achieve reasonable results and so as to be able to determine the position of the constan-
tan leg precisely. This method was used in [37] for the experimental investigation of
the temperatures in the primary and tertiary cutting zone and in the base material.
within the ISO Standard 11562 [38]. The waviness profile is smoothed and yields
long-wave information without roughness data. On the other hand, the roughness
profile consists only of short-wave information without waviness data. The com-
monly used characteristic values of the surface roughness are derived from the
roughness profile. These are the arithmetic mean surface roughness Ra, the ten-
point height Rz and the maximum roughness Rmax. Of course, there are more
roughness values available for various tasks having different specifications, which
will not be mentioned and explained here.
The arithmetic mean surface roughness Ra is an integral value over the whole
measuring length, which is also used in industrial application and in research.
When calculating Ra, peak values, which can be disadvantageous for some appli-
cations, are averaged. In these cases, the ten-point height Rz is more suitable. To
determine this value the whole measuring length is divided into five sections. In
each section, the maximum roughness, i. e. the difference of the maximum and
minimum height of the profile, is determined. The average value of these five val-
ues is called Rz. The maximum roughness Rmax is the difference of the maximum
and minimum height of the profile over the entire measuring length. It is a very
important value for technical applications such as sealing surfaces where peak
values as well as outliers can lead to difficulties.
Further methods for surface assessment are optical measurement techniques
like interferometry or white light interferometry. For the measurement a beam of
light is divided via a beamsplitter to illuminate two surfaces - workpiece and pla-
nar reference. The reflected beams from workpiece and reference are recombined
and directed to interfere at a detector. Depending on the phase shift of the measur-
ing and reference beam typical fringe patterns can be observed, e. g. parallel
fringes for tilted planar surfaces.
For white light interferometry (Fig. 1.15) the workpiece is scanned in vertical di-
rection so that the light reflected from the surface points interfering with the refer-
ence beam at certain heights will be recorded as surface data. The interferometric
measurement principle is only applicable for optical surfaces; it provides the same
roughness values as mentioned above but considers an area instead of a line profile.
Fig. 1.15 White Light Interferometer (left) and measured surface topography of a diamond
turned sample (right)
24 E. Abele et al.
The main advantages of optical measurements are the short measuring time
(only a few seconds) and, most importantly for optical surfaces, the absence of
any mechanical contact during the measurement. Therefore, the surface cannot be
damaged. Another advantage is the possibility of obtaining spatial structure in-
formation for the surface topography instead of only 2-dimensional information
(profilometry). One disadvantage is the low capability of measuring surfaces with
high slopes which can occur e.g. for spherical, aspherical, structured and free-
formed surfaces. If the slopes exceed a certain angle (typically > 15° to 25°) de-
pending on the numerical aperture of the applied objective, the reflected light
will not enter the microscope objective. For these areas, no measurement data is
recorded.
The workpiece is scanned from several different positions during the exposures.
After the scanning process is finished the individual exposures are coarsely-
merged to a 3-dimensional general view. This happens by means of characteristic
features on the surface. Afterwards, a precision alignment algorithm is used. The
calculated 3-D model can be exported. The scanner as well as the scanned part can
be seen in Figure 1.17.
1.5 Conclusion
This chapter presents a general overview of the state of the art for different mea-
surements and test techniques, which are used for the analysis of the interaction
between process and machine tool structure. It has been shown that, depending
upon a special process and machine tool, different measurement and test
26 E. Abele et al.
References
[1] Weck, M.: Werkzeugmaschinen 5: Messtechnische Untersuchung und Beurteilung,
dynamische Stabilität. Springer, Berlin (2006)
[2] DIN55189, Ermittlung von Kennwerten für Pressen der Blechverarbeitung bei sta-
tischer Belastung. Beuth Verlag (1988)
[3] Schindler, J.: Experimentelle Strukturanalyse im Werkzeugmaschinenbau. Antriebs-
technik 34(5), 30 (1995)
[4] Weck, M.: Dynamisches Verhalten spanender Werkzeugmaschinen. Einflussgrößen,
Beurteilungsverfahren, Meßtechnik. Aachen (1971)
[5] He, J., Fu, Z.-F.: Modal Analysis. Butterworth-Heinemann, Oxford (2001)
[6] Ewins, D.J.: Modal Testing: Theory, Practice and Application, 2nd edn. Research
Studies Press Ltd., Hertford-shire (2000)
[7] Heylen, W., Lammens, S., Sas, P.: Modal Analysis Theory and Testing. Katholieke
Universiteit Leuven (2007)
[8] Jansen, T.: Entwicklung einer Simulation für den NC-Formschleifprozess mit To-
russchleifscheiben. Dissertation, Technische Universität Dortmund, Essen, Band 43
(2007)
[9] Derenthal, M.-J.: Bewertung und Optimierung der thermischen Eigenschaften schnel-
llaufender Umformanlagen. Dissertation, Leibniz Universität Hannover (2009)
[10] DIN/ISO 10791-10. Test conditions for machining centres – Part 10: Evaluation of
thermal distortion, Genf. (2007)
[11] ISO 230-3, Test code for machine tools- Part 3: Determination of thermal effects,
Genf. (2007)
[12] Behrens, B.-A., Reuß, C., Vieregge, T.: Thermografie zur Gesenküberwachung.
Schmiede-Journal, Ausgabe, 33–36 (September 2009)
[13] Bernhard, F.: Technische Temperaturmessung. Springer, Berlin (2003) ISBN:
3540626727, 1. Auflage
[14] Rohrbach, C.: Handbuch für elektrisches Messen mechanischer Größen, p. 498. VDI-
Verlag, Düsseldorf (1967)
[15] Thomson, W.: On the Electro-dynamic Qualities of Metals. Philosophical Transac-
tions of the Royal Society of London (1856)
[16] Hoffmann, K.: An Introduction to Measurements using Strain Gauges. Hottinger
Baldwin Messtechnik GmbH, Darmstadt (1989)
[17] Tichy, J., Gautschi, G.: Piezoelektrische Messtechnik, Physikalische Grundlagen,
Kraft-, Druck- und Beschleunigungsaufnehmer, Verstärker. Springer, Heidelberg
(1980)
[18] Klocke, F.: Process Design. Manufacturing Processes 2 – Grinding, Honing, Lapping,
pp. 251–287. Springer, Berlin (2009)
[19] Rosetto, S., Koch, U.: On Investigation of Temperature Distribution on Tool Flank
Surface. CIRP Annals 19(3), 551–557 (1971)
[20] Müller, B.: Thermische Analyse des Zerspanens metallischer Werkstoffe bei hohen
Schnittgeschwindigkeiten. Dissertation. RWTH Aachen (2004)
[21] Vieregge, G.: Zerspanung der Eisenwerkstoffe. Verlag Stahleisen M.B.H., Düsseldorf
(1970)
1 Measurement and Test Techniques 27
[22] De Witt, D.P., Nutter, G.D.: Theory and Practice of Radiation Thermometry. Wiley,
New York (1988)
[23] Kitagawa, T., Kubo, A., Maekawa, K.: Temperature and Wear of Cutting Tools in
High-speed Machining of Inconel 718 and Ti-6Al-6V-2Sn. Wear 202, 142–148
(1997)
[24] Kim, N.K., Gou, C., Malkin, S.: Heat Flux Distribution and Energy partition in
Creep-feed grinding. CIRP Annals 46(1), 227–232 (1997)
[25] Lavine, A.S., Malkin, S., Jin, T.C.: Thermal aspects of grinding with CBN wheels.
CIRP Annals 38(1), 557–560 (1989)
[26] Fischbacher, M.: Schleifen - Möglichkeiten zur Beherrschung der Prozesswärme.
IDR 42(I), 44–47 (2007)
[27] Büttner, A.: Das Schleifen sprödharter Werkstoffe mit Diamant-Topfscheiben unter
besonderer Berücksichtigung des Tiefschleifens. Dissertation. Technische Universität
Hannover, Hannover (1968)
[28] Davies, M.A., Ueda, T., M’Saoubi, R., Mullany, B., Cooke, A.L.: On The Measure-
ment of Temperature in Material Removal Processes. CIRP Annals 56(2), 581–604
(2007)
[29] Shen, B., Xiao, G., Guo, C., Malkin, S., Shih, A.J.: Thermocouple Fixation Method
for Grinding Temperature Measurement. Journal of Manufacturing Science and Engi-
neering 139, 0510141–0510148 (2008)
[30] Hoffmeister, H.-W., Maiz, K.: Wärmemanagement beim Flachprofilschleifen. In:
Hoffmeister, H.-W., Denkena, B. (eds.) Jahrbuch Schleifen, Honen, Läppen und Po-
lieren, vol. 61, pp. 17–26. Vulkan-Verlag, Essen (2004)
[31] Hoffmeister, H.-W., Machanova, I., Maiz, K.: Simulation of Grinding Processes with
FEA. In: CIRP International Workshop on Modelling of Machining Operations, pp.
329–333 (2005)
[32] Hwang, J., Kompella, S., Chandrasekar, S., Farris, T.N.: Measurement of Tempera-
ture Field in Surface Grinding Using Infra-Red (IR) Imaging Systems. Journal of Tri-
bology 125, 377–383 (2003)
[33] Brinksmeier, E., Heinzel, C., Meyer, L.: Development and Application of a Wheel
Based Process Monitoring System in Grinding. CIRP Annals 54(1), 301–304 (2005)
[34] Körtvelyessy, L.V.: Thermoelement Praxis. Vulkan Verlag, Essen (1981)
[35] Frohmüller, R., Knoche, H.-J., Lierath, F.: Aufbau und Erprobung von Temperatur-
messeinrichtungen durch das IFQ im Rahmen des Schwerpunktprogramms Spanen
Metallischer Werkstoffe mit hoher Geschwindigkeit. In: Spanen Metallischer
Werkstoffe Mit Hohen Geschwindigkeiten Kolloquium des Schwerpunktprogramms
der DFG, pp. 108–115 (1999)
[36] Zacher, M.: Integration eines optischen 3D-Sensors in ein Koordinatenmessgerät für
die Digitalisierung komplexer Oberflächen. RWTH Aachen, Dissertation (2004)
[37] Heisel, U., Storchak, M., Stehle, T., Korotkih, M.: Temperaturbestimmung in den
Zerspanzonen. wt Werkstattstechnik Online 100(5), 365–370 (2010)
[38] EN ISO 11562, Geometrical Product Specifications (GPS) – Surface texture: Profile
method – Metrological characteristics of phase correct filters (1998)
[39] Abele, E., Bauer, J.: Kamerabasierte Bahnkorrektur für das Fräsen mit Industrierobo-
tern. wt Werkstatttechnik Online, 9 (2010)
[40] Brecher, C., Denkena, B., Grossmann, K., Steinmann, P., Bouabid, A., Heinisch, D.,
Hermes, R., Löser, M.: Identification of Weak Spots in the Metrological Investigation
of Dynamic Machine Behaviour. Production Engineering – Research and Develop-
ment 9(6), 679–689 (2011)
Chapter 2
Modeling and Simulation
Abstract. One focus of the Priority Program 1180 is the prediction of process
machine interactions. The investigated manufacturing processes as well as the ma-
chine tool behavior and the physical phenomena vary within the projects of this
program. So depending on the issues that were investigated, the modeling ap-
proach that is best suitable for the specific problem has to be applied. To predict
the interactions these models have to be coupled and simulated. Besides the mod-
eling approaches different simulation techniques have also been applied. This
chapter gives an overview of the applied models of the structural machine beha-
vior and the manufacturing processes, the coupling of these models as well as the
simulation techniques that were used.
2.1 Introduction
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 29–51.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
30 C. Brecher et al.
Fig. 2.1 Principle of flexible and moveable flexible multi-body simulation [3]
X0 describes the geometry of the spinning wheel at the idle state and does not de-
pend on the time t. X is time-dependent since the position of a material point
changes due to the rotation, regardless of whether the angular velocity ω is con-
stant or not.
32 C. Brecher et al.
Finally, the deformation of the spinning wheel is given at the deformed (or to-
tal) configuration Bt by the function ϕ0 or ϕ, depending on the choice of the con-
figuration one takes as reference, i. e. B0 or Br respectively. Thus, the coordinates
x of a material point in the total configuration can be given by
x = ϕ0 ( X 0 , t ) = ϕ ( X(t ) , t ) = ϕ ( R(t ) ⋅ X 0 , t ) (2.2)
The total deformation ϕ depends on the time t in two ways: indirectly via X(t) and,
in addition, in an explicit way via t. The dependency on X(t) includes the fact that
the less important part of the kinematics, i. e. the rigid body rotation, is anticipated
in the model formulation. The direct dependency on time t, however, expresses the
most important part of the kinematics, i. e. the transient deformation due to the
process forces. For the position vector x, a total and a partial time derivative x
and ∂ t x can be defined, depending on whether the rigid body rotation is included
or not.
x = ϕ 0 ( X 0 , t ) = ϕ ( X(t ) , t ) = ϕ ( R (t ) ⋅ X 0 , t )
Initial ϕ0 Deformed
Configuration B0 X0
Configuration Bt
x
X
R ⋅ X0 ϕ
Reference
Configuration Br
Fig. 2.2 Configurations corresponding to the ALE approach. In addition to the initial non-
deformed state and the total deformed state there is an intermediate rotated state one refers
to when solving the problem
The stress tensor can be defined in different ways. Relating a force increment
dF to a surface element da, both in the total configuration, results in the Cauchy or
true stress tensor σ. Taking, however, the surface element in the initial or rotated
configuration dA0 or dA leads to the Piola-stress tensor P0 or P respectively.
In the following, the basic relations of the finite element model of the spinning
wheel are given. All equations are written with respect to the rotated configuration
Br. The balance of momentum writes
Div P + b r = ρ r x (2.3)
where ρr is the material density. The external body forces br, caused by gravity for
example, are not important for the present presentation. The total acceleration x
2 Modeling and Simulation 33
can be determined by totally deriving the position vector x twice. After some deri-
vations and transformations one obtains the following relations:
Total acceleration: [
x = G : X ]
+ [ F ⋅ Ω + 2 L ]⋅ X
⊗X +a (2.5)
Herein, the following entities have been used:
Deformation gradient: F = ∂ Xϕ (2.6)
S=2
∂W λ
=
∂C 2
[ ]
J 2 − 1 C −1 + μ 1 − C −1 [ ] (2.15)
In Eq. (2.13)-(2.15), λ and μ designate the Lamé parameters. In addition, the fol-
lowing entities have been used:
∂S ∂ 2W
C=2 =4 (2.18)
∂C ∂C ⊗ ∂C
34 C. Brecher et al.
Now, when the spinning wheel problem is discretized first with respect to space
and then time and when substituting Eq. (2.3) for Eq. (2.4)-(2.12), one can show,
that the forces at an arbitrary node I can be written at time step tn+1 as:
[ ][ F ]
I h h dV
n +1 = − ρ r ∂ X N ⋅ X
FIspn ⋅X
h
Spinning forces: n +1 r (2.19)
Br
FIintn +1 = [F ]
⋅ S hn +1 ⋅ ∂ X N I dVr
h
Internal forces: n +1 (2.20)
Br
⋅ h
n +1 =
Coriolis-type forces: FIcor
Br
2 ρr NI Lhn+1 ⋅ X dVr (2.22)
where the notation (⋅) h means that after spatial discretization the corresponding
entity (⋅) has, of course, to be written in discretized form, i. e. as a function of the
value at node I, weighted by the element shape function N I. The local velocity and
acceleration at time step tn+1 can be written according to the Newmark-scheme as
γ γ γ
v n +1 = ∂ t d n +1 ≈ [d n +1 − d n ] + 1 − v n + Δt − Δt a n (2.23)
β Δt β 2β
1
a n +1 = ∂ tt2 d n +1 ≈
1
[d n+1 − d n ] − 1
vn − − 1 a n (2.24)
β Δt 2 β Δt 2β
Herein, Δt and dn+1 denote the time increment to be appropriately chosen and the
time-dependent nodal position vector to be solved for respectively. Also, it has
been assumed that the state of the wheel is known at time step tn. The parameters β
and γ appearing in Eq. (2.23)-(2.24) are the Newmark parameters to be chosen
such that
(β , γ ) = 1 , 1 (2.25)
4 2
Relations (2.19)-(2.22) in sum give the force residual vector
−R n +1 = Fnspn
+1 + Fn +1 + Fn +1 + Fn +1
int ine cor
(2.26)
[ h
IJ n + 1 = − ρ r ∂ X N ⋅ X
K spn ][ ∂ ]
N J ⋅ X h I dV r
I
Spinning part: X (2.27)
Br
n +1 = [∂ ]
⋅ S hn +1 ⋅ ∂ x N J I dVr
geo I
Geometric part: K IJ xN (2.28)
Br
Material part:
h I
[
IJ n +1 = Fn +1 ⋅ ∂ x N ⋅ C n +1 ⋅ ∂ x N ⋅ Fn +1 dVr
K mat h J h
][ ] t
(2.29)
Br
I 1
Inertia part: IJ n +1 =
K ine ρ
Br
r N
β Δt 2
N J I dVr
(2.30)
γ ⋅ h
IJ n +1 =
Coriolis-type part: K cor
Br
ρr N I
2
β Δt
∂ x N I ⋅ X I dVr
(2.31)
Herein, I denotes the identity matrix. By summing all stiffness parts, the total
stiffness results in
K n +1 = K nspn
+ 1 + K n + 1 + K n + 1 + K n + 1 + K n +1
geo mat ine cor
(2.32)
which again results together with the force residual R in the Newton-Raphson step
K n + 1 ⋅ Δd n + 1 = R n + 1 (2.33)
where Δdn+1 denotes the iterative increment of the nodal displacement vector.
Once the nodal position vector dn+1 is known all other entities, e. g. the stress or
the strain tensor, can be determined.
G ( jω ) = −m ω
1
(2.34)
n n
2
+ jωd n + c n
with the modal mass m, the damping d and the stiffness c. The parameters of these
models can be identified from frequency response functions measured at the real
36 C. Brecher et al.
machine structure. For this, commercial software for a modal analysis or self-
written routines can be used. In many cases, the fitting between the dynamic be-
havior of the real structure and the model is better than the one that can be
achieved by structure based models, especially in the case of multi-degree of free-
dom behavior of the machine, see chapter 1. Since the abstract models contain
only a few degrees of freedom the computation time for simulations is much
shorter than when structure-based models – like FE or MBS models - are applied.
A disadvantage of the analogous models is the loss of information about the
structure of the machine. These abstract models are only valid for a given configu-
ration or state of the machine. Therefore, a change of local parameters or a change
of the position of the axes can only be considered by a new set of modal parame-
ters. A structure-based model can help to identify local parameters of the real
structure that cannot be measured directly. An analogous model is not appropriate
for such a task.
e.g. by Fang et al., e.g. by Kienzle e.g. by Stèphán, Feng e.g. by Altintas
Ernst & Merchant
b⋅h −
Fi = b ⋅ K i1.1 ⋅ h1 mi Fi = b ⋅ K i ⋅ h i + b ⋅ K ie Fi = K ic ⋅ b ⋅ h + b ⋅ K ie
x
Fs = τ s ⋅
sin φc
Ki1.1=f(vc, WZ / WST) Ki=f(vc, WZ / WST) Kic=f(vc, WZ / WST)
1-mi=f(vc, WZ / WST) xi=f(vc, WZ / WST) Kie=f(vc, WZ / WST)
αc
F F F
Cutting Tool
φc
Workpiece
h h h
Especially for the simulation of process stability linear models and models
based on exponential functions have been established to estimate the process
forces. Using such simulation models emerging friction between the workpiece
and the cutting edge can be captured by appropriate additional terms [7, 11].
These linear models assume a linear relationship between process forces and chip
thickness. There are different approaches for this purpose, e. g. by Tlusty [12],
Altintas [13] and Weck [14].
Models based on exponential functions derive process forces out of a non-linear
relationship to the chip thickness. Appropriate models for the description of the
processes have been developed, for example by Stepan [15] and Feng [16].
Investigations have shown that the feed motion-dependence of the dynamic force
variation may explain differences in process stability, which were also partially
observed in practice. The parameterization of the shown force models is
carried out on the basis of cutting tests and therefore valid only for one defined
cutting edge-workpiece material combination. In some cases, the parameterization
can be adapted to different experimental conditions with the help of adjusting
factors [17].
Especially for the depiction of chip formation, burr formation and the chip tem-
perature the cutting simulation using the finite element method is an important
tool [18]. In such a simulation tool, the chip formation is discretized for a small
surrounding between the workpiece and the cutting tool by finite elements in a
sufficiently small mesh size. These simulations are not linear since large dis-
placements and deformations, temperature and strain rate-dependent plasticity of
the workpiece at the cutting point as well as contact between tool and workpiece
must be taken into account. The results of the force calculation using FEM do not
provide sufficiently accurate results for the passive force in particular.
38 C. Brecher et al.
For mapping the cutting forces in order to simulate the process stability of a
machining process in most cases the afore-mentioned analytical models are used.
However, the problem is modeling the dynamic cutting forces using data from sta-
tionary chipping processes. Fig. 2.4 shows the geometric engagement of a cutting
edge for dynamic cutting conditions in one plane. The rapid change of the oscillat-
ing engagement conditions leads to digressive results in cutting force calculation.
Regardless of the approach of modeling process forces (mechanistic, empirical)
a multiplicity of research works have demonstrated, that dynamic effects have to
be taken into account additionally for stability simulation in the case of improving
process models. These models consider the inner chip thickness modulation
(wave-cutting) and the outer chip thickness modulation (wave-removing) as an
independent input parameter as well as a phase shift between the dynamic cutting
force and chip thickness modulation.
Fc
In particular, the variation of the position of the shear plane is quite often used
for the determination of the dynamic cutting forces from existing static models.
Kim and Lee [19], for example, give an analytical description of the shear angle
regarding the inner and outer chip modulation. This dynamic force model
considers the transfer function of the tool and its derivation in the direction of
cutting speed and direction of the chip thickness variation. Although the
mentioned models can be used for stability simulation they have not been applied
for this purpose so far.
Van Brussel [20] and Werntze [21] propose simpler, empirical models. These
take into account the influence of the inner and outer chip thickness modulation in
terms of a proportional relation and a phase shift. The phase shift allows the
modeling of a time lag between changes in chip thickness and the corresponding
changes in force.
2 Modeling and Simulation 39
Table 2.1 Methods for the determination of stresses, forces and energy: dxi - infinitesimal
small dimension, kf – yield stress, V – forming volume, η – deformation efficiency, W –
deformation energy, Fm – temporal mean value of deformation force F, σc, τc – contact
stresses, σm – temporal and local mean value of normal stress [32]
of dxi
Method of deformation energy 0 kf, V, η σm, Fm, W
assumptions
The method of deformation energy is the most elementary method to calculate the
deformation forces. This method is based on the law of energy conservation with
the applied outer energy
Wa = F ⋅ s (2.35)
In this formula, kfm is the mean value of yield stress and ϕ the equivalent plastic
strain. The inner energy is equivalent to the total deformation energy
Wtot = Wid + WFR + WSh + WB , (2.37)
summarizing the expressions for ideal, frictional, shearing and bending energy.
The ratio of ideal deformation energy and total deformation energy is referred to
as the deformation efficiency
Wid
η= . (2.38)
Wtot
1
FExt Wa = FExt ⋅ s u , Wi = k fm V ⋅ φ
η
su A0
with V = A 0 s u and φ = ln
h0 d 0 (A0) A1
1 A
Wa = Wi : FExt ⋅ s u = k fm A 0 s u ⋅ ln 0
η A1
1 A0
FExt = k fm A 0 ⋅ ln (2.39)
η A1
d 1 (A1 )
1
Wa = FD , max ⋅ s u , Wi = k fm V ⋅ φ
η
d0 dp d1
with V = π d 1s0 ⋅ su and φ = ln
d0
d1
FBH FBH
FD Wa = Wi :
s0 1 d
su FD ,max ⋅ s u = η k fm πd1 ⋅ s 0 ⋅ s u ⋅ ln d 0
1
rM
1 d1
FD ,max = k fm π d 1s 0 ⋅ ln (2.40)
η d0
Fig. 2.7 Deep drawing of a cylindrical cup
In both examples, the friction, shear and bending behavior are summarized in
the deformation efficiency factor. More detailed approaches considering the fric-
tion between die and workpiece result in the following formulas, calculating the
maximal forming forces in solid forward extrusion [34]
2 μ A0
FExt = k fm A 0 ⋅ tan α + 1 + ln + π ⋅ d0 ⋅ kf 0 ⋅ μ ⋅ h0 (2.41)
3 sin 2α A1
μπ d 2 ⋅ μFBH s
FD,max = πd m s 0 e 2 1.15k fmI ⋅ ln p + + k fmII 0 . (2.42)
d m πd m ⋅ s 0 2rM
This includes the blankholder force FBH, the mean wall diameter d m = d1 + s 0 and
the outside diameter of the flange when the drawing force has achieved a maxi-
mum value d p = 0.7d 02 + 0.3d 2m .
The presented elementary methods provide a solution for a set of problems
without substantial effort. This is achieved with simplifying assumptions deviating
from the real behavior. If the simplifying assumptions do not meet the require-
ments of accuracy, the Finite-Element-Method (FEM) could be used, however,
with a higher effort and costs. Thus, the following factors can be considered in an
improved way:
• different material characteristics,
• complicated geometry of the parts as well as
44 C. Brecher et al.
In general, independent representations for structure and process are the starting
point for the development of models describing the process-structure interaction.
Depending on the properties of interest engineers can choose among a large varie-
ty of structure models. Mathematically, structure models are understood as equa-
tions of motion corresponding to
• a complex multi-body system,
• an abstract system with multiple degrees of freedom describing the dynamics of
a selected point of a large structure,
• a system of partial differential equations discretized with finite elements in
space,
• a coupled system of ordinary and discretized partial differential equations.
The above-mentioned models can be summarized by the following system of dif-
ferential (algebraic) equations
( )
M (t , q, p M )q = f t , q, q , p f , (2.43)
with pM, pf denoting parameter vectors, which are usually determined fitting struc-
ture simulation results to the corresponding experimental data. The vector q
represents the general set of coordinates. If (2.43) has been directly derived from
modal analysis data, the mass matrix is usually constant and the right hand side is
a linear function of q and q .
Process models relate a geometrical parameter vector g and the relative velocity
v between tool and workpiece to the cutting force vector F acting on the tool. In
milling, the geometrical parameter vector consists of cutting width b and uncut
chip thickness h. A large class of process models can be expressed in terms of the
force they exhibit
~
F = F ( g , v, p g ) , (2.44)
ably from the desired values regulated by the machine control system. However, if
the cutting forces become large, increasing machine structure oscillations induce
larger variations of the geometry parameter vector and the relative velocity. This,
in turn, leads to undesired cutting force variations, which possibly cause a further
increase of the machine structure oscillations. In order to describe these process
structure interaction phenomena both aspects have to be incorporated into the
model equations. The variation of the geometrical parameter can be formulated as
a non-linear functional involving the history of the state variable arising from the
structure model (2.43), i. e.
t
g = [t , q(.)] = g~ (s, q(s ))ds .
(2.45)
t0
The relative velocity between tool and workpiece is usually also a non-linear func-
tion depending on the state variable and the corresponding time derivative, i. e.
v=~ v (t , q, q ) (2.46)
Moreover, an additional term appearing on the right hand side of (2.43) has to be
introduced to model the effect of the varying process forces arising from (2.44)
and (2.45), (2.46). Since the forces usually act on the tool or, in terms of the struc-
ture model, on an element representing the tool an additional model equation has
to be developed to get the appropriate contribution to the right hand side of (2.43).
In a general setting, such an expression is given by
f F = f F (t , q, F ) .
~
(2.47)
Thus, the modified structure model involving the additional term due to the pres-
ence of process forces reads
( ) ~
M (t , q, p M )q = f t , q, q , p f + f F (t , q, F ) (2.48)
A simple system possessing all the important features to model stability problems
in milling is illustrated in Fig. 2.8.
46 C. Brecher et al.
Ff
fz
ϕ h
dx Fc
m
kx
y
ky dy
x
For the present example, the equation of motion representing the machine struc-
ture reduces to the system
m 0 x dx 0 x k x 0 x Fx
= − − + , (2.49)
0 m y 0 d y y 0 k y y Fy
The unknown model parameters are pM = m and pf = (dx,dy,kx,ky)T. Note that (2.49)
is a very simple form of (2.48). The cutting force model corresponding to (2) is a
linear function relating the cutting cross section h ap and the cutting forces acting
on the tip of the cutting edge, i. e.
Ff K f
= a p h . (2.50)
Fc K c
The parameter vector is pg = (Kf,Kc)T. The expression corresponding to (2.45), i. e.
the variation of the geometrical parameter, consists of a stationary part hstat and a
dynamic part hdyn. The dynamic part hdyn represents the modulation of the uncut
chip thickness due to the structure oscillations
h = hstat + hdyn = f z sin ϕ + ( x(t ) − x(t − τ )) sin ϕ + ( y(t ) − y(t − τ )) cosϕ (2.51)
Note that instead of h nonlinear models usually involve the positive part of the un-
cut chip thickness, i. e. h+ = max(h,0). Since the cutting force components in
(2.50) are given in the rotating reference frame of the cutter an additional trans-
formation has to be introduced to get the corresponding force components into the
global reference frame, i. e.
Fx sin ϕ cos ϕ F f
= −γ (ϕ ) , (2.52)
F
y − cos ϕ sin ϕ Fc
2 Modeling and Simulation 47
with γ: [0, 2 π ] {0,1} denoting a function that switches from one to zero, if the
corresponding tooth is not cutting. Again, equation (2.52) can be interpreted as a
simple version of (2.47). Since the expressions for the cutting forces, the uncut
chip thickness, and the transformation of the cutting forces are given explicitly
(2.52), (2.51), (2.50) and (2.49) can be summarized by the following first order
delay differential equation (DDE)
u (t ) = [A − C (t )]u (t ) + C (t )u (t − τ ) + bstat (t ) , (2.53)
where C(t) denotes a non-smooth τ-periodic matrix. The vector bstat(t) represents
the external forces related to the stationary uncut chip thickness hstat. The incorpo-
ration of process structure interaction effects leads to an additional state-dependent
term hdyn in the uncut chip thickness (2.51). Without this term the matrix C(t) va-
nishes and (2.53) reduces to an inhomogeneous linear ordinary differential equa-
tion (ODE) with constant coefficients. Simulating the solution of (2.53) with the
Matlab dde23 solver and comparing it to the corresponding solution v(t) of the
system with C(t) = 0 for two different values of ap reveals the additional benefit of
the interaction model.
0.9 60
0.8
50
0.7
0.6 40
|u−v|/|v|
|u−v|/|v|
0.5
30
0.4
0.3 20
0.2
10
0.1
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
time / [sec] time / [sec]
Fig. 2.9 Stable (a) and unstable (b) solutions of system (2.53)
In the stable case, the solution of (2.53) converges to the solution of the corres-
ponding ODE. Consequently, the relative difference of both solutions converges to
zero. Such an evolution of the relative difference is shown in Fig. 2.9(a). In the
unstable case, the solution of (2.53) does not converge to the solution of the cor-
responding ODE, which remains bounded for all times. Consequently the relative
difference of both solutions diverges, as illustrated by Fig. 2.9(b). In the experi-
ments, the evolution of the DDE-solution is called chatter. While the DDE model
properly reproduces the onset of chatter the ODE model has a bounded solution of
all positive values of ap and is thus not capable of reproducing chatter phenomena.
2.4.2 Simulation
In the previous sections, the different approaches of modeling of the machine and
the process behavior have been presented. The coupling of the models differs
for the different processes. This section gives a short overview of the simulation
48 C. Brecher et al.
techniques, which have been applied within the different projects of the priority
program.
Within the Priority Program 1180, the analysis of the process machine interac-
tions for cutting operations focuses on the coupling of process forces and the
deflection of the tool. As shown in the example of a simple milling model, the
coupling for cutting processes is given by a change of the intersection between
tool and workpiece:
• change of cutting geometry caused by the deflection of the tool/workpiece
• modulation of cutting forces caused by the changing cutting geometry
For the computation of the tool workpiece intersection the following approaches
have been applied:
• intersection between a discrete workpiece (e. g. Dexel model) and the envelope
of the tool (e. g. CSG model)
• intersection between a discrete workpiece and a micro-model of single grains as
well as a macro-model of the grinding wheel for abrasive processes
• analytical computation of the chip thickness, as shown in the example in the
preceding section
In sheet metal forming, the position and deformation of the tool represent the
boundary conditions for the FE simulations of the forming of the blank. The re-
sulting forces at the nodes of the blank model act on the tool causing deflections
and deformations. For the computation of static or dynamic interaction respective-
ly two principle methods have been applied.
• Offline coupling, i. e. iterative computation of process forces and deflections of
the ram
• Integration of the model of the machine in the FE model of the forming process
For the simulation of commercial software tools as well as in-house developments
were used. The coupling was carried out in the same solver or by coupled simulations.
• Exchange of forces and deflections between two simulators with a fixed time
step
• Exchange of forces and deflections between two simulators with a variable time
step
• Exchange of forces and deflections in the same tool (model integration, time
domain simulations)
• Mode-dependent exchange of forces and deflections in the same tool.
• Integration of a force model in the equations of motion. Taking the repetitive
tooth engagement into account leads to a system of delayed differential equations.
modeling techniques
machining results
stability
workpiece: geometry,
surface roughness
model coupling
offline coupling
FE-models code export math tools
FE-models time domain
+ model reduction frequency domain
2.5 Conclusion
An overview of the applied modeling approaches has been provided. It has been
demonstrated which approach is suitable under specific conditions regarding the
machine and process behavior. An overview of the coupling methods and the si-
mulation techniques has also been given. The choice as to which approach has to
be applied depends on the intension of the simulation of the process machine inte-
raction, i. e. the interaction phenomenon that is to be investigated. This chapter
has given only basic information on the principle methods. Detailed information
will be provided in the following chapters, showing the application of these ap-
proaches to the project specific problems.
50 C. Brecher et al.
References
[1] Brecher, C., Esser, M., Witt, S.: Interaction of manufacturing process and machine
tool. CIRP Annals – Manufacturing Technology 58, 588–607 (2009)
[2] Klocke, F., Brecher, C., Sitte, B., Weiß, M.: Analyse der dynamischen Wechselwir-
kungen bei Pendel- und Schnellhubschleifprozessen. In: Jahrbuch Schleifen, Honen,
Läppen und Polieren, vol. 64, Ausgabe, Essen (2010)
[3] Hoffmann, F.: Optimierung der dynamischen Bahngenauigkeit von Werkzeugma-
schinen mit der Mehrkörpersimulation. Dissertation, WZL of RWTH Aachen Univer-
sity (2008)
[4] Oden, J.T., Lin, T.L.: On the general rolling contact problem for finite deformations
of a viscoelastic cylinder. Computer Methods in Applied Mechanics and Engineer-
ing 57, 297–367 (1986)
[5] Tlusty, J.: Analysis of the State of Research in Cutting Dynamics. Annals of the
CIRP 27(2), 583–589 (1978)
[6] Radharamanan, R.: The Measurement of the Dynamic Cutting Coefficients and the
Analysis of Chatter Behaviour in Turning. Katholieke Universiteit Leuven, Disserta-
tion (1977)
[7] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. In: Annals of
the CIRP, vol. 53(2), pp. 619–642 (2004)
[8] Stephenson, D.A., Agapiou, J.S.: Metal Cutting. Theory and Practice. Marcel Dekker
Inc., New York (1996)
[9] Merchant, M.E.: Mechanics of the metal cutting process. Journal of Applied Phys-
ics 16, 318–325 (1945)
[10] Fang, N., Jaeahir, I.S.: Analytical Predictions and Experimental Validation of Cutting
Forces Ratio, Chip Thickness, and Chip Back-Flow Angle in Restricted Contact Ma-
chining Using the Universal Slip-Line Model. International Journal of Machine Tools
and Manufacture 42, 681–694 (2000)
[11] Clausen, M.: Zerspankraftprognose und -simulation für Dreh- und Fräsprozesse.
Hannover. Diss (2005)
[12] Tlusty, J., Ismail, F.: Basic Non-Linearity in Machining Chatter. Annals of the
CIRP 30(1), 299–304 (1981)
[13] Altintas, Y.: Modeling Approaches and Software for Predicting the Performance of
Milling Operations at MAL-UBC. Machining Science and Technology 4(4), 445–478
(2000)
[14] Weck, M., Brecher, C.: Werkzeugmaschinen. Automatisierung von Maschinen und
Anlagen; 6. Auflage. Springer (2006)
[15] Stépán, G., Insperger, T.: Stability of the Milling Process. Periodica Polytechnica Ser.
Mech. Eng. 44 (2000)
[16] Feng, H.-Y., Azeem, A., Wang, L.: Simplified and Efficient Calibration of a Mecha-
nistic Cutting Force Model for Ball-End Milling. International Journal of Machine
Tools and Manufacture 44, 214–268 (2004)
[17] Witt, S.: Integrierte Simulation von Maschine, Werkstück und spanendem Ferti-
gungsprozess. Dissertation. Aachen. RWTH Aachen, Werkzeugmaschinenlabor
RWTH Aachen (2007)
[18] Marsolek, J., Fleischer, J., Schmidt, C., Schermann, T.: Simulation von Zerspa-
nungsprozessen mit Abaqus. Tagungsband ABAQUS Benutzerkonferenz, Nürnberg
(2005)
2 Modeling and Simulation 51
[19] Kim, J.S., Lee, B.H.: An Analytical Model of Dynamic Cutting Forces in Chatter Vi-
bration. International Journal of Machine Tools and Manufacture 31(3), 371–381
(1991)
[20] van Brussel, H.: Dynamische Analyse van het Verspaningsproces. Katholieke Un-
iversiteit Leuven, Dissertation (1971)
[21] Werntze, G.: Dynamische Schnittkraftkoeffizienten. Bestimmung mit Hilfe des Digi-
talrechners und Berücksichtigung im mathematischen Modell zur Stabilitätsanalyse.
RWTH Aachen, Dissertation (1973)
[22] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hoffmeister, H.W., Peters, J.,
Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.: Advances in
Modeling and Simulation of Grinding Processes. Annals of the CIRP 55(2), 667–696
(2006)
[23] Tönshoff, H.K., Peters, J., Inasaki, T., Paul, T.: Modelling and Simulation of Grind-
ing Processes. Annals of the CIRP 41(2), 677–688 (1992)
[24] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and simulation of
process: machine interaction in grinding. Production Engineering 3(1), 111–120
[25] Mackerle, J.: Finite Element Analysis and Simulation of Machining: an Addendum. A
Bibliography (1996-2002). International Journal of Machine Tools & Manufac-
ture 43, 103–114 (2003)
[26] Hou, Z.B., Komanduri, R.: On the Mechanics of the Grinding Process-Part 1. Sto-
chastic Nature of the Grinding Process. International Journal of Machine Tools &
Manufacture 43, 1579–1593 (2003)
[27] Zitt, U.-R.: Modellierung und Simulation von Hochleistungsschleifprozessen. Uni-
versity of Kaiserslautern, Dissertation (1999)
[28] Chang, H.-C., Junz Wang, J.-J.: A stochastic grinding force model considering ran-
dom grit distribution. International Journal of Machine Tools and Manufacture 48,
1335–1344 (2008)
[29] Doman, D.A., Warkentin, A., Bauer, R.: Finite element modeling approaches in
grinding. International Journal of Machine Tools and Manufacture 49(2), 109–116
[30] Werner, K., Klocke, F., Brinksmeier, E.: Modelling and Simulation of Grinding
Processes. In: 1st European Conference on Grinding. Aachen, pp. S.8-1 – S.8-27
(2003)
[31] Denkena, B., Deichmueller, M., Kröger, M., Popp, K.M., Carstensen, C., Schroeder,
A., Wiedemann, S.: Geometrical Analysis of the Complex Contact Area for Modeling
the local Distribution of Process Forces in Tool Grinding. In: Proceedings of the 1st
International Conference on Process Machine Interaction, pp. 289–298 (2008)
[32] Voelkner, W.: Spannungs- Kraft- und Arbeitermittlung beim Umformen. Ferti-
gungstechnik und Betrieb 25(12), 739–743 (1975)
[33] Voelkner, W.: Ein Beitrag zur Streifen- und Gleitlinienmethode. Wissenschaftliche
Zeitschrift der TU Dresden 25(3), 613–618 (1976)
[34] Lange, K. (ed.): Handbook of Metal Forming. Society of Manufacturing Engineers,
Dearborn, MI (1994)
[35] Petzold, W., König, J., Eberlein, L. (eds.): Umform- und Zerteiltechnik. Lehrbrief
Tiefziehen und Drücken, Dresden (1981)
[36] Neugebauer, R. (ed.): Umform- und Zerteiltechnik. Verlag Wissenschaftliche Skrip-
ten, Chemnitz (2005)
Chapter 3
Adaptive Finite Elements and Mathematical
Optimization Methods
Abstract. This chapter focuses on the topics of mathematical research in the pri-
ority program and on the application of the results to engineering problems. First, a
posteriori estimates and adaptive methods based on them are presented for contact
problems. In detail, frictional contact problems using a linear elastic material law,
which are discretized by a dual-dual method, are considered. Using a different ap-
proach, adaptive methods are derived for elasto-plasticity with contact and friction.
Then results for contact problems involving inertial effects are introduced. Second,
parameter identification and inverse problems are considered. After an introduction
into the general problem of setting and solution techniques, the balancing of a rotat-
ing system is discussed as a prototypical industrial application.
3.1 Introduction
Modeling, simulation and optimization are typical steps in computer-aided man-
ufacturing. The mathematical topics addressed in this priority program are fo-
cussed on developing efficient simulation and optimization techniques as well as
on parameter-identification tasks based on inverse problems.
The basic ideas are introduced by using a well-known example. We consider
a heat conduction problem in a perfectly isolated bar: Let us assume that the bar
consists of an inhomogeneous material of conductivity c(x) and specific heat ζ (x),
the bar is heated at its left endpoint and the resulting temperature at the endpoint on
the right is measured. The real temperature distribution inside the bar is denoted by
u , which is a function of time t and a spatial variable x ∈ [0, 1]. The heat distribution
in the bar can be modeled by the one dimensional heat equation, i.e. u satisifes
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 53–77.
springerlink.com c Springer-Verlag Berlin Heidelberg 2013
54 M. Andres et al.
The heat inflow is described by the function q0 and the initial temperature by u0 .
First, we address the task of efficiently approximating the solution u for known
c, ζ , q0 and discuss the various error terms involved. In order to approximate u we
compute a discrete solution ukh using a space-time finite element method (FEM).
However, we are not able to determine the exact solution of the FEM on a computer
but get an approximate solution ũkh . For example, we only determine approximate
solutions of the linear systems.
Thus, three types of errors have to be considered: The modeling error u − u,
the discretization error u − ukh , and the numerical error ukh − ũkh . The modeling
error cannot be estimated using mathematical techniques, one has to validate the re-
sults by experiments instead. The numerical error is negligible, if suitable numerical
techniques are used. In the run-up to the computation, a priori error estimates of the
discretization error show the convergence of the discrete solution ukh to u as well as
the convergence rate. However, a posteriori error estimates provide an estimate of
the discretization error for a computed discrete solution. They also yield informa-
tion about the distribution of the error. Based on this information, adaptive FEMs
modify the scheme such that the discretization error is reduced optimally. Hence, the
efficiency of the simulation increases. In the Sections 3.2, 3.3 and 3.4, some results
concerning adaptive FEMs are presented for contact problems . The application of
these methods in the simulation of engineering problems is discussed in Chapters 6,
7 and 14.
Next, we consider the case, where the parameter c, which is allowed to vary spa-
tially, is not known at all or only a rough estimate is available. The mathematical
technique of parameter identification (inverse problems) provides a systematic ap-
proach to determine such model parameters. The most widely used approach takes
an a priori guess c1 , simulates the resulting data u1 (t, 1) and determines an update
by minimizing the deviation between the simulation and the measured data g(t)
− div σ (u) = f in Ω ,
u=0 on ΓD ,
σ (u) · n = t0 on ΓN ,
(3.3)
un ≤ g; σn ≤ 0; (un − g)σn = 0 on ΓC ,
if |σ t | < μ f |σn |, then ut = 0 on ΓC ,
if |σ t | = μ f |σn |, then ∃ s ≥ 0 : ut = −s σ t on ΓC ,
where the stress tensor σ (u) is connected to the strain tensor ε (u) via Hooke’s law
for linear elasticity. The volume body force f ∈ L2 (Ω ), the prescribed traction t0 ∈
1 1
H− 2 (ΓN ), the positive gap function g ∈ H 2 (ΓC ) and the friction coefficient μ f are
assumed to be given. To handle the Coulomb friction numerically it is approximated
by a sequence of contact problems using Tresca friction, as described by Nečas et
al. in [25]. For this reason, the friction law in (3.3) is replaced by the following
|σ t | ≤ F ; σ t · ut + F |ut | = 0 on ΓC , (3.4)
0
(−4, 1) ΓN t0 = 2 4
−800 (1− x2 + x16 )
ΓD Ω ΓD
ΓC g (4, −1)
The discrete system (3.5) can be solved by a nested Uzawa-type algorithm, see
Glowinski et al. [22] or Ekeland and Témam [20] for an explanation of Uzawa
algorithms. The choice of different mesh-sizes h, h̃ and ĥ within the discrete for-
mulation is necessary due to the discrete inf-sup conditions for the bilinear form
B((v, ξ , ψ , μt ), τ ) + dC,n (μn , τ ), on the one hand, and for the bilinear form q(μt , κ ),
on the other hand. They ensure the convergence of the nested Uzawa algorithm.
Within the algorithm, the new iterates for the normal displacement and the friction
force are projected onto the corresponding convex sets in each iteration step.
As a numerical example, we consider the domain Ω := [−4m, 4m] × [−1m, 1m]
with boundary Γ divided into the Dirichlet part ΓD := {−4m} × [−1m, 1m]∪{4m} ×
3 Adaptive Finite Elements and Mathematical Optimization Methods 57
−3
x 10
2 100
0 0
−2
−200
−4
−300
−6
−400
−8
−500
−10 −600
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
contact boundary ΓC contact boundary Γ
C
3 400
tangential stress [N/m ]
2
2
200
displacement [m]
0 0
−1 −200
−2
−400
−3
−4 −600
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
contact boundary Γ contact boundary Γ
C C
Fig. 3.2 Approximated solutions on the contact boundary with mesh-size h = 0.0078m cor-
responding to 4 million degrees of freedom.
[−1m, 1m], the Neumann part ΓN := [−4m, 4m] × {1m} and the contact part ΓC :=
[−4m, 4m] × {−1m}.
We choose Young’s modulus E := 200 000 N/m2 and shear modulus G :=
80 000 N/m2 . The friction coefficient is μ f = 0.5 and the volume body force is
set to zero. On the contact boundary, we assume the body Ω to come into contact
with a rigid foundation, which has a positive distance g = 0.01m to Ω . In Figure 3.1
we show the domain and the distribution of the boundary.
In Figure 3.2, we display normal displacement and normal stress and tangential
displacement and tangential stress respectively. Due to the large system (3.5), the
number of degrees of freedom grows fast using the h-version. Since our solution
algorithm uses a nested loop to solve the inequalities in (3.5) the solution time is
very sensitive to the number of unknowns. Therefore, we are interested in adaptive
methods based on error estimators, which refine the finite element mesh in those
regions of the domain, where the estimated error is large.
58 M. Andres et al.
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
Fig. 3.3 Meshes created during the adaptive refinement process using the maximal local
error.
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(a) After 1 adaptive iteration step
1
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(b) After 5 adaptive iteration steps
1
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(c) After 7 adaptive iteration steps
Fig. 3.4 Meshes created during the adaptive refinement process using the fixed fraction
method.
addition the boundary condition is changing in the corners. Moreover, the residuum
of the friction force (σh )t − νĥ F leads to a refinement of the elements near the
part of the contact boundary where the domain is in contact. Finally, the indicator
referring to the antisymmetric part of the stress tensor (σh − σhT ) leads to a refine-
ment of those elements near the two corners at the bottom of the domain. As in the
first case, the change of boundary conditions in the corners enforces this refinement
additionally.
The application of adaptive methods for the dual-dual formulation of a frictional
contact problem leads to a faster convergence of the algorithm. First, the two-
fold saddle-point structure of the formulation involves several Lagrange multipli-
ers some of them requiring a different mesh-size. For this reason even the coarsest
discretization exhibits a large number of degrees of freedom. Therefore, a uniform
mesh-refinement strategy can be applied ony few times, which may result in an
unsatisfactory accuracy of the solution in the contact zone. Second, the nested
Uzawa algorithm is very sensitive to the number of unknowns. Hence, applying
adaptive methods decreases the convergence time of the algorithm.
60 M. Andres et al.
cf. [13, 23, 24]. Here, j is the dissipation functional with j(Q) := σy (Q : Q)1/2 and
σy > 0 is the yield stress in uni-axial tension. The equivalent variational inequality
is to find W ∈ K := {Z = (v, P) ∈ W := V × Q | vn ≤ g on ΓC } so that
a((u, P), (v, Q)) := (σ (u, Q), ε (v) − Q)0 + (HP, Q)0,
Ψ (v, Q) := j(Q) dx + μ f |vt | ds,
Ω ΓC
(v, Q) := (f, v)0 + (t0 , v)0,ΓN .
Λ p := {μ ∈ Q | μ : μ ≤ 1},
Λn := {μ ∈ H −1/2(ΓC ) | ∀v ∈ H+ (ΓC ) : μ , v
≥ 0},
1/2
Λt := {μ ∈ L2 (ΓC ) | |μ | ≤ 1},
3 Adaptive Finite Elements and Mathematical Optimization Methods 61
1/2
with H+ (ΓC ) := {v ∈ H 1/2 (ΓC ) | v ≥ 0} and μ , v
denotes the usual dual pairing
of μ and v. We obtain a mixed discretization scheme by exchanging the space W
with the finite dimensional space Wh := Vh × Qh where Vh is the space of continu-
ous, piece-wise bilinear finite elements and Qh is the space of piecewise constant
finite elements on a quadrilateral mesh Th . Furthermore, we replace Λ p , Λn , and
Λt by sets Λ p,h , Λn,H , and Λt,H , which are constructed on the basis of piecewise
constant finite elements. The solution of the discrete mixed scheme is denoted by
(Wh , λ p,h , λn,H , λt,H ). The indeces h and H indicate mesh-sizes of the finite element
discretizations. To construct the set Λ p,h we apply the mesh, which is already used
to define Wh since the resulting inclusion Λ p,h ⊂ Qh guarantees the unique existence
of the discrete Lagrange multiplier λ p,h of the plastic variable. To ensure the unique
existence of the Lagrange multipliers λn,H and λt,H we have to make sure that a
discrete inf-sup condition is fulfilled. This can be done, for instance, by choosing
a coarser mesh-size H ≥ h defining their finite element discretizations, for details
see [38]. To include multiple yield surfaces we have to add some further plastic-
strain variables and dissipation functionals, which can again be resolved by extra
Lagrange multipliers into order to include them in the proposed mixed scheme. The
structure of the mixed formulation (3.9) remains the same. We refer to [16, 17, 40]
for more details.
In many standard discretization approaches for elasto-plasticity, the non-linearity
j is resolved by regularization, which allows for the application of Newton’s
method, cf. [17, 16]. A mixed discretization scheme, as discussed in this section,
results in a finite dimensional saddle-point problem, which can be solved, for in-
stance, by methods of quadratic programming or by simple Uzawa-type algorithms,
see [40, 23].
The main goal of our investigations is to derive a posteriori error estimates for
the proposed mixed scheme. For this purpose, we combine error estimates for mixed
schemes developed for problems in elasto-plasticity, on the one hand, and for contact
problems with friction, on the other hand. It is shown in [40] that the error of the
purely elasto-plastic problem without contact can be estimated by
where 1/2
η := ∑ h2K RK + ∑ h E RE
T ∈T E∈E 0 ∪E
To include contact with friction we have to add some additional terms accounting for
the geometrical and complimentary conditions as well as the frictional conditions,
cf. [36]. We then obtain
62 M. Andres et al.
where ( f )− := − max{0, − f (x)} denotes the negative part of a function f . The term
dev(σ (Wh )−HPh )− σy λ p,h 0 results from the use of quadrilaterals. It vanishes, if
dev(Wh ) ⊂ Qh holds. This is the case, if piece-wise linear finite elements on triangles
are applied. In this case, the estimates for purely elasto-plastic problems without
contact are similar to those for standard schemes, cf. [9, 15, 33].
In a numerical experiment, we apply the a posteriori error estimates within an
adaptive scheme, where we use the fixed-fraction strategy, as described in Section
3.2. We set Ω := [0m, 1m]2 and the material parameters E := 2, 500N/m2, ν := 0.25,
ϑ := 500N/m2, σy := 10N/m2 and μ f := 0.5. Furthermore, we define H := ϑ I
(a) After 3 adaptive iteration steps (b) After 5 adaptive iteration steps
Fig. 3.6 (a) the norm of the plastic strain, (b) norm of the Lagrange multiplier
where I, is the identity tensor. The outer forces are set to zero and the obstacle
function is given as φ (x) := (x − 1/2)2 + 0.98. Figure 3.5 shows meshes created by
successive adaptive refinements on an initial mesh with 256 elements. We observe
refinements towards the contact area on the upper part of the boundary and the lower
corners, where Dirichlet conditions change to Neumann conditions. In Figure 3.6,
the resulting plastic strain and the Lagrange multiplier for it are shown.
ρ ü − divσ (u) = f in I × Ω ,
u(0) = u0 ; u̇(0) = v0 in Ω ,
u = 0 on ΓD , (3.10)
σ (u) · n = 0 on ΓN ,
un ≤ g; σn ≤ 0; (un − g)σn = 0 on ΓC ,
with the initial displacement u0 and the initial velocity v0 . The weak problem for-
mulation is
(ρ ü, v − u) + (σ (u), ε (v − u)) ≥ (f, v − u) (3.11)
where the variational inequality has to hold for all v in the test space V and all t ∈ I.
Two discretization approaches are commonly used: The first one combines a fi-
nite difference method in time with a finite element method in space, see [18, 27].
64 M. Andres et al.
An alternative is given by space-time finite element methods [6, 29]. The temporal
discretization has to be carried out carefully because it is difficult to ensure energy
conservation and numerical stability.
For both approaches, we derive a posteriori error estimates and start our descrip-
tion with the finite difference/ finite element approach. There, Rothes method is
used, i. e. the temporal direction is discretized first. We obtain a sequence of spa-
tial continuous but temporally discrete problems after the time discretization. The
structure of these problems is similar to static contact problems. Consequently, the
same techniques are used for the spatial discretization and for a posteriori error es-
timation. We modify the estimators given in [37] and apply them to the dynamic
problems. They lead to an estimate, which can be evaluated easily and fast. How-
ever the error is only estimated in the energy norm and no temporal effects are taken
into account. The detailed derivation is found in [5, 7, 8]. Currently, we are extend-
ing this approach by goal-oriented error estimators, which are originally designed
for static contact problems (see [39]).
The second approach relies on a space-time Galerkin discretization of the dy-
namic contact problems. We estimate the error in a user-defined error-functional J,
for instance the von Mises equivalent stress in a certain region. By utilizing a so-
called dual problem, which represents the influence of a space-time point on the
error J(u) − J(uh ), the a posteriori error estimate is derived. It consists of the primal
and the dual residual weighted by the approximation error of the dual and the primal
problem, respectively, as well as a term representing the error in the contact stresses
weighted by the dual solution. We obtain information about the spatial and the tem-
poral distribution of the error. However, a discrete solution of the dual problem has
to be calculated, i. e. a linear wave is additionally solved. More details are presented
in [29].
The a posteriori error estimates provide information about the spatial and, in the
dual-weighted residual (DWR) approach, temporal distribution of the discretization
error. To make use of this characteristic we have to localize the estimates, i. e. a
value is assigned to each single mesh cell. Two different techniques can be applied:
Integration by parts, see, e. g., [1], and filtering [10]. By considering hyperbolic
problems, some technical additional steps are carried out to obtain appropriate re-
finement indicators, see [29]. For instance, a special treatment of the hanging de-
grees of freedom is necessary.
Once the spatial and temporal error distribution is known, we utilize adaptive
refinement strategies, which select the mesh cells for refinement. A well-known
strategy is the fixed fraction strategy (see Section 3.2). Apart from this one, we
employ the more complex optimal mesh strategy [34], which optimizes the error
over the effort.
Employing a dynamic mesh approach, i. e. the meshes change between the time
steps, special attention has to be paid to the additional errors due to the mesh
changes. The data transfer has to be carried out in the appropriate way prescribed
by the space-time bilinear form. Typically, we have to calculate L2 -projections. Fur-
thermore, the meshes of two time steps have to be connected closely. To ensure this,
we employ a mesh regularization strategy [29].
3 Adaptive Finite Elements and Mathematical Optimization Methods 65
Fig. 3.7 Schematic view of the bar and the rigid foundation
0.0055
eta
0.005
0.0045
0.004
k
0.0035
0.003
0.0025
0.002
0.0015
0.001
0 0.2 0.4 t 0.6 0.8 1
In the next paragraphs, we will compare the results of the special contact estimator
η and the DWR estimator for the unconstrained problem ηe as well as different
adaptive methods. We use a fixed spatial mesh with 4, 096 cells and perform five
adaptive refinement iterations to compare the temporal meshes. The temporal mesh
based on η is depicted in Figure 3.8. It uses larger time steps for small t and smaller
time steps for large t.
For the discussion of the spatial meshes, we choose a uniform decomposition of
the time interval I with M = 400 time steps. The spatial mesh created in the constant
mesh approach based on η is presented in Figure 3.9. The left end of the contact
zone and the left corners of Ω are well-resolved. In the left corners, we observe
stress singularities due to the geometry and the change of the boundary conditions.
66 M. Andres et al.
Fig. 3.9 Adaptive mesh resulting from the constant mesh approach
(e) m = 400
Fig. 3.10 Meshes created during the adaptive refinement process in the dynamic mesh
approach
7000
eta
6000
5000
Nm
4000
3000
2000
1000
0
0 0.1 0.2 0.3 0.4 0.5
t 0.6 0.7 0.8 0.9 1
In Figure 3.10, the adaptive meshes created in the dynamic mesh approach are
depicted. For m < 200, we mainly observe additional refinements in the left corners,
where the stress singularities occur. In the 200th time-step, the left boundary of the
contact zone and B are well-resolved. The additional refinements are more and more
concentrated in B for m > 200. In Figure 3.11, we can see that the number of mesh
cells increases for t > 0.4. It attains its maximum for t ≈ 0.7 and then decreases.
In Figure 3.12, the convergence behavior for different refinement techniques is
compared. More detailed information about the dynamic mesh approach based on
η is presented in Table 3.1, where
3 Adaptive Finite Elements and Mathematical Optimization Methods 67
0.1
CM eta
CM eta_e
DM eta
DM eta_e
Uniform
J(w)−J(wkh )
0.01
J(w)
0.001
1e-04
10000 100000 1e+06 1e+07 1e+08 1e+09
M
∑ Nm
m=1
Fig. 3.12 Convergence in the output functional J with respect to the total number of mesh
cells for different refinement techniques
The results in Table 3.1 substantiate the accuracy of the a posteriori error estimate.
From Figure 3.12, we deduce that the adaptive methods based on ηe as well as the
dynamic mesh approach based on η are very efficient. They lead to comparable re-
sults. However, the constant mesh approach based on η is less effective. The results
nearly correspond to the results based on uniform refinement.
M
M ∑ NM Erel Ieff e
Ieff
n=1
50 12800 3.725 · 10−2 2.692 2.708
96 42912 1.700 · 10−2 3.119 3.132
182 176552 6.339 · 10−3 2.294 2.300
352 725908 3.351 · 10−3 2.532 2.536
678 4177824 5.664 · 10−4 1.049 1.050
structures, feed rates and depth of cut. In real-life situations some of these parame-
ters are not known precisely or are acceptable only in a small range of manufacturing
conditions.
This general framework of parameter identification can be used for different tasks
such as
• model calibration (parametrization),
• determining optimal structure and process parameters for achieving a desired
result (optimal control),
• determining the causes, if a desired result is not achieved with a given parameter
setting (process monitoring and correction).
In addition, the mathematical formulation of the solution process typically adds
some artificial parameters, such as step sizes or regularization parameters for sta-
bilizing the numerical solvers. Hence, one needs to calibrate the system by deter-
mining optimal values for those parameters. This becomes a challenging task, if
those parameters cannot be measured directly. Hence, they have to be estimated
from indirect measurements, e. g. the determination of damping parameters from
vibration measurements of a mechanical structure. This is the typical task of in-
verse problems: determining an unknown parameter from indirect measurements.
The forward mapping, which relates the parameters to the measurements, is typ-
ically called parameter-to-state mapping, which needs to be inverted to solve the
inverse problems.
There is a large variety of mathematical methods for the determination of such
parameters, if they are scalar. However, the development of mathematically justi-
fied numerical schemes for the identification of distributed parameters (functions,
such as spatially varying material constants or time-dependent process parameters)
is the subject of ongoing research. The technical difficulty in determining these pa-
rameters increases even more, if the parameter-to-state mapping is ill-posed, which
means that different and diverging parameter settings give - within the available
measurement precision - rise to the same data. In mathematical terms, the inverse
problem needs a regularization (stabilization).
In order to describe one of the most successful regularization schemes (the so-
called Tikhonov-regularization) let us introduce a formal notation of the inverse
problem. We denote the searched parameters by p, where we assume that p = p(t, x)
is a function
depending on time and/or location. We assume that p has finite energy,
i. e. p2 (s) ds < ∞ .
Let us assume, that we cannot determine p, e. g. the unbalances of a rotating
system, directly. Hence, we have to rely on indirect measurements such as vibra-
tion measurements, g, at some sensor locations. The relation between p and g is
determined by a model or operator A, which describes the manufacturing process
as well as the measurement device. This model allows us to simulate artificial mea-
surements for different parameter settings p1 . Again, in general, g = g(t, x) is a
distributed parameter, i. e. a function of time and/or location. We formally write
A(p) = g. However, measurement devices typically have limited precision and we
3 Adaptive Finite Elements and Mathematical Optimization Methods 69
where Φ (p) is called the regularization, term which controls the energy of p and at
the same time models some a priori information on the optimal parameter setting.
The classical setting
Φ (p) = p(s)2 ds (3.15)
leads to parameters of minimal energy. This type of least squares approach has been
investigated since the early 1960’s. However, only recently, so-called sparsity reg-
ularization schemes have been introduced: there, one is not interested in restrict-
ing the energy of the parameter p but one aims at determining parameters, which
are“well-localized”. For an illustration of this concept, let us consider a cutting pro-
cess. One might want to determine the optimal, time-varying feed rate of the cutting
tool in order to obtain a pre-defined surface g. In addition, the model A allows us to
simulate the resulting surface g for given feed rate f . Based on experience, a feed
rate f0 is suggested, which, however, leads to a sub-optimal result. The searched for
parameter, in this case, is the change of the feed rate, e. g. fopt = f + p and we want
to determine p such that A( f + p) ∼ g. Enforcing the side condition, one prefers for
the modification of the feed rate f over a minimal time span, leads to a modified
Tikhonov-functional. Solutions of this type are obtained with penalty terms of the
form
Φ (p) = |p(s)| ds . (3.16)
This minor change in the formulation of the minimization procedure has a tremen-
dous impact on the structure of the minimizer as well as on the complexity of the nu-
merical schemes needed for approximating the minimizers: the classical L2 -penalty
term (3.15) allows to use Hilbert space techniques leading to least square approx-
imations, which have been well-developed for decades. In contrast, sparsity pro-
moting L1 -penalty terms (3.16) require optimization techniques, which were first
analyzed in 2004 by [19].
The matrix M denotes the mass matrix, S the stiffness matrix and D the damp-
ing matrix of the system. The vibrations u of each node of the discretization are
collected in the vector u and the corresponding loads in vector p. Each node has 6
degrees of freedom: the displacements u, v and w in each space direction, the tor-
sion angle βz and the cross section slopes βx and βy . Since no information about the
damping properties is available the damping is neglected in a first attempt, i. e. we
consider the equation
Mü(t) + Su(t) = p(t), (3.18)
y
motor coupling spindle
casing
x
Ctorsion casing
Engine
En
rotor z
Caxial
Clateral C
spindle
C
engine
Cengine Ccasing
granite base
We assume that the induced vibration is also harmonic with the same frequency ω ,
i. e., u(t) = u0 eiω t , with a complex and time independent amplitude u0 . By inserting
this into (3.18) we get a (time-independent) expression for the vibration amplitude
of the vibration u0 ,
u0 = (−ω 2 M + S)−1 p0 . (3.19)
3.5.2.3 Balancing
We will now discuss the inverse problem of balancing a rotating system. As men-
tioned in the last section, the mounting of a workpiece onto the spindle often results
in an unbalanced system. This unbalance can be projected onto the two balancing
planes. Reconstructing this projected unbalance distribution allows us to determine
the necessary balancing weights, which have to be mounted in the balancer rings.
The reconstruction is carried out with the help of the above given model. To this
end, the workpiece has to be added to the model, i. e., we have to include an element
in the rotating part of the spindle. Balancing then means computing the balancing
weights as well as their positions from measurements of the vibration.
In order to determine the balancing weights from vibration measurements we
have formulated the problem as an operator equation
A(p) = u. (3.20)
Here, p denotes the unbalance distribution we are looking for and u the vibrations
induced by this unbalance distribution measured at positions, where sensors are
mounted. The measurements for u are obtained from an idle run of the machine,
hence we have no forces from the cutting process and punb and u are harmonic and
related via their absolute values or amplitudes p0 and u0 by (3.19):
instead. The penalty term Φ (p) acts as a stabilizer and prevents large values of
Φ (p). Typical choices of Φ are, e. g.
1/p
Φ (p0 ) = p0 p := ∑ |pi0 | p , 0 < p ≤ 2. (3.22)
i
Problems, which are not continuously invertible, are called ill-posed and require
regularization [21]. The functional (3.21) is well-known as Tikhonov-functional and
forms a regularization method [31, 32]. The parameter α is called the regularization
parameter. It has to be chosen according to the amount of noise in the data and
assures the convergence of pδ0,α to p0 as δ → 0. A popular rule for the choice of α
is Morozov’s discrepancy principle, [35, 30].
3 Adaptive Finite Elements and Mathematical Optimization Methods 73
In order to find the underlying unbalance distribution projected onto the bal-
ancers, we have minimized the functional (3.21) with Φ (p) = p22 . For a numeri-
cal simulation, we place a specified unbalance at the node of the workpiece and use
our model to compute the related vibrations of the platform. We then consider the
inverse task of determining the unbalance (mass, location, phase) from the vibration
data.
In order to simulate a real-life situation it was assumed that the data can only
be obtained at two sensor positions. To simulate measurement errors we disturbed
the data at the sensor positions with noise (20 − 30%). The sensors are fixed in
y-direction on the spindle casing above the two balancers. We then reconstructed
the unbalance projected onto the balancing planes by minimizing (3.21), i. e., we
reconstructed an unbalance distribution, which produces approximately the same
vibrations at the sensors as the original unbalance created by the workpiece but is
located at the two balancer positions. As an example for an unbalance created by
a workpiece, we set 50 gmm at an angle of 30◦ to the zero mark at the workpiece
node in the model. The corresponding amplitudes of the vibrations for a frequency
range of [5, · · · , 50] Hz ’measured’ at the first sensor are plotted in Figure 3.15 (line
without markers). The vibrations were disturbed with a noise level of 26 %, i. e.
(u − uδ )/u = 0.26. In the balancing planes, we reconstructed unbalances of 26
gmm in the first and 26.5 gmm in the second plane, both with a position of 30◦ .
The balancing masses at the balancers can be placed on a radius r = 41.5 mm, thus
the according masses are determined as 0.63 g in the first plane and 0.64 g in the
second, both having to be placed under an angle of 210◦, see Figure 3.16. As shown
in Figure 3.15 (line with star markers), the setting of those balancing masses reduces
the vibration amplitude significantly.
Fig. 3.15 Vibration amplitude for a workpiece unbalance of 50 gmm with 30◦ angle before
and after balancing [12]
74 M. Andres et al.
Fig. 3.16 Balancing masses and angles for a workpiece unbalance of 50 gmm with 30◦ ,
balancing radius 41.5 mm. [12]
The second part of this chapter was devoted to parameter identification and in-
verse problems. The balancing of a rotating system is a prototypical example for
the use of inverse problem theory in industrial applications. The expensive multiple
step process of finding the best balancing weights and their positions can be reduced
to one computation using only one set of vibration measurements. We are paying
for this reduction of costs with the necessity of creating a mathematical model of
the rotating machinery, which relates unbalance distributions in the system to the
resulting vibrations. On the other hand, once such a model is created it can be used
repeatedly, e. g., in ultra precision machining, for every new machining process with
an unbalance-situation originating from the mounting of a new workpiece.
References
1. Ainsworth, M., Oden, J.T.: A Posteriori Error Estimation in Finite Element Analysis.
John Wiley & Sons, New York (2000)
2. Andres, M.: Dual-dual formulations formulations for frictional contact problems in me-
chanics. Ph.D. thesis, Institute of Applied Mathematics, Leibniz Universität Hannover
(2011)
3. Arnold, D.N., Brezzi, F., Douglas Jr., J.: PEERS: a new mixed finite element for plane
elasticity. Japan J. Appl. Math. 1(2), 347–367 (1984),
http://dx.doi.org/10.1007/BF03167064, doi:10.1007/BF03167064
4. Bangerth, W., Rannacher, R.: Adaptive Finite Element Methods for Differential Equa-
tions. Birkhäuser, Basel (2003)
5. Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A.V., Schröder, A.,
Weinert, K.: Space adaptive finite element methods for dynamic Signorini problems in
the simulation of the NC-shape grinding process. In: Proceedings of the 1st CIRP Inter-
national Conference on Process Machine Interactions (PMI), pp. 309–316 (2008)
6. Blum, H., Jansen, T., Rademacher, A., Weinert, K.: Finite elements in space and time for
dynamic contact problems. Int. J. Numer. Meth. Engng. 76, 1632–1644 (2008)
7. Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for dy-
namic obstacle problems. ETNA, Electron. Trans. Numer. Anal. 32, 162–172 (2008)
8. Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for dy-
namic signorini problems. Comput. Mech. 44(4), 481–491 (2009)
9. Bostan, V., Han, W., Reddy, B.D.: A posteriori error estimation and adaptive solution
of elliptic variational inequalities of the second kind. Appl. Numer. Math. 52(1), 13–38
(2005), doi:10.1016/j.apnum.2004.06.012
10. Braack, M., Ern, A.: A posteriori control of modeling errors and discretisation errors.
Multiscale Model. Simul. 1(2), 221–238 (2003)
11. Brandt, C., Krause, A., Brinksmeier, E., Maaß, P.: Force modelling in diamond machin-
ing with regard to the surface generation process. In: Proceedings of the 9th International
Conference and Exhibition on Laser Metrology, Machine Tool, CMM and Robotic Per-
formance, LAMDAMAP 2009, London, June 30-July 2, pp. 377–386 (2009)
12. Brandt, C., Niebsch, J., Ramlau, R., Maass, P.: Modeling the influence of unbalances for
ultra-precision cutting processes. Zeitschrift für Angewandte Mathematik und Mechanik
(2011) (in press)
13. Carstensen, C.: Numerical analysis of the primal problem of elastoplasticity with hard-
ening. Numer. Math. 82(4), 577–597 (1999), doi:10.1007/s002110050431
76 M. Andres et al.
14. Carstensen, C., Dolzmann, G.: A posteriori error estimates for mixed FEM in elasticity.
Numer. Math. 81(2), 187–209 (1998),
http://dx.doi.org/10.1007/s002110050389,
doi:10.1007/s002110050389
15. Carstensen, C., Klose, R., Orlando, A.: Reliable and efficient equilibrated a posteriori
error finite element error control in elastoplasticity and elastoviscoplasticity with hard-
ening. Comput. Methods Appl. Mech. Engrg. 195(19-22), 2574–2598 (2006)
16. Carstensen, C., Valdman, J., Brokate, M.: A quasi-static boundary value problem in
multi-surface elastoplasticity: part 1 - analysis. Math. Methods Appl. Sci. 27, 1697–1710
(2004)
17. Carstensen, C., Valdman, J., Brokate, M.: A quasi-static boundary value problem in
multi-surface elastoplasticity: part 2 - numerical solution. Math. Methods Appl. Sci. 28,
881–901 (2005)
18. Czekanski, A., El-Abbasi, N., Meguid, S.A., Refaat, M.H.: On the elastodynamic solu-
tion of frictional contact problems using variational inequalities. Int. J. Numer. Meth.
Engng. 50, 611–627 (2001)
19. Daubechies, I., Defrise, M., De Mol, C.: An iterative thresholding algorithm for linear
inverse problems with a sparsity constraint. Communications in Pure and Applied Math-
ematics 57(11), 1413–1457 (2004)
20. Ekeland, I., Témam, R.: Convex analysis and variational problems. Classics in Ap-
plied Mathematics, vol. 28, english edn. Society for Industrial and Applied Mathematics
(SIAM), Philadelphia (1999); Translated from the French
21. Engl, H.W., Hanke, M., Neubauer, A.: Regularization of Inverse Problems. Mathematics
and its Applications, vol. 375. Kluwer Academic Publishers Group, Dordrecht (2000)
22. Glowinski, R., Lions, J.-L., Trémolières, R.: Numerical analysis of variational inequali-
ties. Studies in Mathematics and its Applications, vol. 8. North-Holland Publishing Co.,
Amsterdam (1981); Translated from the French
23. Han, W.: Finite element analysis of a holonomic elastic-plastic problem. Numer.
Math. 60(4), 493–508 (1992), doi:10.1007/BF01385733
24. Han, W., Reddy, D.: Plasticity. Springer (1999)
25. Hlaváček, I., Haslinger, J., Nečas, J., Lovı́šek, J.: Solution of variational inequalities in
mechanics. Applied Mathematical Sciences, vol. 66. Springer, New York (1988); Trans-
lated from the Slovak by J. Jarnı́k
26. Ito, K., Kunisch, K.: Lagrange multiplier approach to variational problems and appli-
cations. Advances in Design and Control, vol. 15. Society for Industrial and Applied
Mathematics (SIAM), Philadelphia (2008)
27. Laursen, T.A.: Computational Contact and Impact Mechanics. Springer, Heidelberg
(2002)
28. Niebsch, J., Ramlau, R., Brandt, C.: On the interaction of unbalances and surface quality
in ultra-precision cutting machinery. In: SIRM 2011, Darmstadt, Germany (2011)
29. Rademacher, A.: Adaptive finite element methods for nonlinear hyperbolic problems of
second order. Ph.D. thesis, Technische Universität Dortmund, Published in Verlag Dr.
Hut, München (2009)
30. Ramlau, R.: Morozov’s discrepancy principle for tikhonov regularization of nonlinear
operators. Numer. Funct. Anal. and Optimization 23, 147–172 (2002)
31. Ramlau, R.: A steepest descent algorithm for the global minimization of the tikhonov-
functional. Inverse Problems 18(2), 381–405 (2002),
http://stacks.iop.org/0266-5611/19/i=2/a=312
32. Ramlau, R.: Tigra - an iterative algorithm for regularizing nonlinear ill-posed problems.
Inverse Problems 19(2), 433–467 (2003),
http://stacks.iop.org/0266-5611/19/i=2/a=312
3 Adaptive Finite Elements and Mathematical Optimization Methods 77
33. Rannacher, R., Suttmeier, F.-T.: A posteriori error estimation and mesh adaptation for
finite element models in elasto-plasticity. Comput. Methods Appl. Mech. Eng. 176(1-4),
333–361 (1999)
34. Richter, T.: Parral multigrid method for adaptive finite elements with application to 3d
flow problems. Ph.D. thesis, Ruprecht-Karls-Universität Heidelberg (2005)
35. Scherzer, O.: The use of Morozov’s discrepancy priciple for Tikhonov regularization for
solving nonlinear ill-posed problems. Computing 51, 45–60 (1993)
36. Schröder, A.: A posteriori Error Estimation in Mixed Finite Element Methods for Sig-
norini’s Problem. In: Numerical Mathematics and Advanced Applications. Proceedings
of ENUMATH 2009, pp. 801–808 (2010)
37. Schröder, A.: Fehlerkontrollierte adaptive h- und hp-Finite-Elemente-Methoden für Kon-
taktprobleme mit Anwendungen in der Fertigungstechnik. Ph.D. thesis, Universität Dort-
mund (2005)
38. Schröder, A.: Mixed Finite Element Methods of Higher-Order for Model Contact Prob-
lems (2009); Humboldt Universitt zu Berlin, Institute of Mathematics, Preprint 09-16
39. Schröder, A., Rademacher, A.: Goal-oriented error control in adaptive mixed FEM for
Signorinis problem. Computer Methods in Applied Mechanics and Engineering 200(1-
4), 345–355 (2011)
40. Schröder, A., Wiedemann, S.: Erro restimates in elastoplasticity using a mixed method.
Applied Numerical Mathematics 61, 1031–1045 (2011)
41. Sokolnikoff, I.S.: Mathematical theory of elasticity, 2nd edn. McGraw-Hill Book Com-
pany, Inc., New York (1956)
Part II
Grinding
Chapter 4
High-Performance Surface Grinding
4.1 Introduction
In contrast to machining processes with defined cutting edges, grinding is affected
by a large number of cutting edges. These cutting edges are single grains, whose
number, geometry and relative position to the workpiece are not known. Although
the depth of cut in high-performance surface grinding is in the range of millime-
ters, the depth of cut of the single grains is in the sub-micrometer range. For this
reason, process machine interactions (PMI) in the range of micrometers have great
influence on the single grains and hence on the complete process [1; 2; 3].
To approximate these PMI, a process model was coupled with a machine model.
The process model, a kinematic-geometrical simulation of surface grinding
(KSIM), is able to deliver undeformed chip parameters of each single grain in the
contact zone and a high-frequency signal of the grinding force for each point of the
time discretization. Based on the simulated grinding force, the machine model, a fi-
nite element model for spinning motions (SFEM), simulates PMI of the modeled
machine structure in terms of the resulting displacements of the grinding wheel.
Experimental and numerical studies will be presented. The experimental inves-
tigations have been carried out to gain an enhanced understanding of the influence
of PMI on the process and the process results.
The numerical studies include the determination of undeformed chip parame-
ters and a detailed description of the machine model. After presenting different
coupling strategies of KSIM and SFEM, the result of a combined simulation will
be described.
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 81–100.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
82 J.C. Aurich et al.
Experimental setup
The surface-up grinding experiments were executed on a high performance grinding
machine with high stiffness. Two series of tests were carried out. In the first series, a
resin bonded CBN (cubic boron nitride) grinding wheel (B181) with a diameter of
Ø 400 mm and a width of 12 mm was used. The second series of tests was aiming at
higher loads of the machine system to achieve higher displacements, corresponding
to stronger PMI. To achieve these higher loads, i. e. higher grinding forces, specific
material removal rates higher than those of the first series were necessary. Therefore,
another grinding wheel had to be used. This grinding wheel was an electro-plated
CBN grinding wheel (B251) with a diameter of Ø 400 mm and a width of 24 mm.
The workpiece material used for both series of tests was a heat-treated steel AISI
4140H (42CrMo4V, DIN EN 10132-3, 55 ± 2 HRC) with dimensions of
125 mm x 50 mm x 10 mm.
Common measurement techniques were applied using a dynamometer for
measuring normal and tangential forces and non-contact hall effect probe to meas-
ure the effective spindle power. In addition, the displacements of the grinding
wheel relative to the headstock were measured (to examine the process machine
interactions). The application of optical measurement systems was not possible
due to the high amount of cooling lubricant, which was used during the experi-
ments. Also, tactile sensors could not be used because of the high rotational
speeds. For this reason, the displacements were measured using eddy-current sen-
sors. They provide high-resolution signals and can be used in wet environments.
In the first series of tests using the resin bonded grinding wheel, the measuring
surface used for the sensors was a high-precision shoulder on the grinding wheel.
In the second series of tests using the electroplated grinding wheel, the flange of
the grinding wheel was used as measuring surface.
grinding wheel diameter 400 mm; resin-bonded; B181 V240 (DIN ISO 6106)
grinding wheel speed: 60 - 100 m/s
depth of cut: 0.25 - 1 mm
grinding parameters
feed rate: 600 – 4,800 mm/min
maximum specific material removal rate: 20 mm³/mm s
lubricant cooling 300 l/min at 12 bar; wheel cleaning 100 l/min at 25 bar
The relative dynamic stiffness during the stationary grinding process can be
calculated based on the measured normal grinding forces and displacements . The
calculated relative dynamic stiffness varies from 120 - 200 N/mm (Fig. 4.1).
Variations of the wheel speed and the depth of cut do not influence the relative
dynamic stiffness. The main influence on the relative dynamic stiffness in normal
direction was exerted by the feed rate. Different feed rates result in different ma-
chine loading times. Feed rates of 1,200 mm/min or less reveal a relative dynamic
stiffness of 120 - 160 N/µm. Feed rates of 2,400 mm/min or higher (grinding pass
< 3 s) reveal an almost constant relative dynamic stiffness of approximately
200 N/µm. It seems that the inertia of the traveling column of the grinding ma-
chine of more than 10 tons influences the measured displacements due to the short
time periods of loading.
Fig. 4.1 Relative dynamic stiffness from the first series of tests
Compared to the investigations with the resin bonded grinding wheel the spe-
cific material removal rate was significantly increased. However, the measured
normal and tangential grinding forces were smaller than the forces reached with
the resin-bonding wheel (Fn = 140 - 780 N and Ft = 30 - 260 N). The resulting
maximum measured displacements in normal direction were smaller (4 µm), too.
Similar to the first series of tests, the relative dynamic stiffness during the sta-
tionary grinding process was calculated based on the measured normal grinding
forces and displacements. The relative dynamic stiffness ranged between
110 N/µm and 210 N/µm (see Fig. 4.2). The same correlations for the relative dy-
namic stiffness as for the resin-bonded grinding wheel were observed. Variations
of wheel speed and depth of cut showed minor influence whereas the feed rate had
a strong influence.
Fig. 4.2 Relative dynamic stiffness from the second series of tests
In this second series of tests, the quality of the ground workpieces was eva-
luated. The surfaces were classified into three categories:
• no grinding marks
• grinding marks
• grinding marks and burn
According to these categories, the whole parameter range can be mapped into ac-
ceptable process results (no grinding marks) and inacceptable process results
4 High-Performance Surface Grinding 85
(grinding marks or grinding marks and burn). The classification of the ground sur-
face was carried out visually since measurement with tactile sensors was not suit-
able. Fig. 4.3 shows the results of the mapped parameters.
Similar to the relative dynamic stiffness, the feed rate is the dominating factor
for the occurrence of grinding marks and burn. This can be seen when comparing
different feed rates at constant material removal rates (e. g. 50 mm³/mms at
63 m/s): At a feed rate of 2,000 mm/min there are no grinding marks or burn.
However, at 3,000 mm/min grinding marks occurred. At 4,000 mm/min (and
6000 mm/min) grinding marks as well as burn were detected. The occurrence of
grinding marks can be attributed to the higher feeding path per wheel revolution at
higher feed rates and the resulting higher chip thicknesses. The observed burn was
caused by less cooling lubricant volumes per time unit when applying higher feed
rates. Less grinding marks and burn occurred at higher wheel speeds. This results
from the better cooling lubricant supply at higher wheel speeds and to the smaller
feeding path per wheel revolution.
Despite the correlation of the influence of feed rate on the relative dynamic
stiffness and on the workpiece quality no direct correlation between these two pa-
rameters could be found. Higher relative dynamic stiffness does not necessarily
result in higher or lower workpiece quality. When taking, for example, the feed
rate 5,000 mm/min at grinding wheel speed 100 m/s the relative dynamic stiffness
is approximately constant for all applied depths of cut. In contrast, the workpiece
quality ranges from no grinding marks to grinding marks and burn.
be calculated depending on the process type and the workpiece material. The chip
cross section and the specific grinding force are used to calculate the forces of
each single grain based on a simplified approach of Kienzle. The Kienzle ap-
proach can be used to calculate cutting forces, e. g. for turning or milling
processes. The simplified approach is based on a specific cutting force kc,sim (an
experimentally-determined parameter), the undeformed chip thickness and the
width of the undeformed chip. The total process force can then be calculated by
cumulating the forces from each kinematic grain (i. e. each grain that removes ma-
terial) (Fig. 4.4).
For the simulations to be carried out the modeled grinding wheel with the di-
mension of Ø 400 x 3 mm² was used. By using a smaller grinding wheel width
than the wheel used in the experiments the calculation time of the kinematic simu-
lation could be reduced (without reducing the simulation quality [9]).
Based on the applied manufacturing process (electroplating) a grain protrusion
height of 35 % of the nominal grain size of 251 µm was used to model the grind-
ing wheel. This means that about 163 µm of the grain height were embedded in
the bonding, resulting in an average grain protrusion height of 88 µm. Because of
the statistic nature of the modeling process the grain size distribution leads to sig-
nificant deviations of the average grain protrusion height. This results in grains
with high protrusion heights that remove the whole material, resulting in chip
thicknesses of 10 to 15 µm. Those grains would wear very fast under real grinding
conditions. This rapid wear would lead to a stationary grinding process characte-
rized by a high number of kinematic grains distributing the material removal
process over many grains with less chip thickness. Since in KSIM only one wheel
revolution is simulated with no wear a numerical dressing process was imple-
mented in KSIM to model grinding wheels in a stationary wear state [8]. Within
this numerical dressing process, the grain protrusion height of every single grain is
ideally cut down to a specified limit (dressing height, see Fig. 4.5) to fit real grind-
ing conditions, for the simulations in this study 90 µm.
88 J.C. Aurich et al.
Table 4.3 Model specifications and process parameter for kinematic simulation
modeled grinding diameter 400 mm; width 3 mm; B251; 1,200 grains/cm²
wheel
process parameter grinding wheel speed: 63 - 100 m/s
for kinematic simu- depth of cut: 0.5 - 1 mm
lation
feed rate: 1,000 – 5,000 mm/min
maximum specific material removal rate: 83.33 mm³/mms
Generally, an increase of either the feed rate or the depth of cut results in higher
specific material removal rates. This means that a higher material volume has to
be ground in the same time, resulting in a higher chip thickness and a higher num-
ber of momentarily active grains. The results show a progressive increase for the
chip thickness and a degressive increase of the number of active grains with
increasing feed rate. The increase of active grains is limited due to the constant
contact length at a constant depth of cut, which limits the maximum number of
available grains for the material removal process. For this reason, the chip thick-
ness shows a progressive increase because it has to compensate the degressive in-
crease of active grains with increasing material removal rates.
4 High-Performance Surface Grinding 89
A comparison of the average chip thickness and the number of active grains for
different grinding wheel speeds reveals higher values for lower grinding wheel
speeds. At higher grinding wheel speeds, less material has to be removed within
each grinding wheel revolution. Thus, the chip thickness and the number of active
grains decrease.
The modeling of the machine has been carried out by means of the finite element
method on the base of the so-called Arbitrary Lagrangian Eulerian (ALE) ap-
proach. The system to be modelled consists of a grinding wheel, which rotates
with a constant angular velocity (Fig. 4.7). The notations used are briefly ex-
plained in the following.
90 J.C. Aurich et al.
u* Ro Ri
Workpiece
δ
δ0
u
F0(t)
Fig. 4.7 Model of the grinding wheel. For the machine model there are two important radi-
uses: the inner radius Ri which the wheel is fixed at, and the expanded radius R* that serves
as reference to determine the wheel’s deformation
The grinding wheel has an inner radius Ri, which is assumed to remain constant
throughout the simulation since the grinding wheel is fixed to the spindle. Due to
the centrifugal forces caused by the angular velocity ω the outer radius Ro expands
to the radius R* by the radial expansion u*.
Starting-up processes enabling the grinding wheel to accelerate from the idle
state up to the velocity ω do not have to be considered. It will be assumed that at
the time t = 0 the grinding wheel already rotates with the velocity ω and thus has
the outer radius R*. The depth of cut, which the grinding wheel is initially plung-
ing into the workpiece with, is denoted by δ0. The cutting force resulting from δ0
is denoted by F0. Due to this force, which is assumed to only radially act at one
point of the wheel’s circumference, the grinding wheel deforms in radial direction
by the amount u, defined relatively to the expanded radius R*. In tangential direc-
tion, the force application point is retained by defining an additional bearing. The
deformation u is taken to be positive, if the outer radius R* increases. The depth of
cut, which results when the machine’s deformation is taken into account, is simply
given by δ = δ0 + u. According to this nomenclature one can distinguish initial
values (δ0 and F0(δ0)), which are set up at the beginning of the process, and effec-
tive values (δ and F(δ)), which result during the cutting process, for both the depth
of cut and the cutting force.
The machine model can now be used either independently or in conjunction
with a process model to run both uncoupled and coupled simulations respectively.
Simulation results are presented in a later section.
versa. The coupling algorithm is based on general modeling strategies for coupled
problems, i. e. Neumann and Dirichlet boundary conditions are applied. In the fol-
lowing, four coupling strategies are presented. Strategies 2 and 4 are closely re-
lated to each other.
Coupling Strategy 1 - data exchange and iteration until convergence after
multiple time steps (each wheel revolution)
The process model (KSIM), as presented in chapter 4.3.1, provides simulation re-
sults after every grinding wheel revolution. For this reason, data exchange of
process and machine model is carried out after each grinding wheel revolution.
The resolution of the simulated grinding force signal depends on the grinding
wheel speed and the wheel diameter. Also, one grinding wheel revolution consists
of a number of time steps (here 1,000), comparable to the number of data points
for one revolution in real measurements. When modeling, for example, the grind-
ing wheel presented in 4.3.1 (Ø 400 mm, grinding wheel speed 100 m/s) the simu-
lated grinding force signal for one wheel revolution has a resolution of 80 kHz.
The coupled simulation starts with the generation of a force signal by the
process model. For this force signal an ideal grinding machine is assumed, i. e. no
machine deformations are considered. This force signal is used as input data for
the machine model. The machine model generates displacements of the grinding
wheel for one wheel revolution based on the input force signal. After resetting the
workpiece topography (to simulate the same grinding wheel revolution as before)
the process model is restarted based on the generated displacements by the process
model. In doing so, the penetration of the grinding wheel and the workpiece is
modified, which results in a different grinding force signal. This altered force sig-
nal again serves as input for the machine model and so forth. This iteration is re-
peated until “convergence” occurs, i. e. the simulated force and / or displacement
signals (average values) of two consecutive iteration steps show minor deviations
(Fig. 4.8).
Fig. 4.8 Coupling strategy 1 - data exchange after each wheel revolution
Coupling Strategy 2 - data exchange and iteration until convergence for each
time step
The principle of coupling strategy 2 is similar to that of coupling strategy 1. The
difference is that the data exchange between the process and machine model
(grinding force and displacement) is conducted after each time step within one
grinding wheel revolution instead of the complete revolution (Fig. 4.9).
92 J.C. Aurich et al.
Fig. 4.9 Coupling strategy 2 - data exchange and convergence for each time step
In other words, coupling strategy 1 is applied for each single time step until con-
vergence is reached. When convergence of force and / or displacement is achieved
for the current time step the iteration cycle will be started again for the next time
step. When using 1,000 time steps for one wheel revolution coupling strategy 2 con-
sists of 1,000 iteration cycles, each with numerous iteration steps. This results in
very large simulation times so that a simplification of each iteration cycle represents
an interesting alternative solution, as will be discussed in coupling strategy 4.
The main advantage of coupling strategy 2 is the possibility to simulate only a
part of one grinding wheel revolution. That means computing time can be saved, if
less than one wheel revolution is needed.
Coupling Strategy 3 - asynchronous data exchange for each time step without
iteration
In coupling strategy 3, no iteration cycles are used. As in coupling strategy 2, the
force signal at time step tn is used as an input for the machine model. The calcu-
lated displacement then serves as input for the process model, however, at time
step tn+1 (Fig. 4.10).
Fig. 4.10 Coupling strategy 3 - asynchronous data exchange for each time step
4 High-Performance Surface Grinding 93
Coupling Strategy 4 - data exchange and iteration until convergence for each
time step based on an empirical cutting force model
As already mentioned, this coupling strategy corresponds to coupling strategy 2
and represents coupling in a strong form. That means that a converged coupled so-
lution is achieved at every time step. The coupling iterations are, however, carried
out based on the assumption that the process force depends linearly on the cutting
depth, i. e.
F0 (δ 0 ) = c ⋅ δ 0 (4.1)
where c denotes the “process stiffness”. This assumption has been undertaken
based on the linear relation between the chip cross section and the specific grind-
ing force, to be found in the Kienzle approach, for example, and furthermore
enables a considerable acceleration of the coupled simulation.
Eq. (4.1) is only valid for an ideally stiff grinding wheel, which the cutting
depth is constant for throughout the whole process. For a real deformable grinding
wheel, however, the grinding force is not a function of the cutting depth set up but
of the actual (or effective) one. It will be assumed that the initial depth of cut δ0
leads to a wheel deformation u, which simply has to be added to δ0. By multiply-
ing the sum [δ0 + u] by the process stiffness c an adjusted force value is then
obtained and can be subsequently used to calculate the deformation again. This
coupling procedure can be written as:
In summary, the models of the process, the machine and the process machine
interactions presented above constitute a complete simulation procedure, which
can now be used to run coupled simulations.
smaller than the values gained after the first iteration (1,050 N and 7.2 µm), i. e.
smaller than the values corresponding to the uncoupled solution.
Coupling strategy 1 is well suited for the case of quasi-static loading. In simu-
lating a high-performance grinding process it has been observed that the high
loading and deformation rates lead to an abrupt change of the state of single grits
from contact to non-contact. This effect led to excessive fluctuations in the simu-
lated force and deformation values causing the coupled simulation to diverge (de-
spite the reasonable average values). For this reason, coupling strategy 1 is not
suitable for coupled simulations involving vigorous dynamics.
In order to investigate dynamic properties of the coupled process machine sys-
tem a coupling strategy, which guarantees a strong coupling has to be used, i. e.
either strategy 2 or 4. The first one comprises more modeling details, the last is
more economic in terms of simulation times. In addition, the fast Fourier trans-
form of the force signal simulated in KSIM suggests carrying out the coupled
simulation within a wide range of loading frequencies. In the following, the force
signal provided by KSIM is substituted by a set of harmonic depths of cut with
different frequencies ranging from 0 to 20 kHz. This set results in a set of harmon-
ic loads, which are now able to excite the grinding wheel at different frequencies.
The process machine interaction is then simulated using coupling strategy 4. The
corresponding simulation model is based on the parameters listed in Tab. 4.4. All
of them represent parameters commonly encountered in real grinding processes.
For the process stiffness c a high value is used in order to emphasize the effect of
coupling.
Table 4.4 Parameters for the coupled simulation, grouped into machine, process and coupl-
ing parameters
Machine model:
Inner radius Ri : 63.5 mm
Outer radius Ro : 200 mm
Elastic modulus E : 210 GPa
Poisson ratio ν : 0.3
Material density ρ : 7.85 g/ccm
Boundary conditions : – No displacement at the inner boundary.
– No tangential displacement at the point the
load is acting on.
Initial (setup) load F0(t) : Results readily form the product [c ⋅ δ0(t)].
Process model:
Angular velocity ω : 200 rad/s, i. e. roughly about 2,000 rpm.
Depth of cut δ0(t) : Harmonic function with a fixed mean value of 5
µm, a fixed amplitude of 1 µm and variable fre-
quencies ranging from 0 to 20 kHz.
Coupling model:
Process stiffness c : 200 N/µm
4 High-Performance Surface Grinding 95
The time-dependent force and deformation signals are exploited in the follow-
ing by means of their frequency and amplitude. Since the amplitude of a time de-
pendent signal is generally not constant with respect to time the amplitude mean
value over the whole signal is used.
The frequency spectrum of the grinding wheel can be obtained by varying the
excitation frequencies and simulating the resulting mean deformation amplitude
for each frequency. Also, the amplitude spectrum can be normalized by the de-
formation amplitude, which results when the grinding wheel is loaded by a quasi-
static force. A normalized representation enables different simulation models to be
compared to each other.
Fig. 4.12a shows how coupling shifts the eigenfrequencies gained from an un-
coupled simulation to higher values. Note that the frequency spectrum obtained from
a coupled simulation can no longer be considered as a property of the grinding
wheel but of the whole system consisting of grinding wheel, process and interaction.
To answer the question as to why the eigenfrequencies increase when coupling
is included one considers the applied force and the resulting deformation in time
domain, Fig. 4.12b,c. All forces shown are normalized by dividing them by the
same value, namely by the amplitude of the initial load F0(t), so that the initial and
the effective force can be readily compared to each other with regard to both am-
plitude and frequency. However, all deformations are normalized by dividing
them by their maximum value within the considered time interval. This has the
advantage that, in normalized form, forces and deformations can be represented in
the same diagram. A comparison of the frequencies is also possible. However, the
deformation amplitudes resulting from different simulations cannot be readily
compared to each other since deformations are not normalized by the same entity.
Also, only some special loading frequencies will be presented here:
• the loading frequency of f0 = 12.8 kHz, which corresponds to an eigen-
frequency of the uncoupled process machine system and
• the loading frequency of f0 = 13.3 kHz, which corresponds to an eigen-
frequency of the coupled process machine system.
f0 = 12.8 kHz
Although this loading frequency corresponds to an eigenfrequency of the grinding
wheel as given in Fig. 4.12a there is no significant increase of the deformation
amplitude throughout the coupled simulation. Fig. 4.12d shows that through cou-
pling the effective force signal acquires smaller amplitudes than the initial one be-
cause a deformed grinding wheel generates a smaller depth of cut than an unde-
formed one. In the first excitation cycle, the maximum amplitude of the force
effectively acting at the grinding wheel is only about 18 % of the initial force. An-
other more important aspect is the fact that through coupling the force signal ac-
quires a higher frequency. This is the reason why the shifting effect observed in
Fig. 4.12a occurs in direction of higher and not smaller frequencies. Fig. 4.12b
shows that the force, which is effectively applied, and the resulting deformation
oscillate with different frequencies. Under this circumstance, the deformation am-
plitude remained almost constant over the whole loading history. So, for reso-
nance, effective force and deformation must first synchronize. This happens at the
frequency of 13.3 kHz.
96 J.C. Aurich et al.
a
1.6
1.4
1.2
Normal Force in kN 1.0
0.8
0.6
0.4
0.2
0
0.02
0.015
0.01
0.005
0
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 38
Frequency in kHz
1.2 0
1.0 -2
Deformation in µm
0.8 -4
Force in kN
0.6 -6
0.4 -8
0.2 -10
0 -12
1 2 3 4 5 6
Iteration
Fig. 4.11 Results of the coupled simulation based on coupling strategy 1. (a) Force signal
simulated by KSIM. (b) FFT-transform of the force signal. (c) Convergence behavior of the
coupled simulation.
4 High-Performance Surface Grinding 97
a Non-coupled Coupled
log | A / A(static,non-coupled) |
Mean Deformation Amplitude
100
10
b Radial Deformation c
12.8 kHz 13.3 kHz
Effective Force
Force /Displacement
1.0 4
Normalized
0.5 2
0 0
-0.5 -2
-1.0 -4
10 Time in ms 10.25 10 Time in ms 10.25
d e
Loading Frequency: 12.8 kHz Non-coupled
1.0
Normalized Force
0.5
Force
-0.5
-1.0
0 Time in ms Time
01
Fig. 4.12 Results of the coupled simulation based on coupling strategy 4. (a) Frequency
spectrum. (b)-(c) Effective force and displacement in time domain for special loading fre-
quencies. (d) Influence of coupling on the loading frequency. (e) Influence of coupling on
the variation of the effective force over time.
98 J.C. Aurich et al.
f0 = 13.3 kHz
As shown in Fig. 4.12c, effective force and resulting deformation oscillate for this
loading frequency (after some loading cycles) with the same frequency. Only
through their synchronisation can the coupled process machine system get into
resonance.
Another important difference between the coupled and uncoupled system is that
force and deformation are inversely influenced by each other. In the case of cou-
pling, the deformation patterns resulting from a harmonic load, a beat for example,
is mirrored into the force pattern and can thus also be retrieved there (Fig. 4.12e).
This is not the case for an uncoupled simulation.
In Fig. 4.13, the eigenmodes of the grinding wheel are presented. They are just
briefly described in the following. At the first eigenfrequency, the grinding wheel
is oscillating in vertical direction and thus behaves approximately like a rigid
body. The second eigenform represents an alternating dilation and contraction in
horizontal and vertical directions. In the third eigenmode, the grinding wheel takes
a triangular form. In the forth one, the outer wheel boundary expands and con-
tracts uniformly. Finally, polygonal wheel forms result at higher eigenfrequencies.
Fig. 4.13 Eigenmodes of the spinning grinding wheel (ω = 200 rad/s). The contour plots
represent the field of the radial deformation. Positive values mean an increase in the ex-
tended radius of the outer wheel R* due to deformation.
4.4 Conclusion
Extensive experimental research was conducted to examine PMI during grinding.
In doing so, displacements of the grinding wheel were measured and the resulting
relative dynamic stiffness was mapped for a large parameter field. High stiffness
was identified for high feed rates for different grinding wheels. The examination
of the workpiece quality revealed grinding marks and burn when applying high
4 High-Performance Surface Grinding 99
feed rates. The experiments conducted enhance the understanding of the cause-
and-effect chain in grinding and the influence of PMI on the process and the proc-
ess result.
Also, a complete simulation procedure enabling the modelling of the process,
the machine and their interaction was discussed in detail. There are differences be-
tween results gained from uncoupled and coupled simulations, which can be
summarized as follows:
In the uncoupled process machine system:
• Loading frequencies, which correspond to eigenfrequencies of the ma-
chine, lead to a critical (non-stable) process.
• The grinding force results in a machine deformation that, however, has no
inverse influence.
• The knowledge of the frequency spectrum of the machine provides in-
formation about critical loading frequencies, which are not allowed, if the
process has to be kept stable.
In contrast, simulations of the coupled process machine system revealed that:
• Coupling is more complex than just scaling the amplitude and the fre-
quency of the initial load.
• A loading frequency, which corresponds to an eigenfrequency of the ma-
chine, does not necessarily lead to a critical process.
• Vice versa, a loading frequency, which lies outside the eigenfrequencies,
can be critical.
The last three aspects suggest that the knowledge of the natural frequencies of the
machine does not generally suffice to judge, whether a harmonic load is critical or
not. This again underlines the importance of simulation techniques in general and
of coupled simulations in particular.
Concerning simulation techniques one can say that there are quite efficient
simulation techniques to simulate the deformation behavior of the grinding wheel
due to transient process forces. This is primarily enabled by the use of the ALE-
approach instead of the conventional Lagrangian one so that the kinematically less
relevant rigid body motion is a priori omitted. This also enables the application of
the force at a fixed point without contact-searching steps.
There is still need to bring simulation and experimental techniques closer to-
gether. Providing confident information about the real workpiece surfaces is still
not reliably possible. Also, real-time simulations of high-performance grinding
processes are currently not available.
The physical phenomena taking place during PMI are quite complex and not
always accessible for measuring setups so that efficient simulation techniques are
of great importance to investigate the grinding process. Although the simulation
techniques presented in this chapter do not completely fit experimental results
they can still be considered as an important step in understanding the phenomena
involved in the PMI.
100 J.C. Aurich et al.
References
[1] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hofmeister, H.-W., Klocke, F.,
Peters, J., Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.:
Advances in Modelling and Simulation of Grinding Processes. CIRP Annals - Manu-
facturing Technology 55(2), 667–696 (2006)
[2] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and Simulation of
Process - Machine Interaction in Grinding. Production Engineering 3(1), 111–120
(2008)
[3] Herzenstiel, P., Bouabid, A., Steinmann, P., Aurich, J.C: Experimental Investigation
and Computational Simulation of Process-Machine Interactions during High-
Performance Surface Grinding. In: Proceedings of 1st International Conference on
Process-Machine Interaction, PZH Hannover, pp. 267–278 (2008)
[4] Warnecke, G., Zitt, U.: Kinematic Simulation for Analyzing and Predicting High-
Performance Grinding Processes. CIRP Annals – Manufacturing Technology 47(1),
265–270 (1998)
[5] Zitt, U.: Modellierung und Simulation von Hochleistungsschleifprozessen. Disserta-
tion, TU Kaiserslautern (1999)
[6] Braun, O.: Konzept zur Gestaltung und Anwendung definiert gesetzter Hochleis-
tungsschleifscheiben. Dissertation, TU Kaiserslautern (2008)
[7] Aurich, J.C., Herzenstiel, P., Sudermann, H., Magg, T.: High-performance dry grind-
ing using a grinding wheel with a defined grain pattern. CIRP Annals – Manufacturing
Technology 57(1), 357–362 (2008)
[8] Herzenstiel, P., Aurich, J.C.: Numerical and Experimental Investigations of a Grinding
Wheel with a Defined Grain Pattern. In: CIRP Conference on Modeling of Machining
Operations, vol. 12, pp. 567–574 (2009)
[9] Aurich, J.C., Herzenstiel, P., Kirsch, B., Steffes, M.: Experimental and Numerical Stu-
dies of a Surface Grinding Process. In: Proceedings of the 2nd International Confe-
rence on Process Machine Interaction (PMI) Session G1 (2010)
Chapter 5
Process Machine Interaction in Pendulum
and Speed-Stroke Grinding
Abstract. The complex interaction of process forces and machine structure affects
the quality of ground workpieces, especially in highly-productive machining
processes, if machines are operated at their limits. In speed-stroke grinding, the
highly-dynamic process forces are caused by high workpiece velocities and high
acceleration of the machine table. These forces are influenced by the process pa-
rameters, the material properties, the coolant application and the grinding tool
specification. The paper describes the approach to simulate the process machine
interaction in speed-stroke grinding by a coupled model. The machine is modeled
by a multi-body simulation, which can depict the static and dynamic behavior of
the machine for every working position. This machine model is coupled with an
analytical-empirical force model, which predicts the process forces regarding the
process parameters, the coolant application and the workpiece material. The ma-
chine control system is implemented in the model as well. The ability to model a
speed stroke grinding process, including the machine, the control system and the
process itself can be used to predict and improve the workpiece quality regarding
the measurement accuracy minimizing time and cost intensive experiments.
5.1 Introduction
In many cases, grinding is the only economical process to machine parts. Materials,
which are especially difficult to machine, are ground to achieve high productivity
and meet the requirements in terms of surface qualities. This paper focuses on
speed-stroke grinding. Speed-stroke grinding is defined as a surface grinding
process with high table speeds. The table speed of most of the conventional grind-
ing machines is limited to 30 m/min. Pendulum grinding with table speeds higher
than 50 m/min is defined as speed-stroke grinding [1]. In speed-stroke grinding the
chip formation mechanisms change to bigger chips causing lower specific grinding
energies at high table speeds. The resulting lower heat transfer to the workpiece can
be used to improve the workpiece quality or the productivity of grinding processes.
Speed-stroke grinding can be applied for machining highly heat-resistant materials
and ceramics as well as hardened steel [2, 3, 4].
High table speeds result in short contact times between the grinding wheel and
the workpiece. Short run-in and run-out phases lead to high gradients of process
forces. The values of these forces are influenced by the complex interaction
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 101–119.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
102 M. Weiß, F. Klocke, and H. Wegner
between the process parameters, coolant application and the mechanics of chip
formation specific to the behavior of different workpiece materials. The material
removal in grinding is realized by an amount of single grinding grits with an unde-
fined shape. Several grinding grits are in contact with the workpiece surface at the
same time, realizing material removal. Hence, it is a complex challenge to predict
the forces occurring in the grinding process, their influence on the excitement of
the machine structure and, thus, the workpiece quality.
In the following, an approach is shown, where the machine is modeled by a
flexible multi-body simulation and the occurring forces by an analytical-empirical
force model. These two models are coupled to comprehend the process-machine
system. As an input for the simulation model and to validate the simulation results
experimental investigations have to be conducted. The results of the analysis of
the process and the machine behavior were used to set up the simulation model.
Afterwards, experimental investigations of certain influences on the machine be-
havior and process forces resulted in a more detailed and more exact model of the
whole process and lead to a potential of optimization for industrial pendulum and
speed-stroke grinding processes.
Height adjustment
Height scale
Angle and position
y
adjustment
Needle nozzle
z Grinding wheel
Workpiece
Multi component
force dynamometer
x Machine table
Fig. 5.1 CAD drawing of machine interior with precision nozzle height adjustment
The static and dynamic behavior of the machine was examined by an experi-
mental modal analysis of the machine tool. A hydraulic actor unit was mounted to
the machine. This actor unit applied dynamic loads with different frequencies to
the grinding wheel. A piezo-electric sensor unit measured the response of the ma-
chine structure to the dynamic loads. This procedure was repeated with different
sensor and actor positions in x, y and z direction so that the cross relations be-
tween the excitations could be measured. More information about modal analysis
can be found in Section 1.2 of this book. The data was used to parameterize the
machine model. Thus, a comparison between the modeled and the real machine
behavior can be carried out.
10-1 3. eigenfrequency 78 Hz
FRF w m:2abs:+X/w m:Piezo:+X
Synthesized FRF w m:2abs:+X/w m:Piezo:+X
Elasticity Gyy [µm/N]
2. eigenfrequency 58 Hz
1. eigenfrequency 38 Hz
10-3
0°
Synthesized FRF w m:2abs:+X/w m:Piezo:+X
-180°
0 100 200 300 400 500 600 700 800 900
Frequency [Hz]
Fig. 5.2 Frequency response function (Gyy) of test machine, preload 1,000 N
Process forces
In the experiments, the dynamic forces during pendulum and speed stroke grind-
ing processes are measured by a multi-component force dynamometer and record-
ed on a PC. The measurement of specific normal and tangential forces and the
acoustic emission (AE) signal during two passes of the grinding wheel is shown in
Fig. 5.3. It was recorded in the middle of a speed-stroke process at a ground vo-
lume of V’w = 500 mm3/mm. The AE-signal is important for the analysis of the
data. Every time the grinding wheel gets in contact with the workpiece a peak in
the AE-signal can be detected. Thus, the exact time frame from the first to the last
contact of the grinding wheel with the workpiece can be specified.
To explain different effects the measurement is separated into nine phases. At
the beginning of Phase 1 the workpiece is on the left side of the machine and acce-
lerates towards the grinding wheel. Because of the moment of inertia the specific
tangential forces show a characteristic peak during the acceleration. At the begin-
ning of Phase 2 the workpiece is engaged by the grinding wheel. With a work-
piece velocity of vw = 80 m/min and a depth of cut ae = 15 µm the run-in time is
very short. It takes less than 2 ms from the first contact of the grinding wheel until
the maximum depth of cut is reached. However, the normal force needs a certain
time to reach its stationary level at the end of phase 2.
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 105
50
40
30
20
10
1 3 5 7 9
0
2 4 6 8
40
F‘t / N/mm
Specific
20
0
-20
-40
46,7 46,8 46,9 47 47,1 47,2 47,3 47,4
Time / s
AE-Signal
100
80
60
40
20
0
1 3 5 7
vs vs vs vs
vw vw vw vw
Run in
1 Table acceleration 2 Coolant phase + up grinding
Run out
3 Up grinding 4 Coolant phase
5 Table deceleration – Change table direction – Table acceleration
Run in
6 Coolant phase 7 Down grinding
Run out
8 Coolant phase + down grinding 9 Table deceleration
Fig. 5.3 Force and acoustic emission signals during speed-stroke grinding
Responsible for this effect is the coolant. With the shown machine setup the
coolant nozzle is on the right side of the grinding wheel. If the workpiece is
approaching from the left side, the main coolant flow is not able to get into the
contact zone. The coolant jet hits the workpiece from the right side. Until the
workpiece moves further the coolant jet builds up a hydro-dynamic pressure in
and in front of the contact zone, as described in [6, 7]. Depending on flow rate,
coolant speed, nozzle angle, nozzle position and process parameters the coolant
adds a significant value to the normal force [8]. Besides that, the hydrodynamic
coolant pressure influences the dampening behavior of the system [9].
During phase 3 the process is nearly stationary. At the end of this phase, the
normal forces drop rapidly because of the disengagement of the grinding wheel.
After this, the coolant forces drop with a time delay in phase 4. As the gap
106 M. Weiß, F. Klocke, and H. Wegner
between the workpiece and the grinding wheel gets wider while the workpiece
moves apart the measured normal force decreases. In phase 5, the workpiece table
decelerates, changes its direction and accelerates in the opposite direction, which
can be seen in the tangential force progression. Phase 5 shows that the level of
acoustic emission is lower than in phase 1 and 9 even though the grinding wheel is
not in contact with the workpiece. The reason for this phenomenon is that the coo-
lant jet hits the workpiece from the side during phase 1 and 9. When the grinding
wheel is at the right side of the grinding wheel there is no coolant splashing
against the workpiece.
Phase 6 is characterized by an increase of normal forces while the workpiece is
not in contact with the grinding wheel yet. This is related to the dynamic coolant
forces, which build up because of the narrowing gap between the workpiece
and the grinding wheel. The run-in takes place at the beginning of phase 7. Before
the workpiece runs out of contact the forces drop as a result of the deflected
coolant jet.
The described effects show the complexity of forces occurring during the grinding
process. To model the forces the main influences have to be taken into consideration.
In order to quantify the effects of process parameters workpiece material and coolant
application numerous systematic experiments have been conducted.
The measured force signal contains a lot of information about the process and
the influences mentioned above. To separate the effect of the chip formation from
the coolant induced forces the data has to be analyzed as shown in Fig. 5.4. On the
upper left, measured specific normal forces plotted against the grinding wheel cir-
cumferential speed vs up and down grinding are shown. The forces drop with
higher rotation speed of the grinding wheel from above 30 N/mm to 20 N/mm.
The coolant forces were measured after every fifth stroke of the process by
conducting one stroke without a depth of cut ae. The results are shown exemplarily
in Fig. 5.4 on the upper right. The coolant induced forces increase up to a certain
point – here 80 m/s – before they begin to decrease at higher circumferential
speeds. For conventional processes a similar trend was described by Heinzel [10].
Higher grinding wheel circumferential speeds lead to a higher flow rate through
the contact zone. Hence, less coolant ponds in front of the contact zone and the
pressure in this area drops. As a result, the specific coolant normal forces de-
crease.
If the specific coolant normal forces F’nKSS are subtracted from the measured
specific grinding normal forces F’n, the specific cutting normal forces F’c,n can be
identified. This is necessary to evaluate the separate effects and to simplify
the modeling of the occurring forces in grinding as every effect can be examined
separately.
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 107
20 20
10 10
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120140
Grinding wheel circumferential Grinding wheel circumferential
speed vs / m/s speed vs / m/s
40
Specific cutting normal
Up-grinding
force F‘c, n / N/mm
30 Down-grinding
20
10
0
0 20 40 60 80 100 120 140
Grinding wheel circumferential
speed vs / m/s
Besides the hydrodynamic pressure, which builds up in and in front of the con-
tact zone, additional coolant forces can be induced due to the ground geometry.
An example is shown in Fig. 5.5. Narrow gaps between parts of the grinding
wheel, which are not in contact with the workpiece, and the workpiece lead to ad-
ditional hydro-dynamic forces. The smaller the gap, the higher these forces. If
such workpiece geometries are ground, the additional coolant forces need to be
considered.
In addition to the forces, the coolant has an influence on the dampening beha-
vior of the whole process machine system. To evaluate these effects the following
experimental setup was used. A workpiece made up of the two different materials
is ground. On one side, the workpiece consists of hardened 100Cr6 and on the
other side of Ureol, an easy to machine and homogenous modeling material. This
workpiece was ground with high table speed. The machining of the modeling
workpiece material creates very small forces compared to the bearing steel.
108 M. Weiß, F. Klocke, and H. Wegner
Grinding wheel
Additional coolant
forces
Total depth of
cut
FC
Workpiece
Contact area
Gap between grinding
wheel and workpiece
The experimental setup is shown in Fig. 5.6. As a result, the forces drop rapidly
at the junction of the two materials. If no coolant is used, a deviation of the profile
height can be measured on the surface of the soft material. When using coolant the
deviation is much smaller due to dampening and pre-load effects of the fluid in
and in front of the contact zone. Although several experiments with this setup
were accomplished quantitative statements about the dampening behavior could
not been derived yet as the collected data scatters in a big range. As a conse-
quence, the experimental setup and measurement need to be improved.
Without coolant
Profile / µm
Height/ µm
Easy to machine 25
of
modeling material
0
0
Profile
(Ureol)
-25
0 25 50 75 100
Distance / mm
With coolant
Profile / µm
25
Height of
00
100Cr6 -25
0 25 50 75 100
Distance / mm
50
Inconel 718
Specific cutting normal
40
force F‘c,n / N/mm
30
100Cr6
20
10 vw = 20 m/min vw = 80 m/min
vw = 50 m/min vw = 120 m/min
0
0 200 400 600 800
Hardness / HV
Material: Grinding parameters: Coolant application:
100 Cr 6 (AISI 52100) vs = 120 m/s Emulsion 4 %
and Inconel 718 ae = 60, 24, 25, 10 µm Needle nozzle (11°) +
Grinding wheel: Q‘w = 20 mm3/mms clearing nozzle
CBN 10B 181 VP Flow rate: 110 l/min
The material removal in grinding results from various contacts between the
grinding wheel and the workpiece. In addition, the grinding grits have irregular
shapes and they move on a certain path around the centre of the grinding wheel.
Because of this, some of the grits just plow through the workpiece surface, some
grits form chips and some are not in contact with the workpiece surface. A large
110 M. Weiß, F. Klocke, and H. Wegner
amount of the energy is transferred into elastic and plastic deformation of the
workpiece material. Hence, the grinding wheel properties, depending on influ-
ences like grit size, bond type or dressing conditions, influence the occurring
forces significantly. Different grinding wheels were tested and need to be imple-
mented in the modeling as well, as shown in the following chapter.
Machine tool
MBS
DBS DBS
s(t) s(t), v(t)
For the designed model the simulation of the process forces and the control sys-
tem is implemented into Matlab Simulink. In the following, the machine and the
process model are described in more detail.
Machine model
For the design of a MBS model of a machine the following steps have to be con-
ducted. The procedure of the development of a MBS machine model is shown in
Fig. 5.9.
Comparison with
experiments
2. FE-model
Abstraction and
meshing of machine
components Modelling of
moving parts
Fig. 5.9 Procedure for the development of a multi body machine model
Based on CAD data of the machine the relevant parts need to be abstracted.
Small bore holes and small radii are removed because they have no significant in-
fluence on the stiffness or the mass of the system. As a result, the FE-meshing of
the finite element model (FE-model), which is performed afterwards, is less com-
plex. Next, guiding systems, spindle bearings and screw connections are approx-
imated by simple spring elements. Afterwards, a comparison between the
measured machine behavior and the simulated machine is conducted to evaluate
the required stiffness of the joints. Furthermore, the relevant eigenmodes and ei-
genfrequencies can be depicted. Fig. 5.10 shows the 2nd eigenmode of the machine
system at 58 Hz exemplarily.
For the MBS-model the FE structure parts have to be converted into flexible
bodies. A flexible body allows the description of its flexible properties at defined
force transmission points. The calculation of the forces at these knots and the elas-
tic deformation of the parts is implemented in MSC.Adams and calculated by a
developed sub-routine for every position of the spindle and the machine table in
the machine. The special aspect about the developed MBS-model is the possibility
of realizing a movement of certain parts. This allows the simulation of movements
of the spindle and the machine table like in the real machine. More detailed infor-
mation about the machine model and the subroutines can be found in [13, 14].
112 M. Weiß, F. Klocke, and H. Wegner
2nd Eigenmode
Eigenmode of original machine Eigenmode simulation model
Fig. 5.10 Comparison between measured and simulated eigenmode and eigenfrequency
Process model
The task of the process model is to calculate the occurring forces during the grind-
ing process, regarding the position of the grinding wheel, the grinding parameters
and other boundary conditions. Many grinding force models have been developed
over the last decades. Usually, a technological study of grinding processes results
in the development of a model, which is valid for a closely limited field with given
boundary conditions, as grinding is a very complex process of material removal
[15]. Many models to predict the grinding forces are presented in [15, 16, 17].
Most of these models consider the kinematic of the grinding process and an empir-
ical additive, which represents the influences of the grinding wheel and the work-
piece material.
For the prediction of the grinding forces during speed-stroke grinding an
adapted grinding force model was developed. The schematic representation of the
model is shown in Fig. 5.11. Depending on the workpiece position the current ma-
terial removal is calculated by using a chip longitudinal section Al, which
represents the intersection of the grinding wheel and the workpiece during one
time step of the simulation. This longitudinal section Al is multiplied by an expe-
rimental investigated force-ratio factor, which is dependent on the process
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 113
parameters, the material properties and the grinding wheel specification, to get the
cutting forces. Simultaneously, the coolant forces are calculated. These forces are
dependent on the workpiece position and the process parameters as well, which
was shown before. Additionally, coolant application parameters like nozzle angle,
nozzle type and coolant flow rate are considered [18]. The coolant forces and the
cutting forces are summarized to the total process forces.
Material properties
kc
vw Fc = Al ⋅ k c
Chip longitudinal vs
F [N]
vs Process parameters
vw and tool properties
Total process forces
Workpiece position
v KSS Nozzle
α
Q KSS
Coolant forces
Coolant parameters
In the following, the single modeling steps are described in more detail.
Fig. 5.12 shows the calculation of the chip longitudinal section during the run-in
phase. In position I, the grinding wheel first gets into contact with the workpiece.
During the run-in phase the contact length increases and thus the material removal
per time step also increases. This increasing material removal is proportional to
the area, which is bordered by the contact length between two time steps. Towards
time step III the maximum depth of cut is reached and the chip longitudinal sec-
tion does not change in case of an even workpiece surface. Grooves in the work-
pieces can also be represented by the model.
114 M. Weiß, F. Klocke, and H. Wegner
t0 t1 t2
x Surface with
deviations
Position: I II III
ΔAl2 Δt
ΔAl1 Ideal surface
Grinding wheel ae,0
ae,1
ΔAl3 ae,2
Al lg
vw Workpiece
Fig. 5.12 Chip longitudinal section during run-in of the grinding wheel
With constant time steps Δt the change of the longitudinal chip section can be
described by geometrical values. For a constant workpiece velocity vw the longi-
tudinal chip section can be described as a function of the position x according to
equation 1 during the run-in phase.
A l,max 2 ⋅ A l,max (1)
A l,Run-In (x) = − 2
x2 + x ± ΔA li
lg lg
Al,max is the maximum chip longitudinal section, which is reached after the run-in
of the grinding wheel at the maximum depth of cut ae. ΔAli is the deviation of the
chip longitudinal section Ali due to an uneven surface or the geometry. At the run-
out, the chip longitudinal section is calculated with the following equation:
A l,max 2 ⋅ A l,max (2)
A l,Run-Out (x) = 2
x2 − x + A l,max ± ΔA li
lg lg
If the workpiece has no deviations at the surface, the chip longitudinal section
stays constant. Fig. 5.13 shows the cutting force characteristic of one stroke during
grinding. In speed-stroke grinding, the run-in and run-out phases are in the range
of milliseconds. If the table speed is 120 m/min at a depth of cut of 10 µm, these
phases take 0.935 ms, for example. If there are deviations in the surface, an addi-
tional ΔAli needs to be added for the calculation.
chip longitudinal
section / Al
Al,max Overrun
Al,Run-in Al,Run-out
0
Grinding wheel position x
Fig. 5.13 Characteristics of total chip longitudinal section during one stroke
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 115
1
100 specification and
80 10B 181 V circumferential speed vs
2
60 Determine specific
40 2
B251 V force ratio factor
20 1
0
0 20 40 60 80 100 120
Grinding wheel circumferential
speed vs / m/s
Adaption of kc
Set workpiece material
3
Specific force ratio
233 %
Determine adaption
100Cr6
200 4 4
factor
100
3 vw = 20 m/min vw = 80 m/min
vw = 50 m/min vw = 120 m/min
0
0 200 400 600 800
Hardness / HV
Fig. 5.14 Adaption of force ratio factor kc
116 M. Weiß, F. Klocke, and H. Wegner
As the different influences are taken into account the kc factor has to be adapted.
An example for the functionality of the model is shown in the following. Fig. 5.14
illustrates the calculation of the adapted force ratio factors. Different grinding
wheel specifications lead to specific force ratio factors kc at different circumferen-
tial speeds vs. The data was generated in several experiments for different grinding
wheels. In the example in Fig. 5.14, a grinding wheel with a grit size of 181 µm is
used.
The first diagram at the top of the Figure shows the determination of the specif-
ic force-ratio factor for a vitrified, bonded CBN wheel with a grit size of 181 µm
at a circumferential speed of 120 m/s. The specific force-ratio factor is
59,000 N/mm3, assuming defined standard conditions. These standard conditions
are hardened and tempered 100Cr6 as the workpiece material and a combined
needle nozzle with a clearing nozzle with a total coolant flow rate of 110 l/min.
After the “standard” force-ratio factor kc has been calculated it needs to be adapted
by the influences mentioned above. In the example in Fig. 5.14, the nickel base al-
loy Inconel 718 is chosen. As mentioned above, the material, which is difficult to
machine, leads to higher cutting forces. Thus, the force-ratio factor is adapted. In
this case, the forces are higher by a factor of 2.33. This leads to a total specific
force-ratio factor kc,tot of 137,470 N/mm3.
The coolant application affects the forces in two ways. On the one hand, the
coolant application has an influence on the chip formation, which leads to varying
forces. This is also considered by an adaption of the force ratio factor kc. On the
other hand the hydrodynamic pressure adds a value to the total forces. The simula-
tion program calculates these forces depending on the actual coolant conditions.
The coolant forces are calculated in the model by using the workpiece position as
the input factor as well. In combination with the process parameters and the coo-
lant application, parameters like coolant flow rate and nozzle angle, the occurring
forces can be predicted similar to the cutting forces. These are the most important
parameter adjustments for the coolant supply [20].
In a last step, the total process forces for the current time step are summarized.
Then the value is transferred to the machine model, which calculates the static and
dynamic displacement of the grinding wheel. With the displacement, the cycle
starts again with the calculation of the forces for the next time step. Hence, it is
possible to record the tool path and the resulting surface.
A comparison between a measured and simulated surface is shown in Fig. 5.15.
The measurement shows the characteristic super-elevation at the beginning. The
fast-run in of the grinding wheel results in a rapid raise of the process forces. On
this side of the workpiece, the system is pre-loaded by the hydrodynamic coolant
forces. Accordingly, the profile height on the left side is higher at this side com-
pared to the right. After a short period of vibration the workpiece is quite even.
The peak in the middle of the measurement of the real surface is used as a refer-
ence as the measurement length of the used surface measurement device is less
than 100 mm. On the right side of the workpiece the profile height drops. This is
related to the decreasing coolant force at the right side of the workpiece, as de-
scribed above. The height of this step varies depending on the process parameters
and coolant application.
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 117
30 Measurement
reference
Measured surface
Profile height [µm]
20 Simulated surface
10
-10
Fig. 5.15 Measured and simulated workpiece surface after speed-stroke grinding
5.4 Optimization
The approach to model the interaction between the speed-stroke grinding process
and the machine structure, as shown in this paper, can be used to improve produc-
tion processes. Furthermore, it builds the foundation for further research for the
understanding of the grinding process, especially speed-stroke grinding.
It was shown that the coupled simulation allows the prediction of the shape of
the speed-stroke ground workpiece, taking the process parameters, the coolant ap-
plication and the workpiece material into account. If the process result can be pre-
dicted without conducting time and cost-consuming experiments, a faster start of
production can be realized. With the knowledge of the maximum deviation of the
workpiece measurements the process time can be optimized. The roughing process
leads to a certain surface. The grinding minimum allowance for the finishing
strokes can be determined and, thus, the minimum cycle time can be depicted.
The developed coupled simulation can be adapted for other grinding processes.
On the one hand, the grinding force model needs to be adapted for different grind-
ing processes. In particular, the calculation of the current material removal rate
needs to be adapted to other kinematics. On the other hand, a new machine model
has to be built up for other machines. Depending on the process reduced machine
models with less build-up effort may be sufficient. The advantage of flexible mul-
ti-body simulations is the ability to simulate the machine behavior in every posi-
tion of the machine slides, the modeling effort, however, is very high, which is
shown above. In a lot of surface grinding processes the spindle moves within a li-
mited area in the machin only. As a consequence, the simulation of the machine
by means of a multi-mass spring system can be sufficient. This system can be es-
tablished in a substantially shorter time frame.
118 M. Weiß, F. Klocke, and H. Wegner
Another way to use the results of this work is to already evaluate the behavior
of machines in the design phase. Based on the CAD data the machine can be mod-
eled and the process behavior as well as the process result can be estimated with-
out building and testing the new machine. Structural weaknesses and illegitimate
process parameters can be determined, specific to the grinding processes.
References
[1] Duscha, M., Klocke, F., D’Enremont, A., Linke, B., Wegner, H.: Investigation of
Temperatures and residual stresses in speed-stroke grinding via Fea Simulation and
practical tests. In: Proceedings in Manufacturing Systems, vol. 5(3), pp. 143–148
(2010)
[2] Zeppenfeld, C.: Schnellhubschleifen von gamma-Titanaluminiden. Gamma-
Titanaluminiden. RWTH Aachen University, Aachen (2005)
[3] Nachmani, Z.: Randzonenbeeinflussung beim Schnellhubschleifen. Dissertation,
RWTH Aachen University, Aachen (2008)
[4] Uhlmann, E., Sammler, C.: Schnellhubschleifen von Hochleistungswerkstoffen. Jahr-
buch Schleifen, Honen, Läppen und Polieren, vol. 64, pp. 155–168. Vulkan-Verlag,
Essen (2010)
[5] Weck, M.: Dynamisches Maschinenverhalten bei der Zerspanung mit undefinierter
Schneide. Werkzeugmaschinen 5 – Messtechnische Untersuchung und Beurteilung,
pp. 326–362. Springer, Berlin (2001)
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 119
[6] Beck, T.: Kühlschmierstoffeinsatz beim Schleifen mit CBN. Dissertation, RWTH
Aachen University, Aachen (2002)
[7] Wittmann, M.: Bedarfsgerechte Kühlschmierung beim Schleifen. Dissertation, Uni-
versity of Bremen, Bremen (2007)
[8] Brinksmeier, E., Heinzel, C., Wittmann, M.: Friction, Cooling and Lubrication in
Grinding. Annals of the CIRP, Keynote Paper 48(2), 581–598 (1999)
[9] Marksoud, T.M.A., Mokbel, A.: Suppression of chatter in grinding using high-
viscosity coolants. Proceedings of the Institution of Mecanical Engineers, Part B
(Journal of Engineering Manufacture) 216, 113–123 (2002)
[10] Heinzel, C.: Methoden zur Untersuchung und Optimierung der Kühlschmierung beim
Schleifen. Dissertation, University of Bremen, Bremen (1999)
[11] Klocke, F.: Process Design. Manufacturing Processes 2 – Grinding, Honing, Lapping,
pp. 251–287. Springer, Berlin (2009)
[12] Esser, M., Brecher, C., Witt, S.: Interaction of manufacturing process and machine
tool. Annals of the CIRP, Keynote Paper 58, 588–607 (2009)
[13] Hoffmann, F.: Optimierung der dynamsichen Bahngenauigkeit von Werkzeugma-
schinen mit der Mehrkörpersimulation. Dissertation, RWTH Aachern University, Aa-
chen (2008)
[14] Klocke, F., Brecher, C., Sitte, B., Weiß, M.: Analyse der dynamischen Wechselwir-
kungen bei Pendel- und Schnellhubschleifprozessen. Jahrbuch Schleifen, Honen,
Läppen und Polieren 64, 53–65 (2010)
[15] Tönshoff, H.K., Peters, J., Inasaki, I., Paul, T.: Modelling and Simulation of Grinding
Processes. Annals of the CIRP, Keynote Paper 41(2), 677–688 (1992)
[16] Brinksmeier, E., et al.: Advances in Modeling and Simulation of Grinding Processes.
Annals of the CIRP, Keynote Paper 55(2), 667–696 (2006)
[17] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and simulation of
process: machine interaction in grinding. In: Production Engineering – Research and
Development, vol. 3(1), pp. 111–120. Springer, Heidelberg (2009)
[18] Weiß, M., Klocke, F., Wegner, H.: Influence of coolant and workpiece material prop-
erties on dynamic grinding forces in speed stroke grinding. In: Proceedings of the 2nd
International Conference on Process Machine Interaction, Vancouver, G4 (2010)
[19] Klocke, F., Duscha, M., Hoffmann, F., Wegner, H., Zeppenfeld, C.: Machine –
Grinding Wheel –Workpiece Interaction in Speed-Stroke Grinding. In: Proceedings
of the 1st International Conference on Process Machine Interaction, Hannover, pp.
259–266 (2008)
[20] Wittmann, M., Heinzel, C., Brinksmeier, E.: Evaluating the Efficiency of Coolant
Supply Systems in Grinding. In: Production Engineering – Research and Develop-
ment, Braunschweig, vol. XI(2), pp. 39–42 (2004)
Chapter 6
Simulation of Process Machine Interaction
in NC-Shape Grinding
Abstract. The study focuses on the NC-shape grinding process when using toroid
grinding wheels and its simulation. First, the experimental investigation with re-
spect to the machine structure and its dynamic behavior, the process forces as well
as the temperature distribution in the workpiece and the grinding wheel are dis-
cussed. That forms the basis for the modeling and simulation of the NC-shape
grinding process. The simulation consists of a geometric-kinematical simulation
coupled with a finite element simulation. To validate the simulation, comparisons
between the quantities measured and the corresponding calculated values are car-
ried out. Subsequently to this validation the transferability of the simulation to
other grinding processes is studied. Furthermore, the simulation is utilized to op-
timize grinding processes, especially with respect to the NC data.
6.1 Introduction
Grinding processes are mainly used as surface finishing processes. The chip re-
moval is usually carried out by rotating grinding wheels with bonded abrasive
grains. The process is characterized by high cutting speeds and various contacts
between the workpiece and the single grains. Grinding is usually the last step in
the production chain of a workpiece and, accordingly, this phase plays a very im-
portant role. On account of this, defects caused by the grinding process would be
associated with high costs and must be avoided.
A toroid grinding wheel can be flexibly used for NC-shape grinding. Based on
the semi-circular profile of the grinding wheel and a line-by-line movement it is
possible to machine geometrically-complex surfaces. The different parameters re-
lated to NC-shape grinding with a toroid grinding wheel are demonstrated in
Fig. 6.1. The feed motion in the NC-shape grinding process is controlled by NC-
data, which provides excellent possibilities for grinding free-formed workpiece
surfaces. The use of CNC-technology and an adapted path control allows the
grinding of planes as well as profiles and complex shapes. The contact area be-
tween grinding wheel and workpiece varies during engagement along the tool
paths. In the case of complex engagement, the contact area is not measurable
without additional software tools. The shape grinding process is determined by the
shape of the tool, the kinematic conditions of the cut, the chip removal and the
surface produced [1].
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 121–141.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
122 D. Biermann et al.
Fig. 6.1 Different parameters of NC-shape grinding with a toroid grinding wheel [2]
The interaction between grinding wheel, workpiece and machine tool in an NC-
shape grinding process using a toroid grinding wheel is examined in the present
experiments. The research presented in this paper describes the process behavior
of the shape grinding process. The description of the interaction between grinding
wheel, workpiece and machine tool in a NC-shape grinding process requires vari-
ous simulation models and corresponding experiments.
In addition to obtaining a basic understanding of the process and its simulation,
the application of the simulation and its transferability to other processes are the
focus of the current research. Here, application means that the simulation results
are used to optimize the NC-shape grinding process. One approach we mainly
work on this purpose is the modification of the NC-data in such a way that grind-
ing errors are reduced or eliminated. First results related to path planning are pre-
sented in this chapter, where the depth of cut is increased to compensate for the
deflection of the grinding machine. It is considerably more difficult to reduce the
effect on the workpiece caused by oscillations, which result from the dynamic be-
havior of the machine, because restrictions relating to the machine tool control of
the grinding machine have to be taken into account. With respect to the practical
application of the simulation, the transferability of this simulation approach to dif-
ferent grinding machines and other grinding processes is of decisive importance.
The simulation is applied to other grinding wheels than the toroid one and the
workpiece material varied. Thereby, the parameters in the simulation model are
identified and it becomes obvious which parameters have to be adjusted. Further-
more, the dependency of the model's parameters on the modifications is described.
In the next section, we will discuss the experimental basis of the simulation and
describe the experiments relating to the dynamic behavior of the machine as well
as to measurements of force and temperature. The topic of Section 6.3 is the simu-
lation approach and a detailed description of single constituent parts of the simula-
tion. Furthermore, the determination of the model's parameters using numerical
methods is explained. Section 6.4 focuses on the validation of the simulation re-
sults. The application of the simulation to path planning and the transferability of
the simulation to other grinding processes are examined in Section 6.5. Finally, we
will draw some conclusions and provide an outlook on further developments.
6 Simulation of Process Machine Interaction in NC-Shape Grinding 123
to determine the maximum traversing speed when the distance between the NC-
data points is Δ = 0.5 mm have shown that the machine reaches a maximum work-
piece velocity of vw = 2238 mm/min because the drive dynamics of the machine
using this NC-data points is limited. This has to be taken into account during the
shape grinding processes [2].
Fig. 6.2 Eigenmodes of the grinding machine investigated as determined by the modal
analysis along with a representation of the extreme position of deflection [2]
6 Simulation of Process Machine Interaction in NC-Shape Grinding 125
Two eigenmodes play an important role when dynamic forces in a vertical di-
rection affect the machine. Engaging the grinding wheel with the workpiece caus-
es a waviness with an eigenfrequency of f = 60.2 Hz. A strong vertical oscillation
of the spindle tower can be seen at this frequency, so the grinding spindle with its
tower oscillates along the y axis. The second eigenmode has an effect in the ver-
tical direction at an eigenfrequency of f = 143 Hz and represents its effect through
a vertical oscillation of the spindle. The spindle executes a horizontal oscillating
movement parallel to the machine table at an eigenfrequency of f = 84.6 Hz [2].
Fig. 6.4 Friction coefficients determined for different process parameters along the grinding
path
Fig. 6.5 Dependency of temperature on the distance between the thermocouple and the con-
tact area [8]
Fig. 6.6 Temperature development over workpiece length during shape grinding
the grinding wheel is determined by the NC-data. The resulting process force
Fres in the current time step is calculated from the geometric-kinematical data.
To calculate this force an empirical grinding force model is used, which is only
able to predict a resulting process force. No local or microscopic effects are taken
into account.
The data for the part of the workpiece surface which might be in contact with
the grinding wheel is passed on from the geometric-kinematical simulation to the
finite element simulation. This information is used to describe the contact con-
straints of the dynamic Signorini problem in the current time step. Then, the dis-
crete problem is solved and the normal contact stress at that time is determined.
The resulting accumulated contact force Fcon is calculated from the normal contact
stress and this value is returned to the removal predictor. The most important parts
of the grinding machine, the grinding wheel and the spindle, are taken into ac-
count in the finite element simulation. The stiffness of the other parts of the ma-
chine is modeled by elastic bearings.
The finite element simulation is able to calculate the accumulated contact force
Fcon. Thereby, the contact area is represented in detail. Fcon depends directly on the
actual depth of cut ae,act. So it is expected that a good approximation of ae,act has
been found, if Fcon and Fres agree. The predicted global process force Fres is stored
in the removal predictor.
Two steps are performed in the removal predictor. First, it tests whether the ac-
cumulated contact force Fcon and the predicted process force Fres match. If that is
the case, the cycle is left and the next time step is started. Otherwise, a corrected
value for the actual depth of cut ae,act is predicted and passed on to the geometric-
kinematical simulation. Here, the surface of the workpiece is modified according-
ly, and the cycle restarts. See [9] for more details.
Force model
An empirical process force model was implemented in the geometric-kinematical
simulation based on a series of experiments. A specific force model for the NC-
shape grinding process based on the contact area of the geometric-kinematical
simulation was developed. This empirical force model is examined under ideal
conditions with deformation, vibration and temperature not being taken into
account [3].
The process forces increase with an accumulating workpiece velocity in an ex-
ponential way. This behavior is analogous to the behavior of the cutting forces
with defined cutting edges. Therefore, it is possible to determine a force model for
the grinding process, which is based on the Kienzle equation [14], usually em-
ployed for other cutting processes.
The process forces of the NC-shape grinding are calculated fairly exactly for
the geometric-kinematical simulation by the process force model. On average, the
results differ by 6 % from those of the measurements. Greater deviations only ap-
pear for large contact areas. By implementing the process force model into the
geometric-kinematical simulation the process forces, the contact area and the
workpiece shape can be calculated and used as the basis for the finite element si-
mulation [3].
Fig. 6.9 Drawing and finite element model of the spindle and the grinding wheel
The algorithm for solving the discrete problems is complex. An extensive pres-
entation of this theme can be found in [16]. It includes an outer fix-point algo-
rithm, where a quadratic program with nonlinear constraints has to be solved in
each iteration. Therefore, a sequential quadratic programming (SQP) method is
applied. The rotation of the grinding wheel complicates the calculation. An arbi-
trary Lagrangian Eulerian (ALE) approach is used to take the rotational effects in-
to account. This is also described in [16].
To discretize the thermo-elastic problem of the workpiece two points have to be
considered. The first one is that the basic domain changes during the calculation
due to the material removal. Consequently, the initial data of a new time step must
be interpolated on the new mesh. The second point is directly related to this. A
new finite element mesh has to be constructed in each time step. This is carried
out by meshing the dexel model of the workpiece stored in the geometric-
kinematical simulation.
The finite element simulation is complex and therefore time consuming. A
posteriori error estimates have been derived and adaptive finite element methods
have been developed to achieve a given accuracy at minimal numerical costs. Two
different approaches are considered: In the first one, the error is estimated in the
global energy norm [21-23]. This approach is easy to implement and leads to good
meshes but the error is not measured in the relevant physical quantities. This dis-
advantage has been overcome by dual-weighted residual (DWR) based error esti-
mators, see [18, 24]. More details can be found in Section 3.4.
6.3.3 Coupling
The target depth of cut ae,tar is given by the NC-data. The machine, however, is not
able to realize this depth of cut. Instead, an actual depth of cut ae,act is measured,
which is smaller than the target depth of cut. The geometric-kinematical simula-
tion is only able to simulate the target depth of cut. It is combined with the finite
element simulation to obtain a better approximation of ae,act.
The process force Fres is predicted by the geometric-kinematical simulation and
the contact force Fcon is calculated by the finite element simulation. A good ap-
proximation of the actual depth of cut is found, if the contact force and the pre-
dicted process force are equal. The problem is to find such an approximation of
the actual depth of cut. The solution algorithm is described in this section.
6 Simulation of Process Machine Interaction in NC-Shape Grinding 133
The accumulated contact force Fcon is calculated from the actual normal contact
stress by integration. The contact force Fcon is a function of the actual depth of cut.
We minimize the deviation between Fres and Fcon measured in the 2-norm squared
as dependent on the actual depth of cut by using the golden cut algorithm. The op-
timal point is then the actual depth of cut sought. For an alternative approach and
more details, see [9].
6.3.4 Parametrization
Many of the model parameters, the frictional resistance coefficient and the heat
distribution factors, for instance, cannot directly be determined by experiments.
However, they must be accurately defined to ensure precise simulation results.
Numerical parameter identification techniques are used to calculate the model pa-
rameters and to overcome the difficulties related to direct measurement.
Based on an experimental design, several suitably chosen experiments are car-
ried out. Some quantities, which can be determined with small measurement error
and calculated by the simulation, are measured. The experiments are reproduced
numerically in the simulation, where some initial guesses, e. g. from literature, are
used for the parameters. The difference between the simulation results and the
measured results is calculated in an appropriate norm. Then, optimization algo-
rithms are applied to minimize the deviation, where the parameters are the optimi-
zation variables. The optimal solution then gives the parameters sought.
This technique has been used to determine the damping parameters and the
stiffness of the bearings. The experiments are described in [2] and the parameter
identification in [20]. For the determination of the parameters connected to the
model of the thermal effects, see [8, 15].
of cut was set at ae = 40 µm. A non-water-based coolant was used. The process
kinematics was down-grinding.
The simulation was carried out as described above. The termination tolerance
was set at tol = 10 N. This seems to lead to a rough approximation. Experience
with the simulation, however, has shown that a smaller tolerance changes the val-
ues of the actual depth of cut by only about ae,act = 0.001 µm. Thus, the tolerance
is accurate enough in relation to the other model errors. Three to four passes of the
simulation cycle are needed to reach the given prescribed tolerance.
In Fig. 6.10, a cut through the workpiece is displayed. The measured result is
also provided for comparison. Temperature effects are neglected in this calcula-
tion. One can easily see that the results are good for the quasi-stationary process
phase. However, to get an accurate result during the run-in damping has to be con-
sidered in the simulation. For more details, see [2].
Fig. 6.11 Comparison of the tangential forces using the force model and the FEM based on
the frictional model
Fig. 6.13 Comparison of measurement and simulation results for one pass
depth of cut along the grinding path. It can be seen that the simulation with path
planning over a grinding path having constant engagement conditions attains the
mean target cut of depth (Fig. 6.14). The positioning of the workpiece in the
machine has to be examined and the traverse velocities of the different axes taken
into account when path planning in order to integrate the path planning into
the simulation for the changing engagement conditions and to realize the path
planning for every time step.
Fig. 6.15 Engagement conditions using different grinding wheels in the geometric-
kinematical simulation
138 D. Biermann et al.
The forces for different process parameters are measured. Based on the mean
value of the forces measured, the force coefficients at every workpiece speed for
all grinding wheel contours could be determined. The simulated forces are in good
agreement with the measured forces when flat surfaces are produced using cylin-
drical grinding wheels as well as when slanted surfaces are produced using a pro-
file grinding wheel. Comparing the forces, it becomes obvious that there is a great
variance in the forces in the area of the quickly-changing contact surfaces when
concave surfaces are produced. Here, only small depths of cut can be achieved us-
ing cylindrical and profile grinding wheels (Fig. 6.16). Therefore, it is impossible
to predict the forces in the simulation. The development of forces along the sur-
face contours can be simulated using toroid grinding wheels because, due to the
relatively large depth of cut, the contact surfaces are not so greatly changed as
they would be for the depths of cut used with cylindrical and profile grinding
wheels.
Fig. 6.16 Comparison of the measured and simulated forces along a grinding path using a
cylindrical grinding wheel for a concave workpiece
One can see that when the temperature in the process is measured the tempera-
tures behave as described in literature and in Sect. 6.2.3 when using both grinding
wheel contours. An example is shown in Fig. 6.17.
A further analysis of the grinding processes is now in progress. On the one
hand, the temperatures measured using the different grinding wheel contours are
being compared with the temperatures calculated in the finite element simulation.
On the other hand, other workpieces are being investigated.
6 Simulation of Process Machine Interaction in NC-Shape Grinding 139
Fig. 6.17 Dependency of temperature on the distance between the thermocouple and the
contact area using a profile grinding wheel
References
[1] DIN EN 8589-11. Schleifen mit rotierendem Werkzeug - Einordnung, Unterteilung,
Begriffe. Beuth Verlag (2003)
[2] Jansen, T.: Entwicklung einer Simulation für den NC-Formschleifprozess mit To-
russchleifscheiben. Dissertation, Technische Universität Dortmund. Vulkan Verlag,
Essen (2007)
[3] Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A., Weinert, K.: Ex-
perimental analyses to develop models for NC-shape grinding with a toroid grinding
wheel. In: Denkena, B. (ed.) The 1st International Conference on Process Machine In-
teractions, Hannover, Germany, pp. 279–287, 3.9–4.9 (2008)
[4] Brecher, C., Esser, M., Witt, S.: Interaction of manufacturing process and machine
tool. CIRP Annals 58(2), 588–607 (2009)
[5] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hoffmeister, H.-W., Peters,
J., Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.: Advances
in modelling and simulation of grinding processes. CIRP Annals 55(2), 667–696
(2006)
[6] Klocke, F.: Manufacturing Processes 2: Grinding, Honing, Lapping. Springer, Berlin
(2009)
[7] Tönshoff, H.K., Peters, J., Inasaki, I., Paul, T.: Modelling and simulation of grinding
processes. CIRP Annals 41(2), 677–688 (1992)
[8] Biermann, D., Blum, H., Rademacher, A., Scheidler, A.V.: Simulation of thermal ef-
fects in NC-shape grinding of free formed surfaces using toroid grinding wheels. Part
I: Experimental results. In: Proceedings of the CIRP 2nd International Conference
Process Machine Interactions, Vancouver, BC, Canada, pp. 10.6–11.6 (2010); digital
published
[9] Weinert, K., Blum, H., Jansen, T., Rademacher, A.: Simulation based optimization of
the NC-shape grinding prozess with toroid grinding wheels. Production Engineering –
Research and Development 1(3), 245–252 (2007)
[10] Stautner, M.: Simulation und Optimierung der mehrachsigen Fräsbearbeitung. Tech-
nische Universität Dortmund. Vulkan Verlag, Essen (2006)
[11] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and simulation of
process machine interaction in grinding. Production Engineering – Research and De-
velopment 3(1), 111–120 (2009)
[12] Biermann, D., Mohn, T.: A geometric-kinematical approach for the simulation of
complex grinding processes. In: CIRP International Conference on Intelligent Com-
putation in Manufacturing Engineering, Innovation and Cognitive Production Tech-
nology and Systems, Naples, Italy (2008)
[13] Weinert, K., Blum, H., Jansen, T., Mohn, T., Rademacher, A.: Angepasste Simula-
tionstechnik zur Analyse NC-gesteuerter Formschleifprozesse. ZWF, Zeitschrift für
wirtschaftlichen Fabrikbetrieb 101(7/8), 422–425 (2006)
[14] Kienzle, O.: Einfluss der Wärmebehandlung von Stählen auf die Hauptschnittkraft
beim Drehen. Stahl und Eisen 74, 530–551 (1954)
[15] Biermann, D., Blum, H., Rademacher, A., Scheidler, A.V.: Simulation of thermal ef-
fects in NC-shape grinding of free formed surfaces using toroid grinding wheels. Part
II: Modeling and FE-discretization. In: Proceedings of the CIRP 2nd International
Conference Process Machine Interactions, Vancouver, BC, Canada, pp. 10.6–11.6
(2010); digital published
6 Simulation of Process Machine Interaction in NC-Shape Grinding 141
[16] Blum, H., Kleemann, H., Rademacher, A., Schröder, A.: On solving frictional contact
problems part II: Dynamic case. Ergebnisberichte Angewandte Mathematik 377,
Technische Universität Dortmund (2008)
[17] Blum, H., Jansen, T., Rademacher, A., Weinert, K.: Finite elements in space and time
for dynamic contact problems. International Journal for Numerical Methods in Engi-
neering 76, 1632–1644 (2008)
[18] Rademacher, A.: Adaptive Finite Element Methods for Nonlinear Hyperbolic Prob-
lems of Second Order. Dissertation Technische Universität Dortmund. Verlag Dr.
Hut, München (2009)
[19] Blum, H., Kleemann, H., Rademacher, A., Schröder, A., Wiedemann, S.: SOFAR:
Scientific object oriented finite element library for application and research. Technic-
al report, Technische Universität Dortmund (2002),
http://www.mathematik.uni-dortmund.de/lsx/research/
software/sofar/index.html
[20] Biermann, D., Blum, H., Rademacher, A., Schäckelhoff, M., Scheidler, A.V., Wei-
nert, K.: Bestimmung der Materialparameter des Spindel-Schleifscheiben-Systems
mittels numerischer Parameteridentifikation. Ergebnisberichte Angewandte Mathe-
matik 376T, Technische Universität Dortmund (2008)
[21] Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A.V., Schröder, A.,
Weinert, K.: Space adaptive finite element methods for dynamic Signorini problems
in the simulation of the NC-shape grinding process. In: Denkena, B. (ed.) 1st Interna-
tional Conference on Process Machine Interactions, Hannover, Germany, pp. 309–
316, 3.9.–4.9 (2008)
[22] Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for
dynamic Signorini problems. Electronic Transactions on Numerical Analysis 32,
162–172 (2008)
[23] Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for
dynamic Signorini problems. Computational Mechanics 44(4), 481–491 (2009)
[24] Rademacher, A., Schröder, A.: Goal-oriented error control in adaptive mixed FEM
for Signorini’s problem. Computer Methods in Applied Mechanics and Engineer-
ing 200(1-4), 345–355 (2011)
Chapter 7
Modeling of Process Machine Interactions
in Tool Grinding
7.1 Introduction
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 143–176.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
144 M. Deichmueller et al.
tool grinding it is essential to model both the generation of forces and the genera-
tion of deflections.
In the future, the process machine interaction should be considered in the plan-
ning phase of tool grinding processes to allow the fabrication of high quality tools
with low tolerances and without extensive prototype grinding. Within an interdis-
ciplinary research project of production engineers, mechanical engineers and ma-
thematicians, the authors developed a multi-scale simulation approach, which
represents the process-machine interactions in tool grinding in the microscopic
and the macroscopic scale.
Tungsten carbide tools are produced using deep grinding processes. Within one
production step, the flutes are ground, which can be up to several millimeters
deep, see Figure 7.1.
Fig. 7.2 Modeling concept for the simulation of the process machine interactions in tool
grinding, interfaces and responsible research institutes
is used. Via two axes, the spindle and grinding wheel can be positioned axially in
y and z-direction, whereas the position of the machine table can be defined
over one axial (x-direction) and two rotation axes (A and C), as illustrated in
Figure 7.3.
During the grinding process, forces occur, which deform the machine structure.
Since this has an effect on the quality of the produced workpiece, analyses were
carried out to characterize the deformation of the clamping of workpiece and
grinding wheel and the machine structure.
Fig. 7.3 Axis configuration of the used 5-axis CNC tool grinding machine (Walter Helitron-
ic Power)
Fig. 7.5 Measured deformation of workpiece and machine table for a constant load of
F = 100 N
Since the stiffness of the clamping support has an influence on the total dis-
placement of the workpiece, the elasticity of the machine table is implemented in-
to the workpiece model. Therefore, two spring elements are considered in the
model with a distance of 40 mm within the clamping, see also [4, 5].
Fig. 7.6 Measured surface and Fast Fourier analysis of a grinding wheel topography with
graining D91
150 M. Deichmueller et al.
In each simulation step, the current workpiece’s cross sections are analyzed for
calculating the geometrical moments of inertia Iyy, Izz and Iyz at discrete positions xi
along the workpiece axis x. For example, the geometrical moment of inertia Izz can
be calculated with
bD ⋅ h3
IzzD = D + y2 ⋅ b ⋅ h ,
12 D D D (1)
I zz = ΣI zzD
The static oblique bending of the workpiece is calculated by integrating the li-
near beam equations
of the ground flute and depth of cut ae.. In Figure 7.9, the first eigenfrequency of
a workpiece with longitudinal flutes is shown exemplarily. The variation of the
workpiece characteristics is visible. A finite beam element approach is used as
workpiece discretization, according to the dexel model, to represent the changing
properties, compare [11].
Fig.7.9 Simulated workpiece frequency characteristics depending on the geometry for lon-
gitudinal ground flutes
Each beam element has six degrees of freedom to represent bending and torsion
of the workpiece axis. The properties are assumed to be constant within one ele-
ment but can change over time. The equation of motion can be written as
(3)
Mq + Dq + Kq = f ,
154 M. Deichmueller et al.
with the mass matrix M, the damping matrix D, the stiffness matrix K, the vector
of the generalized coordinates q and the vector of forces and moments f. Because
the system of coupled equations has the size of 6n × 6n, with n the number of
elements, it is very large and takes long to solve. By means of a modal reduction,
the system can be reduced and the equations become linearly independent, which
simplifies the solving process. The modal transformation of the system in the
modal space is carried out by a transformation matrix Φ = qˆ 1 ,qˆ 2 ,...,qˆ n with
qˆ i the eigenvectors of the system. Via the substitution of the node displacements
q = Φqmod (4)
and the multiplication with the transformation matrix from the left side the equa-
tion of motion can be written in modal space as
By using special eigenvectors the modal mass matrix is equal to the identity
matrix; and the modal stiffness matrix has a diagonal shape
= diag{ω12 , ω22 ,..., ωn2 } with the eigenvalues as entries. Depending on the num-
ber of considered eigenvectors, the size of the system is reduced in modal space;
however the information about the neglected modes is also lost.
The reduction of Equation (5) to a system of first order with the substitution
z1 = q mod and z 2 = q mod is
d z1 0 E z1 0
= − − + −1
− − z 2 f mod ,
1 1
dt z2 (6)
z = z + f (t ).
z (t ) = et z (t0 ) + e (t −τ ) f(τ ) dτ .
0
(7)
size Δt. The matrix depends on the geometry of the workpiece and has been
recalculated frequently.
After solving Equation (7) in modal space, the solution of the deformation has
to be retransformed into the physical space by Equation (4).
Fig. 7.10 a) Calculated material removal rate Qwa(i) [mm³/s] and b) calculated equivalent
chip thickness heq [µm] for a tool grinding process (vft = 30 mm/min, d = 10 mm,
R = 62.0 mm)
are measured under the machine table, see Figure 7.3, but experiments suggest
that the stresses vary throughout the contact zone. In order to include this varia-
tion, a process force model has to account the locally different contact conditions.
The contact zone is discretized into small sub-domains by the dexel model (see
7.2.5.1). The range of parameters for the engagement of single grains can be cal-
culated using the geometrical analysis explained in 7.3.1. and [12]. This analysis
also provides a method for the combination of the microscopic stresses stored in a
database (see Figure 7.2) to the stress distribution within the contact zone. Here,
the approach is to model the contact of grains and workpiece on a microscopic
scale for the different parameters of those sub-domains. This is carried out by
coupling variational inequalities describing the elasto-plastic behavior of the
workpiece and its contact with abrasive grains. Linear kinematic hardening is as-
sumed for the workpiece on microscopic scale denoted by Ω. This domain is fixed
on the free boundary ΓD; and the contact boundary is denoted by ΓC. For the case
of notation, only one time-step of the quasi-static problem with initial values set to
zero is considered here (for further details see [13]). The primal problem of elasto-
plasticity with linear kinematic hardening including contact is to find a displace-
ment field u and a plastic strain P such that
− div σ (u, P) = f in Ω,
σ (u, P) = ε (u) − P ,
u = 0 on Γ D ,
σ n (u, P ) = t0 on Γ N , (9)
u n ≤ g ; σ n ≤ 0; (u − g )σ n = 0 on Γ C ,
σ (u, P) − P ∈ ∂j (P).
Here, j (Q) := σ y (Q : Q)1/ 2 , is the non-differentiable part of the minimum plas-
Ω
tic work function with the yield stress σy > 0 in uni-axial tension, σ denotes the
stress, is the elasticity tensor and is the tensor describing the hardening. For
simplicity, the surface traction t0 and the outer volume force ƒ are assumed to be
zero from here on. M : N := MijNij denotes the usual scalar product of M and N.
The part of the boundary ΓN is assumed to be disjoint with ΓD and ΓC.
Since the unique analytic solution W := (u, P) for this partial differential equa-
tion and inclusion is generally unknown, a numerical approximation has to be
computed. For this purpose, the mathematical problem shown in (9) is reformu-
lated as a variational inequality; and then it is equivalently written as a saddle-
point problem introducing Lagrange multipliers. Moreover, the spaces, where the
solution is searched, are replaced by discrete ones. This results in the problem to
find Wh, λC,H, λp,h such that
a( Wh , Z h ) + ΓC
λC , H v h ·n + σ
Ω
y λh, p : Qh = 0
(10)
ΓC
( μC , H − λC , H )u h ·n + σ y ( μ p ,h − λ p ,h ) : Q h ≤ 0
Ω
158 M. Deichmueller et al.
a( W, Z) := (ε (u ) − P ) : (ε ( v) − Q) + P : Q,
Ω
where n is the outer unit normal. It can be shown that the mathematical problem
shown in (10) has a unique solution (Wh, λC,H, λp,h). λC,H is the approximation of
the normal stress on the contact boundary, whereas λp,h is the approximation of
the plastic stress, cf. [14,15]. The finite dimensional space Wh is constructed on
hexahedral meshes. In order to ensure the uniqueness of λC,H, the discretization of
ΓC is carried out with a mesh-size H > h, see [16]. An error estimator presented in
Chapter 3 and [17] and an adaptive refinement algorithm allowing possible multi-
level hanging nodes, cf. [18,19], are used in order to improve the accuracy of the
discrete solution for a fixed number of degrees of freedom.
Another numerical approximation approach is the regularization of the varia-
tional problem obtained from Problem (9), which results in an energy minimization
problem over K := {v ∈ H 01 (Ω ) 3 v ( x )·n ( x ) ≤ g ( x ) on Γ C } of the smooth functional
1
(Z) := a(Z, Z) + jδ (Z) (11)
2
where jδ ( Z) := σ y (Q : Q + δ 2 )1/2 , and δ the regularization parameter. In order to
Ω
use the Newton-Raphson method on the whole space V instead of K, the penalty
functional
1
2 ΓC
Ψ ( v ) := ( v·n − g ) 2+ (12)
is introduced. Here, ρ is the penalty parameter and (ƒ)+ := max(0,ƒ) denotes the
positive part of a real valued function ƒ. The functional Ψ is added to the energy
functional ; and the finite element space Wh is used to compute a discrete ap-
proximation of Wρ,δ. The main drawback of this approach is its dependence on δ
and ρ, cf. [20, 21]. The normal component of the microscopic contact forces Fmicro
is computed from the Lagrange multiplier λc via
Fmicro = λC . (13)
ΓC
If the regularized energy functional and the Newton-Raphson method are used, the
stress must be computed from u with Hooke’s law and the normal stress is inte-
grated over the contact boundary. Once parameterized, the computed values are
stored in the database and missing entries have to be interpolated or newly com-
puted. This parameterization of the microscopic force model is carried out via the
measurement of single grain scratches. Figure 7.11 shows the deformation com-
puted by the finite element approach with regularization after a single grain
scratch. The maximum infeed was set to 5 µm; and the grain moved on a circle
with a diameter of 150 mm. The red color indicates areas where the plowing effect
occurs, i.e. material piles up next to the scratch line. However, the resultant cutting
depth is lower than the nominal depth of cut due to the elastic deflection of tool
and workpiece.
7 Modeling of Process Machine Interactions in Tool Grinding 159
Here, the results of main interest are the forces occurring during scratching.
Figure 7.12 shows the computed forces for different parameter configurations. The
change in the module of elasticity has the biggest influence on the forces.
This reduction of the module of elasticity could be interpreted as damage to the
material. This damage is likely to occur during the scratching itself (for an exam-
ple of brittle material, see [22]) or could be introduced by the pre-treatment of the
workpiece.
Fig. 7.12 Force during scratch test calculated with microscopic Finite Element simulation
Fig. 7.13 Strain calculation by a Finite Element simulation with a priori crack
Fig. 7.15 Relationship between mechanical stresses σn (solid line, ♦) and σt (dashed line,●)
and heq/dlg for a tool grinding process (vft = 15 - 50 mm/min, ae = 0.01 - 1mm, vc = 18 m/s)
The calculated stresses are fitted to a polynomial equation applying the least
squares method.
h eq h eq 2
σ t = 0.63 ⋅105 ⋅
N
+ 3.43 ⋅109 ⋅
d lg d lg mm²
This relationship is applicable for the parameter range shown in Figure 7.15. The
grinding force model has been implemented into the simulation system to visual-
ize the occurring stresses in tool grinding. Figure 7.16 shows the distribution of
the normal stresses in the contact area. It can be seen that the highest pressure acts
at the area with high equivalent chip thickness heq, cp. Figure 7.10 b).
162 M. Deichmueller et al.
Fig. 7.16 Calculated distribution of normal stress σn [N/mm²] for a tool grinding process
(vft = 30 mm/min, d = 10 mm, R = 62.0 mm)
Fig. 7.17 a) Grinding Wheel made by FOS Messtechnik GmbH equipped with a tempera-
ture sensor b) electronics inside the grinding wheel with detector, batteries and wireless
equipment for sending the data to external receiver during grinding
Figure 7.18 shows measurement results using the sensor integrated grinding
wheel. A flute with a helical angle of 30° was ground into a tungsten carbide
workpiece with a diameter of 16 mm. It can be seen, that the temperature signal
(black line) rises to its maximum level of about 210°C when the grinding wheel is
entering the workpiece. The time delay of the temperature signal is due to the fact
that the measurement range of the temperature sensor starts at about 150°C. The
explanation for the higher signal level at the beginning of the measurement is that
the sensor is calibrated for the temperature of tungsten carbide. At the beginning,
there is no contact between grinding wheel and the tungsten carbide workpiece
and the signal is not valid. The process forces, which are measured in machine
tool coordinates (see Figure 7.3), and the measured temperature show similar cha-
racteristics: during the constant engagement conditions both, process forces and
temperatures stay constant. This characteristic implies that the process is thermo-
mechanically stable with constant heat flows due to the flood supply of coolant.
Fig. 7.18 Process forces and temperatures measured with the sensor integrated grinding
wheel (vf = 25 mm/min, vc = 18 m/s, workpiece diameter d = 16 mm)
164 M. Deichmueller et al.
With the help of the sensor integrated grinding wheel it is possible to analyze
the influence of the coolant supply on the grinding process. Figure 7.19 shows the
variation of the coolant jet velocity and the resulting temperature in the contact
and the ratio µ between tangential force Ft and normal force Fn. It can be seen,
that the lowest temperature and friction occurs when the coolant jet velocity is
equal to the cutting speed vc.
Fig. 7.19 Influence of velocity of coolant on temperature and grinding force ratio (process:
vft = 30 mm/min, vc = 18 m/s, workpiece diameter: 16 mm, helix angle: 30°, coolant: min-
eral oil)
Fig. 7.21 Calculated (black lines) and measured grinding forces (oscillating lines) for the
grinding of two flutes of a tungsten carbide tool. (Grinding wheel: Q-Flute-2 D54 C75,
R = 62 mm, Workpiece: Tungsten Carbide, L = 100 mm, d = 10 mm, Process: Down
Grinding, vft = 30 mm/min, vc = 18m/s)
166 M. Deichmueller et al.
Fig. 7.22 Comparison of simulated cross-section using dexel-model and real cross-section
(photograph made with digital microscope); detail shows part of the contour with highest
deviation
workpiece, are predicted precisely by the simulation model, 3) The grinding forces
slightly rise during full-contact phase, both in simulation and measurement. This
effect is caused by the increasing stiffness of the workpiece when approaching the
workpiece clamping and the resulting higher actual depth of cut at the end of each
flute. 4) When grinding the second flute the simulated magnitude of Fy does not
exactly match the measured values.
7 Modeling of Process Machine Interactions in Tool Grinding 167
Apart from the grinding forces, the resulting cross-section of the workpiece has
been simulated and compared with measurements. A Keyence optical microscope
was used to create a high resolution picture of the cross-section of the workpiece
tip, where the maximum shape error occurs. As can be seen in Figure 7.22, the dif-
ference between simulation and measurement is relatively low, which verifies the
high accuracy of the developed and implemented static process machine interac-
tion model. However, the cross-section resulting from grinding a flexible work-
piece is different from the ideal cross-section, which can be simulated using a ri-
gid workpiece, compare Figure 7.30.
The wavy wheel surface causes fluctuations in the process forces due to varying
contact conditions. In Figure 7.23, calculated process forces with an ideal round
grinding wheel and with a grinding wheel including an out-of-roundness of
24 µm, a harmonic wave of 10th order with an amplitude of 1 µm and stochastical-
ly distributed grains of about 0.6 µm are shown. For the simulations, the dynamic
module from Section 7.3.5.3 is included. Additionally, the measured force signals
are also illustrated.
In the first few seconds, the contact length between grinding wheel and work-
piece increases as well as the mean value and the amplitude of the process forces
until the contact length gets quasi-constant. Comparing the mean values it is visi-
ble that the calculation with ideal round grinding wheel fits the measured data well
for the y-direction, whereas for the x-direction the forces are overestimated. The
calculation with the wavy grinding wheel shows a smoother change between run-
ning-in period and constant interval, similar to the measurements. For the y-
direction the forces are underestimated by about 10 N, whereas in x-direction the
simulation is in accordance with the measured data.
When comparing the amplitudes of calculated and measured forces, the high
fluctuation at the beginning of the experiments is not expressed by the simulation.
This is due to ideal initial conditions for the simulation. During the experiments,
168 M. Deichmueller et al.
Fig. 7.23 Calculated process forces Fx and Fy with an ideal round and a wavy grinding
wheel compared with measured data
influences of the coolant, for example, are also included in the measured force
signal, which causes larger amplitudes at the beginning of the process. For the
normal forces in y-direction, the increasing measured force amplitudes are also not
represented by the simulation but quasi-constant amplitudes are reached. This cha-
racteristic can also be related to the coolant, which has a higher influence in nor-
mal direction than in tangential direction. The accuracy could be increased by
means of additional analyses about the influence of coolant.
7 Modeling of Process Machine Interactions in Tool Grinding 169
The consideration of the time histories of measured forces and the resulting
time-varying coefficient of friction show significant periodic fluctuations, which
depend on grinding wheel rotation, amplitude of the process parameters, feed, cut-
ting speed and feed rate. Worth mentioning is the measured phase shift between
the normal and tangential load, which is found in the measurements vary in severi-
ty and comparative measurements with other materials and fixations. As an exam-
ple, the relationship between the local friction coefficient and the acting normal
force FN is displayed for different depths of cut in Figure 7.24.
Fig. 7.24 Consideration of the timing behavior of the friction coefficient as a function of the
normal force and the depth of cut
When comparing the results of the workpiece geometry by including the dynamic
behavior of workpiece and grinding wheel, a difference in the geometry to the
ideal surface is visible, as shown in Figure 7.25.
This difference has its maximum at the tip of the workpiece, when the forces
cause the largest bending and decrease towards the clamping. When comparing
the surface with single measured data points, the calculated geometry error is un-
derestimated but shows qualitatively the same shape. The difference can be ex-
plained by the influence of the coolant, which is not represented in the simulation.
By considering an additionally measured coolant-force component of about 10 N,
as discussed in Section 7.6.2.1, the bending of the workpiece increases and the
geometry error gives a better approximation of the measured data points.
170 M. Deichmueller et aal.
Fig. 7.25 Calculated workpiiece geometry predicted by the dynamic module with and witth-
out coolant consideration
Fig. 7.26 Influence of dexel grid distance on simulation accuracy for calculating the work-
piece cross section of a ground two-flute twist drill (10 mm diameter, 30 ° helix angle)
Fig. 7.27 Influence of simulation time step on simulation accuracy for calculating the
workpiece cross section of a ground two-flute twist drill (dexel grid distance: 0.1 mm)
Fig. 7.28 Influence of feed velocity on calculated process forces accuracy for calculating
the workpiece cross section of a ground two-flute twist drill (simulation time step: 0.5 s,
dexel grid distance: 0.1 mm, for process parameters, see Figure 7.21)
7 Modeling of Process Machine Interactions in Tool Grinding 173
7.6.4 Optimization
The main objective of the research on process machine interaction is to understand
the occurring effects during grinding and to predict process characteristics and
process outcome. With this knowledge, it is possible to optimize grinding
processes already in the process planning phase. In this case, the developed mod-
els have been used to adapt the path of the grinding wheel to reduce the shape er-
rors in the ground workpiece.
The optimization procedure starts with a simulation of the static workpiece def-
lection using the simulation model described in Section 7.6.1. During the simula-
tion, the workpiece deflection is computed at the point, where the grinding wheel
exits the contact area. At that point, the final flute profile is generated. Therefore,
it is a good position to evaluate resulting shape error. A characteristic deformation
profile for a two-flute tool is shown in Figure 7.29. It can be seen that the deflec-
tion constantly rises during the entrance phase of the grinding wheel and then de-
creases quadratically during full engagement due to the increasing stiffness of the
workpiece near the clamping. The deflection v in y-direction of the second flute is
significantly higher because of the reduced stiffness of the workpiece after grind-
ing the first flute.
The next step is the segmentation of the calculated workpiece deformation into
line elements. These elements are then translated into NC commands, which
change the movements of the grinding wheel. With the new NC-program, the si-
mulation of the deformation is repeated until the desired shape of the workpiece is
achieved. In this research, two optimization criteria are used to evaluate the cur-
rent shape of the workpiece. The first criterion is the difference of the core radius
of simulated and ideal cross-section. The second criterion is the dexel-wise com-
parison of the shape of a simulated ideal cross-section.
174 M. Deichmueller et al.
When the current NC-program leads to a workpiece shape fulfilling the quality
criteria, the optimization procedure is stopped. As an exemplary result, Figure 7.30
compares the cross-sections at the tip of a workpiece for two different simulation
runs with the ideal cross-section. The first cross section (left) has been simulated
using the original NC-program. A relatively large deviation of the resulting work-
piece shape from the ideal workpiece shape is visible with a maximum deviation of
160 µm, which corresponds to the deflection presented in Figure 7.29.
The cross-section on the right of Figure 7.30 has been computed using an
adapted NC-program, which was the result of one optimization loop. It can be
seen that the error has been reduced significantly with a maximum deviation of 35
µm. The results indicate that the presented optimization procedure is able to im-
prove the workpiece quality in tool grinding, considering the existing process ma-
chine interactions.
References
[1] European Cutting Tool Association: Zerspanungswerkzeuge: Marktvolumen welt-
weit, Fertigung Sonderausgabe – Werkzeuge 12, 7 (2010)
[2] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and Simulation of
Process - Machine Interaction in Grinding Production Engineering, vol. 3, pp. 111–
120 (2008)
[3] de Payrebrune, K., Kröger, M., Deichmueller, M., Denkena, B.: Investigation on the
Dynamics of Tool Grinding, Machine Dynamics Problems, vol. 33, pp. 92–104
(2009)
[4] Popp, K., Kröger, M., Deichmueller, M., Denkena, B.: Analysis of the Machine
Structure and Dynamic Response of a Tool Grinding Machine. In: 1st International
Conference on Process Machine Interactions, pp. 299–307 (2008)
[5] Popp, K., Kröger, M., Deichmueller, M., Denkena, B.: On Contact Modeling of
Workpiece and Grinding Wheel with Nonlinear Elements. In: 7th EUROMECH Solid
Mechanics Conference, pp. 281–282 (2009)
[6] Chang, H.-C., Wang, J.-J.J.: A stochastic Grinding Force Model considering Random
Grit Distribution. International Journal of Machine Tools and Manufacture 48, 1335–
1344 (2008)
[7] van Hook, T.: Real-Time Shaded NC Milling Display ACM SIGGRAPH Computer
Graphics, vol. 20, pp. 15–20 (1986)
[8] Denkena, B., Böß, V.: Technological NC Simulation for Grinding and Cutting
Processes Using CutS. In: 12th CIRP Conference on Modelling of Machining Opera-
tions, vol. II, pp. 563–566 (2009)
[9] Biermann, D., Mohn, T.: A Geometric-Kinematical Approach for the Simulation of
Complex Grinding Processes. In: 6th CIRP International Conference on ICME (2008)
[10] Denkena, B., Tracht, K., Yu, J.-H.: Advanced NC-Simulation based on the Dexelmo-
del and the HRMC-Algorithm Production, Engineering, vol. XIII(1), pp. 91–94
(2006)
176 M. Deichmueller et al.
[11] de Payrebrune, K., Kröger, M.: Modal Reduction of Dynamics of Tool Grinding. In:
2nd International Conference on Process Machine Interactions (2010)
[12] Denkena, B., Deichmueller, M., Kröger, M., Popp, K.M., Carstensen, C., Schroeder,
A., Wiedemann, S.: Geometrical Analysis of the Complex Contact Area for Modeling
the local Distribution of Process Forces in Tool Grinding. In: 1st International Confe-
rence on Process Machine Interactions, pp. 289–298 (2008)
[13] Han, W., Reddy, D.: Plasticity. Springer (1999)
[14] Brokate, M., Carstensen, C., Valdman, J.: A Quasi-Static Boundary Value Problem in
Multi-Surface Elastoplasticity. I: Analysis, Math. Methods Appl. Sci. 27(14), 1697–
1710 (2004)
[15] Schröder, A., Wiedemann, S.: Error Estimates in Elastoplasticity using a Mixed Me-
thod, Humboldt Universität zu Berlin, Institute of Mathematics, Preprint 11-01 (2011)
[16] Schröder, A.: Mixed Finite Element Methods of Higher-Order for Model Contact
Problems. Humboldt Universität zu Berlin, Institute of Mathematics, Preprint 09-16,
submitted to SINUM (2009)
[17] Schröder, A., Wiedemann, S.: Humboldt Universität zu Berlin, Institute of Mathemat-
ics, Preprint 11-02 (2011)
[18] Schröder, A.: Constraints coefficients in hp-FEM. Numerical Mathematics and Ad-
vanced Applications, 183–190 (2008)
[19] Schröder, A.: Constrained approximation in hp-FEM: Unsymmetric subdivisions and
multilevel hanging nodes. LNCSE, vol. 1(76), pp. 317–325 (2009)
[20] Han, W., Reddy, D.: On the Finite Element Method for Mixed Variational Inequali-
ties arising in Elastoplastiity. SINUM 32(6), 1778–1807 (1995)
[21] Brokate, M., Carstensen, C., Valdman, J.: A quasi-static Boundary Value Problem in
Multi-Surface Elastoplasticity. II: Numerical Solution, Math. Methods Appl.
Sci. 28(8), 881–901 (2005)
[22] Wang, C.Y., Clausen, R.: Marble Cutting with Single Point Cutting Tool and Di-
amond Segments. Machine Tools & Manufacture 42, 1045–1054 (2002)
[23] Lemaitre, J., Desmorat, R.: Engineering Damage Mechanics. Springer (2005)
[24] Deichmueller, M., Denkena, B., de Payrebrune, K.M., Kröger, M., Wiedemann, S.,
Schroeder, A., Carstensen, C.: Determination of Static and Dynamic Deflections in
Tool Grinding using a Dexel-Based Material Removal Simulation. In: 2nd CIRP In-
ternational Conference on Process Machine Interactions (2010)
[25] Tönshoff, H., Peters, J., Inasaki, I., Paul, T.: Modelling and Simulation of Grinding
Processes. CIRP Annals - Manufacturing Technology 41, 677–688 (1992)
Part III
Cutting
Chapter 8
HPC - Stability Simulation
8.1 Introduction
For a long time, the technologies of High Performance Cutting have contributed to
the fact that material removal rates in milling processes have greatly increased due
to new cutting materials and tool geometries [14]. The machine technology has al-
so been adapted to the requirements [1], [7]. Lightweight design coupled with a
high degree of rigidity permits high acceleration values and feed rates in the main
axes. Main spindle systems have also been subject to further development and
today offer rotating speeds and torques permitting highly-efficient machining
operations.
However, the productivity of machine tools is often limited, not by the installed
drive performance but by the regenerative vibrations, the so-called chatter [4],
[12]. In such cases, the machining process becomes instable due to the force com-
pliance interaction between the machine tool structure and the cutting process [3],
[6]. Under such extreme operating conditions, the result is occasionally an escalat-
ing shortening of the durability of individual machine components and breakage
of tools. In addition, instable processed workpieces are rejected due to surface
damage.
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 179–201.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
180 C. Brecher, R. Hermes, and M. Esser
In this field, the simulation of stability charts offers a cost and time-efficient
possibility of parameterizing machining processes with regard to their stability be-
havior or to use the process stability as an assessment criteria in the analysis and
design of machine systems [5], [11]. Today, there are numerous simulation possi-
bilities for predicting optimized process parameters, whose limited quality of re-
sults, however, prevents a sensible application to the planning of production. The
same applies to computer-aided evaluation of stable machining performance in the
design phase of machine systems [16].
The aim of the work accomplished in the priority program SPP 1180 “HPC sta-
bility simulation” and described in this article is the improvement of the accuracy
of simulated stability charts to determine process-optimized working parameters.
For this purpose, the sub-systems machine tool structure and cutting process
as well as their simulative depiction in the stability simulation regarding their in-
fluences on errors on the simulation results are determined first. Based on a sys-
tematic analysis, the potential for improvement is shown and implemented in the
modeling of the sub-systems. Besides the individual consideration of the machine
structure and machining process, their coupling and the method of stability simu-
lation concerning potential influences of errors are also analyzed. Here, the influ-
ences on simulation accuracy due to pre-determined simulation parameters are
analyzed as well as the program structure of the simulation models used in the
WZL regarding calculation time-efficiency and usability.
The knowledge obtained by the work carried out is summarized in the newly-
developed PrimeCut simulation software. Besides efficiently computing a stability
analysis of machining processes, the software allows for a calculation of surface
structures of processed workpieces.
position of the rolling element. This effect causes a change in the radial and axial
bearing rigidity. The influences of thermal expansion and centrifugal force in
spindle bearings were investigated on individual test rigs [8] in the works and their
significance was shown in the work of [17].
Apart from the influence of similar spindle components, there is a change in the
compliance behavior of main spindle systems under revolution due to rotor-
dynamic effects at the spindle shaft itself (Fig. 8.1). Gyroscopic and centrifugal
forces influence the natural frequency and compliance amplitudes of the dynamic
behavior at the TCP. Dynamic process loads and the resulting bending of the spin-
dle shaft cause additional balancing influences. The spindle shaft with its slim
structure does not resemble the idealized Laval runner, yet follows the rules of ro-
tor-dynamics.
ω ω
Axis of
gyroscope
Resultant
displacement Static and dynamic unbalance
Reference: Mode shape of spindle nose Dynamic unbalance Static unbalance
Axis of gyroscope
Fig. 8.1 Rotor-dynamics of the spindle shaft and the tool [11]
For slim rotors, the resonance frequency is reduced due to the occurring gyros-
copic force. In a resonance case, the TCP conducts an elliptical movement in the
same direction of rotation as that of the tool. For external excitation of the rotating
spindle shaft, for instance due to dynamic process loads or imbalances because of
the bending of the shaft, a deflection in the direction opposite to the rotation is su-
per-imposed. Decisive for the simulative modeling of stability behavior is the ac-
tual dynamic compliance of the main spindle under the influence of rotation. For
this reason, the focus of the work being carried out is on the metrological determi-
nation of this behavior.
Rigidity [N/µm]
Loose bearing TCP 300 radial
Claw clutch with ball lining Fixed
bearing 200
High speed spindle in open construction without axial
100
peripheral components
Belt drive gear and claw clutch 0
0 10000 20000 30000
nmax = 24.000 1/min Spindle speed [1/min]
Fig. 8.2 Test rig for the assessment of a rotating system [11]
Clutch
Spindle shaft
Loose bearing
support
Impulse exciter
Ball lining (shock pendulum)
Fixed bearing
support
Mounting plate
Fig. 8.3 Modal analysis of the test rig spindle under rotation [11]
The first section of the signal comprises the duration of the impulse excitation
and the decay of the deflection response. The second section serves as a reference
and contains only parts of the interference signal. The signal parts are aligned with
each other by means of a Best-Fit process and the interference signal is eliminated
by means of subtraction.
Fig. 8.4 shows the metrological acquired dynamic compliance behavior at TCP
for various speeds of revolution. Also shown here are the measurements at rest
and for 1,000 1/min in order to determine variation of the compliance behavior be-
tween still-standing spindle systems and systems with a very slow revolution
speed. Thermal effects as well as centrifugal force, out-of-balance and gyroscopic
effects exert no influence at low spindle speeds of 1,000 1/min. However, the
load-bearing lubrication film, which forms in the bearing under rotation, influ-
ences the damping properties of the bearing and thus the dynamic compliance be-
havior of the overall system. Thus, the measurement shows a slight change in the
resonance peaks to higher damping and less compliance.
In Fig. 8.5, the frequency ranges of the dominant mode shapes, here the shaft
bending of first, second and third order, are depicted. These mode shapes can
cause instabilities in the milling process but change their respective characteristics
in form and value of the amplitude superposition in dependence of the spindle
speed. All three mode shapes change in different ways at higher speeds of revolu-
tion. The bending vibration of the first order at 1,000 Hz reduces its amplitude to
approximately 40 % and splits into two mode shapes. A clean differentiation of
the two resulting peaks in the modal analysis is not possible. At 24,000 1/min, a
new mode shape arises at approximately 800 Hz, which can also be identified in
the modal analysis as a bending vibration of first order.
184 C. Brecher, R. Hermes, and M. Esser
1
First order Second order Third order
bending vibration bending vibration bending vibration
Compliance [µm/N]
0.1
<
0.01
Spindle speed
0
0.001
1000 1/min
180
15.000 1/min
90 24.000 1/min
Phase [°]
-90
-180
0 1000 2000 frequency [Hz] 4000
In the frequency range of the bending of the second order, at around 2200 Hz,
the amplitude rises and the phase response becomes flatter when the spindle speed
increases. The third dominant mode shape at 2,500 Hz shows a behavior resulting
from a significant influence on the gyroscopic forces. The resonance peak splits
into two peaks. A creation of a second resonance peak can also be clearly recog-
nized in the phase response. Fig. 8.5 shows the evaluation of the modal analysis
for the mode shapes of third order. For low spindle speeds, a definite oscillation of
the spindle tip in the direction of the excitation force can be seen in the results of
the modal analyses (second figure from the left). The mode shape changes at high-
er spindle speeds. A circular movement of the TCP occurs from the linear bending
vibration. This circular movement in the respective low-resonance frequency is in
n n
dx Fx
dy
Mode shapes of third order bending vibration
Measurement in six n [min-1] f [Hz] n [min-1] f [Hz] n [min-1] f [Hz]
levels
0 3515 15.000 3458 15.000 3531
Detection of two
directions 1.000 3497 24.000 3424 24.000 3552
the same direction as the rotation of the spindle and for the respective higher fre-
quency opposed to the spindle rotation (third and fourth figures from the left). The
behavior then corresponds to those of the gyroscopic effect expected for strong
angular deviation of the shaft.
1
Compliance [µm/N]
10-2 0.1
24
10-3
18 in]
0 /m 0.01
00
1000 12
d [10 Measurement
2000 ee
3000
6 sp Simulation
dle 0.001
Frequency [Hz] 4000 0 in 0 1000 2000 3000 4000
Sp Frequency [Hz]
For measuring the compliance behavior on original main spindle systems the
same measurement setup is used as applied to the auxiliary spindle test rig. How-
ever, this method is not suitable for the measurement at the original milling tools
because the cutting edges and chip spaces of the mill permit neither contact excita-
tion nor an exact measurement of the deflection. For this purpose, the milling tools
are replaced by cylindrical carbide rods, so-called dummy tools, and mounted into
the spindle system to be investigated. For depicting the real compliance behavior,
the dummies have the same moment of inertia as the original tools (Fig. 8.7 left).
Comparable measurements between the original and dummy tool with the spindle
at rest provide very good approximations of compliance behavior.
Fexc Fexc 1
mm
mm
0
Phase [°]
-90
Dummy- and original- tool
cemented carbide in tool holder
D = 20 mm, Lges = 163 mm -180
0 1000 2000 Frequency [Hz] 4000
The measurement results for a dummy tool are shown exemplarily in Fig. 8.7
for various spindle speeds. As for the auxiliary spindle test-rig, the measurement
results show a significant change of dynamic behavior over the speed range. Be-
cause of the good comparability of the measuring results at the original and dum-
my tools, the measurement results determined on the dummy will be used for the
stability calculations in the following. The simulation is taking into account the ef-
fects of spindle rotation and gyroscopic effects on the original main spindle sys-
tem. Fig. 8.8 shows the comparison between two simulated stability charts with
and without taking into account the influence of the spindle speed as well as the
cutting depth limit determined in machining tests. A shift of the stability maxi-
mum due to the consideration of spindle speed and an improvement of simulation
results in comparison with the stability limit determined in the machining tests can
be clearly identified. Clear improvement potential by taking into account rotation-
al speed-dependent compliance behavior is provided mainly in the region of high-
speed cutting of aluminum. Due to the rigid structure of the machine and the high
rotation speed of the mill (stable region beyond the highest stability minimum),
the dynamic behavior of the machine structure in the lower frequency range in
8 HPC - Stability Simulation 187
comparison with the dynamic behavior of the cutting tool and main spindle system
is irrelevant for the occurrence of dynamic chatter vibrations.
Apart from the depiction of the dynamic machine behavior, modeling of the
dynamic process behavior presents the second important aspect of stability simula-
tion. Linear, exponential and mechanical force models have been the state of the
art, [3], [20], for a long time but they ignore potential influences in the analytical
description of the process forces. Some authors [15], [18] point out that a real ma-
thematical relationship between chip thickness and dynamic cutting force is not
sufficient in all cases.
30 Comparison
Cutting tests
Cutting depth ap [mm]
of measurement
24 Stable cut and simulation
Chatter Power
li mit Cutting tool:
18 Chatter frequency
Mitsubishi
12 BXD 4000
D = 32 mm
2 cutting edges
6
Material:
0 Simulation with spindle aluminium 7010
Chatter frequency fR [Hz]
More detailed investigations show that the dynamic deflection of the cutting edge
inducing a highly-dynamic chip thickness variation must be included in the mod-
eling of the machining process. Examples are the consideration of the process
damping (effect of the contact between tool flank and workpiece surface) or the
deduction of an effective direction-corrected cutting speed [2] [9], [13]. The use of
a test-rig is necessary in order to be able to determine the influence of the cutting
dynamics on the cutting forces for a large range of process parameters. The in-
fluencing dimensions of the dynamic process force creation defined in the test rig
experiments are analyzed in further procedures to derive an analytical model with
improved reproduction accuracy of the cutting force generation at the cutting
edge.
188 C. Brecher, R. Hermes, and M. Esser
the three spatial directions. The inertia forces of the tool holder and the cutting
edge also acquired by the force sensors during the oscillation movement are ma-
thematically compensated by a down-streamed measuring program.
y z
Spindle position measurement
with rotary encoder on the spindle x
Fig. 8.9 Guide and workpiece concept for dynamic cutting tests [11]
hardware for signal conditioning. Apart from this, the effects of the mechanical
structure behavior and the inertia of the test-rig components have been taken into
account. For this purpose, a suitable user interface in the LabView software envi-
ronment has been created for the operation of the test-rig as well as for data acqui-
sition and data evaluation.
Support for
piezo actuator
Piezo actuator
Adapter
Flexure joints
y z
ard
ebo
Bas
Fig. 8.10 Design of the overall test-rig for dynamic cutting tests [11]
x o ( t ) = x̂ i ⋅ sin(ωt − ε) (8.2)
′ ,c (t ) = Fˆdyn
Fdyn ′ ,c ⋅ sin(ωt + ϕ Fc )
(8.3)
= xˆ i (Rci sin(ωt + ϕ ci ) + Rco sin(ωt + ϕ co − ε ) )
8 HPC - Stability Simulation 191
′ , p ⋅ sin(ωt + ϕ Fp )
′ , p (t ) = Fˆdyn
Fdyn
= xˆ i (R pi sin(ωt + ϕ pi ) + R po sin(ωt + ϕ po − ε ) )
(8.4)
The modeling approach is based on a complex dynamic force influence from the
chip outer side as well as from the chip inner side, whose wave profiles are shifted
against each other. Eq. 8.1, Eq. 8.2. xi and xo describe the compliance for the inner
and outer sides of chip thickness modulation. xo is shifted backwards to xi in terms
of time about the positive angle ε. All phase positions are referenced to the mod-
ulation of the inner chip thickness, thus to the current relative compliance between
tool and workpiece. p with reference to the chip width are calculated. The interre-
lation between chip thickness modulation and the resulting dynamic cutting force
F’dyn,c resp. passive force F’dyn are shown in Fig. 8.11. They apply for the equation
of the cutting force (Eq. 8.3) and also for the calculation of the passive force (Eq.
8.4) with different parameters. The resulting dynamic force is comprised of a part,
which the outer chip thickness modulation is responsible for, and which is scaled
in relation to this with Rjo and shifted about φjo in phase, as well as a correspon-
dingly scaled (Rji) and phase shift (φji) of the inner chip thickness modulation. The
two dynamic force components are added to the total passive or cutting force re-
spectively. With that, the displacement of the cutting edge exclusively along a sin-
gle direction is possible.
Reference force
inner and outer chip
Cutting tool displacement
φo
Ro
F‘o(t)
F‘i(t) φi
Ri
Xi(t)
Inner chip
thickness modulation Time
With that, the dynamic displacement of the cutting edge exclusively along the
modulation of both sides of chip thickness, the parameters for the inner and outer
modulation cannot be determined directly from individual force and deflection
measurements. Instead, the parameters are numerically calculated from the mea-
surement results of several cutting tests. For this purpose, cutting tests with the
192 C. Brecher, R. Hermes, and M. Esser
same process parameters but various phase angles ε are performed. The variation
of the phase angles is set over a small change of rotation speed between the tests.
The round and square symbols in Fig. 8.12 (left) represent the individual experi-
ments with different phase angles ε under otherwise identical conditions.
Fig. 8.12 on the left side for the machining of steel material C45E shows the
progression of the cutting forces referenced to the cutting width and cutting depth.
The phase angle given is referenced to the modulation on the inner cutting side
(xi). Because of this reference the phase angles of the cutting force are usually
attributed to the region of the second and third quadrants, thus between -90°
and -270°. It must be noted that the absolute value of the cutting force reaches its
maximum approximately at a phase angle of ε = 180° and almost reaches a value
of zero when the upper and lower chip thickness modulation are in phase (ε = 0°).
However, the passive force hardly varies. Thus, it is mainly caused by the modula-
tion of the inner chip thickness.
-90 750
2000
-180 0
1000
-270 -750
Chip shape:
-360 -1500
0 90 180 270 360 -3000 -2000 -1000 0
Phase angle ε [°] Real part [N/mm²]
Cutting force: Amplitude phase Passive force: Amplitude phase Cutting force Passive force
Fig. 8.12 Illustration and fitting of the dynamic cutting forces [11]
The extended lines represent the prediction of the introduced force model,
whose parameters have been calculated with the aid of the measuring values ac-
cording to a best-fit process. Thus, the equation system can only be uniquely
solved with four scalable results for four unknown parameters, it can be deter-
mined that two arbitrarily selected measurement results (respectively amplitude
and phase angle, that means four scalar values) are always sufficient to define the
four model parameters, respectively for cutting and passive force. On the right
side in Fig. 8.12, the same measuring values are presented in the complex plane.
The shown curves illustrate the rotation of the force vectors from the outer mod-
ulation around the peak of the force vector of the inner modulation. Here, the
small diameter of the passive force curve shows that the passive force is generally
determined by the inner chip thickness modulation and has no linear dependency
on the chip thickness. Conventional force models cannot represent this behavior.
8 HPC - Stability Simulation 193
The cutting force model introduced in the complex domain offers extended
possibilities depicting the real process force behavior in a simulative manner and,
as a result, to take into account numerous variables of process parameterization
and dynamic machine behavior. Within the test-rig investigations, the relevance of
individual influences on the progression of the process force has been investigated
and parameterizations for the calculation of the process forces have been deter-
mined. As an example, the potential of the dynamic cutting force model in the de-
piction of the influence of clearance flank wear on the process force is shown. Fig.
8.13 shows the resulting model parameters for the complex dynamic cutting force
model. Accordingly, at 120 Hz the tool wear has no influence. However, the influ-
ence rises drastically at higher oscillation frequencies. Cutting and passive force
coefficients of the inner chip thickness modulation increase strongly. The influ-
ence of the outer chip thickness modulation becomes less via the frequency.
Oscillation frequency
f = 120 Hz f = 600 Hz f = 1200 Hz
Amplitude coeff. [N/mm2]
4000
Test parameters
3200 RRoc vc = 200 m/min fU = 0.1 mm/U
RRic Material: C45E tool: R245 M-PM
2400
1600
800
0 R
Rop
45 R
Rip
0
ϕϕoc
Phase coefficients [°]
-45 ϕϕic
90
45
0
ϕop
-45 ϕip
-90
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Wear mark width [mm] Wear mark width [mm] Wear mark width [mm]
The phase shift of the passive force of the inner chip thickness modulation also
changes. For higher frequencies, the process can be damped and stabilized by
means of the effect of the time-shifted modulation of the inner and outer chip
thickness. A visualization of the effects of the process damping is given in Fig.
8.14. Here, simulation calculations for surface-overlapping between tool and
workpiece are carried out for a planar cutting process at a wavelength of 1.7 mm
and vibration amplitude of 0.05 µm. It is assumed that the collision between clear-
ance flank and workpiece surface causes additional forces, which lead to the ef-
fects shown in Fig. 8.13. This explains the positive phase shift of the passive force
to the inner chip thickness modulation. The force precedes the chip thickness as
there is already clearance flank contact with the inward movement of the cutter.
Thus, the cutter is already subject to the maximum reactional force before reach-
ing the deepest point.
194 C. Brecher, R. Hermes, and M. Esser
Vibration R = 50µm
α = 11°
α = 11° α = 11°
Width of wear mark
= 0,05mm
2.5 Overlap area 2.5 2.5
Vc [10-3 mm²]
2.0 2.0 2.0
0 0.4 0.8 1.2 1.6 0 0.4 0.8 1.2 1.6 0 0.4 0.8 1.2 1.6
Vc Cutting path [mm]
Cutting path [mm] Cutting path [mm]
Fig. 8.14 Flank face contact with a rounded and a worn tool [11]
The deviations of the cutting experiments in the region of low revolutions can
thus be derived from different effects. At low cutting speeds, damping effects can
become more effective as these are taken into account in the parameter identifica-
tion. Different friction effects in chip formation, which may be responsible for
such deviations, are described by Dudzinski and Molinari [10]. In addition, espe-
cially for low cutting speeds, adhesion effects with particles of the workpiece ad-
hering to the tool also influence the dynamics of the cutting process and thus the
stability limit.
Apart from the accuracy of the modeling and the input parameterization, the me-
thod of stability simulation itself has an influence on the quality of the simulation
results. The coupling of the modeling approaches described in the previous
chapters for process and machine behavior in time domain is described in the fol-
lowing. For the simulative investigations described here, the dynamic compliance
behavior of the machine structure is used as input parameter in form of a matrix of
direct and cross-frequency response functions. In the multi-dimensional stability
simulation, these frequency response functions (FRF) are taken into account by
means of the projection of process forces into the Cartesian spatial directions of
the machine system. However, these FRF cannot be processed by time-based si-
mulation algorithms and must therefore be treated as differential equations. The
overall behavior of the machine structure is thus depicted as an accumulation of
damped single-mass oscillators. All single-mass oscillators belonging to one direc-
tion are loaded with the same force in the simulation; the resulting deflections are
added together. The measured or simulated dynamic behavior of machine tools
can be described in this manner with a high degree of accuracy.
The dynamic cutting rigidity is coupled with the dynamic machine compliance
in the simulation model. Fig. 8.16 describes this coupling for a three-dimensional
milling process and several cutting edges. The model is illustrated in a simplified
manner concerning the consideration of the coordinate directions. Here, on the one
hand, a planar milling process with feed motion of the milling cutter in the x-y-
level is assumed. On the other hand, the dynamic cross compliances of the FRF
are ignored in the figure. The direction factors di,xd are set to zero while the respec-
tive cutting edge is not engaging into the workpiece. In this case, a force feedback
does not occur at this cutting edge. If a cutting edge is in engagement, the generat-
ed profile xdi is used for the cutting force calculation at the corresponding cut. The
profile xdi is delayed by the dead time Ti,i+1 between two cutting sequences and the
profile of the chip outer side is passed to the next cutter i+1. The value of the dead
times Ti,i+1 are the same for equally-divided cutters but can deviate from each oth-
er for a non-uniform pitch. Here, the cutting forces are determined over the linear
dynamic cutting force coefficient kcb by multiplying the dynamic part of the chip
196 C. Brecher, R. Hermes, and M. Esser
thickness with kcb and the chip width b. The angle of the cutting force compared
to the chip thickness change (ßorth) is included in the direction factor dF,Fj, which
projects process forces from the coordinate system of the cutting edge to the ma-
chine coordinate system.
On the right side of Fig. 8.16, the sub-model of the second cutting edge is
shown in a sectional view. The lower part of the picture also displays the sub-
model of the same cutting edge but the model is extended such that the developed
complex force model can be shown. Here, xd1 and xd2 serve as input dimensions.
However, both are taken into account separately and multiplied with the force
coefficient Rpq for calculating the dynamic cutting and passive force and shifted
timewise by the phase angle φpq. The phase shift in the shown control loop is then
also taken as a dead time element. The dead time necessary for the depiction of the
phase shift is therefore dependent on the frequency of the respective active mode
shape.
Process model milling with n cutting edges Depiction of one cutting edge
Fx xges xd1
GFx,x Cutting edge 2
Fy yges Fx u2 xd2 xges
GFy,y dF,Fx b kcb dx,xd
Fy - yges
u1 xd1 dF,Fy dy,xd
dF,Fx b kcb - dx,xd
xd2
dF,Fy dy,xd
Essential enhancement for the complex force model
T1,2 xd1
Cutting edge 2, complex force model
u2 xd2 Fx F xd2 xges
dF,Fx b kcb dx,xd dFc,Fx c b - Fic ϕic Ric dx,xd
- Fy yges
dF,Fy dy,xd dFc,Fy Fip dy,xd
ϕip Rip
T2,3 Foc
With di,xd = 0 ϕoc Roc
dFp,Fx
if cutting edge i Fop
is not in contact dFp,Fy b - ϕop Rop
Fp
Tn-1,n
xdn xd2
un
dF,Fx b kcb - dx,xd
Assumption of a plane milling process
dF,Fy dy,xd
Neglecting the cross frequency response
Tn,1 Neglecting the nominal depth of cut
The resulting dynamic force parts are summarized into dynamic cutting and
passive force. They are projected by means of cutting angle-dependent direction
factors into the Cartesian directions x and y of the machine tool and act again on
the structural compliance. The simulation structure was extended in order to set
the correct dead time to depict the phase angle φpq for a resonance frequency and
to take into account the identified influence of the frequency dependency on the
dynamic process force progression. On the right, Fig. 8.17 shows that simulation
loops must be created for the different mode shapes of the machine tool. Thus, an
interpolation of the model parameters is carried out for each mode shape and the
dead times for depicting the phase angle are calculated as well.
8 HPC - Stability Simulation 197
15
Comparison
Surface profile [µm]
10 of measurement
and simulation
5 Cutting tool:
Sandvik
0 R390-32T16-11M
-5 D = 32 mm
3 cutting edges
-10
32 34 36 Groove length [mm] 40 Material:
steel C45E
Profile spectrum [µm]
6 Measurement
Simulation Process:
4 n = 1200 1/min
ae = 32 mm
Profile frequency referred ap = 5.4 mm
2 to feed rate fz = 0.12 mm / tooth
0 Chatter vibrations
0 5 10 Frequency [Hz] 20 at approx. 50 Hz
fchatter n
Work piece
Disc springs
Spindle speed n = 900/min
Depth of cut ap = 3 mm
Flexure joints Base plate Exciter freq. fr = 120 Hz
8.6 Conclusion
Within the framework of the SPP 1180, the possibilities for the stability simula-
tion of High-Performance-Cutting (HPC) processes have been improved by inves-
tigating the input data for the parameterization of the simulation models and the
methods of simulation itself. The depiction of the machine tool dynamics and of
the dynamic behavior of the cutting processes has been extended.
The dynamic behavior of the machine structure was emphasized to improve the
stability simulation. The main effect identified in the HPC processing of alumi-
num alloy is the change of the dynamic behavior of spindle-bearing systems dur-
ing high rotation speeds. A measurement setup and the corresponding analysis
software were developed to evaluate this effect. Besides, the rotation speed de-
pendent dynamic behavior could be investigated in detail by means of an auxiliary
spindle test-rig comprising an accessible spindle shaft for modal analysis. The
investigations have shown that substantial deviations from the behavior of the
spindle at rest occur at higher spindle speeds. Although current tools for structural
simulation can represent individual rotor-dynamic effects, they are not yet suitable
for the identification of the structural behavior at high spindle speed. The possibil-
ities of the metrological determination created are thus necessary in order to carry
out stability simulations for milling processes at high spindle speeds.
Furthermore, investigations of the dynamic process behavior were presented. A
test rig was developed and built up for direct metrological determination of the
dynamic cutting rigidity. This permits the exact and independent variation of
process and oscillation parameters for the orthogonal milling process. A large
number of test-rig experiments with different cutting edges and workpiece mate-
rials were the basis for a presentation of the dependencies in the determination of
dynamic milling forces. Especially cutting with rounded or blunt cutters has
200 C. Brecher, R. Hermes, and M. Esser
shown that the usual linear and real depiction of the dynamic cutting rigidity is not
sufficient as input parameter for stability simulation. Much rather, a phase shift
between force and chip thickness modulation in the cutting process, i.e. a complex
linking of both dimensions, must be taken into account. A corresponding model
has been presented and parameterized for several combinations of cutting edges
and materials. The use of conventional models has shown to be sufficient for cut-
ting aluminum with sharp cutters and high cutting speeds. The depiction and
processing of steel materials, however, must be carried out with the more complex
models.
Beyond the simulation of the process stability, the prediction of the surface pro-
file of instable machined workpieces can also be useful for the identification of re-
cursive weak points. An appropriate method should be derived for this purpose in
the remaining project time of the SPP 1180.
References
[1] Abele, E., Munirathman, M., Tschannerl, M.: Hochgeschwindigkeitsbearbeitung
(HSC) - technologische Leistungsfähigkeit und noch zu überwindende Hindernisse.
In: Proceedings CIRP Int. Conference High Performance Cutting, Aachen (2004)
[2] Altintas, Y., Eynian, M., Onozuka, H.: Identification of Dynamic Cutting Force Coef-
ficients and Chatter Stability with Process Damping. Annals of the CIRP, Jg.57, H.1,
S.371–S.374 (2008)
[3] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. Annals of
the CIRP, Jg. 53, H.2, S.619–S.642 (2004)
[4] Andrae, P.: Hochleistungszerspanung von Aluminiumknetlegierungen. Diss. Univer-
sität Hannover (2002)
[5] Brecher, C., Esser, M., Paepenmüller, F.: Motor Spindles for HPC: Testing and Chat-
ter Simulation. In: Proceedings 2nd CIRP Int. Conference High Performance Cutting,
Vancouver, Canada (2006)
[6] Brecher, C., Esser, M., Witt, S.: Interaction of Manufacturing Process and Machine
Tool. Annals of the CIRP, H. 58 (2009)
[7] Brecher, C., Witt, S.: Static, Dynamic and Thermal Behaviour of Machine Tools with
Regard to HPC. In: Proceedings CIRP Int. Conference High Performance Cutting,
Aachen (2004)
[8] Butz, F.: Gestaltung der Loslagerung von Werkzeugmaschinenspindeln. Shaker, Aa-
chen (2007)
[9] Clancy, B.E., Shin, Y.C.: A Comprehensive Chatter Prediction Model for Face Turn-
ing Operation Including Tool Wear Effect. International Journal of Machine Tools
and Manufacture Jg.42, S.1035–S.1044 (2002)
[10] Dudzinski, D., Molinari, A.: Metal cutting and high speed machining. Kluwer, Bosten
(2002)
[11] Esser, M.:: Stabilitätssimulation für das HPC-Fräsen. Dissertation. Aachen. RWTH
Aachen, Werkzeugmaschinenlabor, WZL (2010)
[12] Großmann, K., Mühl, A., Löser, M.: Prognose von Stabilitätsgrenzen für das Fräsen.
ZWF Jg.101, H.7/8, S.416–S.421 (2006)
[13] Kals, H.J.J.: On the Calculation of Stability Charts on the Basis of the Damping and
the Stiffness of the Cutting Process. Annals of the CIRP, S.297–S.303 (1971)
8 HPC - Stability Simulation 201
[14] Munirathnam, M., Kreis, M.: Das Gesamtpaket muss stimmen - Trendbericht: Sys-
temdenken bei der HSC / HPC - Zerspanung steigert die Produktivität. In: WB
Werkstatt und Betrieb, H. Bd 139 Heft 6, pp. S.42–S.50 (2006)
[15] Peters, J., von Vanherck, P., van Brussel, H.: Die Messung der dynamischen
Schnittkraftkoeffizienten. In: Fertigung, H. Bd 3 Heft 2, pp. S.57–S.65 (1972)
[16] Schulz, A.: Systeme zur Schwingungsdämpfung von Werkzeugmaschinen. In: VDI-
Fortschrittberichte, Jg. 651 pp. S.40–S.56 (2005)
[17] Spachtholz, G.: Erweiterung des Leistungsbereiches von Spindellagern. Apprimus,
Aachen (2008)
[18] Tlusty, J.: Analysis of the State of Research in Cutting Dynamics. Annals of the
CIRP, Jg. 27, H. 2, S.583–S.589 (1978)
[19] van Brussel, H.: Dynamische Analyse van het Verspaningsproces. Katholieke Un-
iversiteit Leuven, Dissertation (1971)
[20] Weck, M., Brecher, C.: Werkzeugmaschinen. Automatisierung von Maschinen und
Anlagen, vol. 6. Springer, Auflage (2006)
[21] Werntze, G.: Dynamische Schnittkraftkoeffizienten. Bestimmung mit Hilfe des Digi-
talrechners und Berücksichtigung im mathematischen Modell zur Stabilitätsanalyse.
RWTH Aachen, Dissertation (1973)
Chapter 9
Development of a Stability Prediction Tool
for the Identification of Stable Milling Processes
Abstract. This chapter deals with a new mathematical model to characterize the
interaction between machine and workpiece in a milling process. The model consists
of a multi-body system representing the milling machine and a linear thermo-elastic
workpiece model. An extensive experimental analysis supported the development
of the governing model equations. A numerical solution strategy is outlined and
complemented by simulations of stable and unstable milling processes including
workpiece effects. The last part covers the development of a new algorithm for the
stability analysis of large milling systems.
9.1 Introduction
The interaction of process and structure is the main reason for the unwanted chat-
ter phenomena in milling. Hence, the determination of stable cutting conditions for
given structures and the design of more efficient milling machines are important re-
search fields in production technology. Since the mathematical description of model
components like machine, workpiece and process is well understood by now, the
main challenge is to integrate them into a coupled model and to simulate the result-
ing system with tailored numerical algorithms to reproduce the stability limits and
the dynamical characteristics of the real system correctly.
Accordingly, the goal of the present project was the development of a coupled
model to study the dynamics of milling processes by means of time domain sim-
ulations. To cope with a necessarily long simulation time the machine structure is
treated as a multi-body system and the workpiece as a thermo-elastic body. Compre-
hensive experimental studies performed in close cooperation between engineers and
mathematicians provided the basis for the mathematical modeling, the identification
of model parameters and the validation of the numerical results.
The main achievements of the present work are the consistent mathematical mod-
eling of a complex milling system including the mathematical analysis of the derived
equations [10], the numerical implementation of a time-domain simulation system
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 203–224.
springerlink.com c Springer-Verlag Berlin Heidelberg 2013
204 D. Hömberg et al.
and its experimental validation, experimental studies, which have revealed a tem-
perature dependency of the stability limits.
The chapter is organized as follows: In Section 9.2, we will present the differ-
ent experimental studies carried out during the modeling procedure, as well as the
parameter identification and for the validation of the simulation results. Section 9.3
covers the derivation of the model equations, the development of numerical algo-
rithms and the presentation of simulation results. The last section is devoted to some
concluding remarks including the discussion of perspectives and further issues.
machine table
Fx Fy Fz
Tool:
N End Mill Cutter
z = 1; d = 8.0 mm
mm
Work piece:
Aluminium Alloy
30 EN AW-7075
Process Parameters:
15 ae = 8.0 mm
fz = 0.2 mm
n = 17200 rpm
0
0 750 1500 Hz 3000
Frequency f
Fig. 9.2 FFT of stable and unstable cutting processes
According to Sims [27], stable and unstable process states were determined by
the investigation of the ratio between the absolute value of the largest force am-
plitude, which can be allocated to the excitation frequency, and the largest force
amplitude of the occurring chatter frequency. If the ratio exceeds a value of 10 %,
the corresponding process is classified as unstable. In addition, the resulting surface
roughness, the occurrence of chatter marks and noise during the machining pro-
cess were also taken into account. A detailed description of the procedure is given
in [29].
9 Development of a Stability Prediction Tool 207
mm z = 1; d = 8.0 mm
Work piece:
Aluminium Alloy
2 EN AW-7075; T = 25oC
Process Parameters:
1 ae = 8.0 mm; fz = 0.2 mm
0
15600 16700 17800 rpm 20000
Spindle speed n
The variation of the stability limit is rather small in the analyzed range of spindle
speed. The maximum value is a p = 2 mm at spindle speeds around n = 17, 000 rpm
and n = 19, 400 rpm. For other spindle speeds, a p = 1.5 mm is the measured stability
limit.
The graph depicted in Figure 9.3 has been computed using the standard stabil-
ity prediction method proposed by Altintas et al. (see e.g. [2] or [1]). A mobility-
frequency-response function measured by means of hammer excitation and cutting
force coefficients determined from cutting force measurements under stable cutting
conditions have been provided as input data.
two orders of magnitude smaller than the corresponding value at the tool center
point [22]. Consequently, the effect of the dynamometer has been assumed to be
negligible.
However, recent results of machining tests have shown that the dynamic behavior
of the dynamometer-workpiece combination seems to have a much larger impact on
process behavior as expected. Although the amplitudes of the response functions
are very small, current experimental process analysis has shown that the shifting
of eigenfrequencies by use of a dynamometer seems to play a major role for the
stability behavior. Alternatively, the work piece accelerations during the machining
tests could be monitored to detect unstable processes. By defining a new stability
criterion this method would allow to remove the dynamometer and thus refine the
chatter detection.
Tool:
End Mill Cutter
N
z = 1; d = 8.0 mm
mm Work piece:
Aluminium Alloy
10 EN AW-7075
Process Parameters:
5 ae = 8.0 mm; ap = 1.0 mm
fz = 0.2 mm
n = 16000 rpm
0
0 750 1500 Hz 3000
Frequency f
Fig. 9.4 Comparison of stability behavior considering different geometries.
during the machining tests thermocouples type K were fixed on the sides of the
workpiece to control the local work piece temperature. In addition, the temperature
was monitored using a thermal imaging camera type Infratronics IR600. Due to this
approach it was possible to execute cutting tests with defined and homogeneous
heating conditions (Δ T = 1 K) of the workpiece.
Fig. 9.5 The dependency of the specific cutting force Kc on the preheating of the workpiece
z = 1; d = 8.0 mm
3 Work piece:
Aluminium Alloy
2 EN AW-7075
T = 25…100 oC
1 Process Parameters:
ae = 8.0 mm; fz = 0.2 mm
0
15600 16700 17800 rpm 20000
Spindle speed n
Fig. 9.6 Stability limits in dependency of the temperature
previously computed solution and pursue the simulation. Consequently, the model-
ing and the simulation process consists of two parts. While the first part deals with
the phenomena occurring during one tooth period, the second part focuses on the
construction of a series of workpiece reference domains and thus on an implemen-
tation of the method of steps.
with gE representing the external forces and torques applied via joints on each body.
The parameter vector η contains the free coupling parameters, such as joint stiffness
or joint damping, while M denotes the state dependent mass matrix and fI represents
the inertia forces. The last term incorporates the cutting force vector F acting on the
cutter. The generalized coordinates q describe the relative motion between the rigid
bodies. Due to the presence of pre-defined parameters, such as feed and spindle
rotation speed the time appears explicitly in the expressions for M, fI , fE and gE .
Since the system is organized in a tree-like structure an iterative algorithm (c.f. [7])
can be applied to evaluate the equations of motion during a time-domain simulation.
headstock
x-slider
y-slider
upper
spindle
lower
spindle
toolholder
z y
x tool
on the solution of the coupled system. As shown in [15], the equations of thermo-
elasticity read
ρ utt = div(σ ) + sm
E, (9.2)
2
σ = λ tr(ε )I + 2με − 3K αΘ I with K = (λ + μ ), (9.3)
3
1
ε= ∇u + (∇u)T , (9.4)
2
ρ cvΘt = κΔΘ − 3K α T0 div(ut ) + seE (9.5)
where λ and μ are the Lamé constants, ρ denotes the mass density, T0 the reference
temperature, cv the specific heat, κ the heat conductivity and α is the thermal expan-
sion coefficient. The function u represents the deformation field, I identity matrix
and tr(.) denotes the trace operator.
The heat equation has been formulated in terms of the deviation Θ from the
reference temperature T0 . Thus, the actual temperature T is given by T = Θ + T0 .
The workpiece is fixed on a large and rigid machine table, a configuration, which can
be approximated by imposing Dirichlet boundary conditions on the corresponding
part of the workpiece boundary, i.e.
F̂ = a p K̂(Tce ) h, (9.8)
where a p denotes the depth of cut. The vector K̂ denotes an empirical parameter, the
so called specific cutting force, which is usually assumed to be constant or depend-
ing on the cutting speed. However, as shown in Section 9.2.5, the cutting forces
decrease due to higher workpiece temperatures. This effect has been incorporated
into the cutting force model by multiplying the constant specific cutting force vector
std
K̂ = [K f , Kc , K p ]T by an empirical function involving the work piece temperature
Tce at the cutting edge, i. e., we define
(cTce )−b ,
std
K̂(Tce ) = K̂ (9.9)
with further fit parameters b and c. The uncut chip thickness h generally depends
on the current position of the cutting edge, the workpiece deformation and the
shape of the workpiece surface created by the preceding tooth. In the mathemati-
cal models delay terms are often used to incorporate the effect of the surface shape
(see [10]). However, to allow for a numerically stable and realistic coupling to the
workpiece, an alternative approach is pursued here. The workpiece surface shall
be constructed employing a real material removal model similar to the approaches
presented in [9], [32].
To derive a formula for the uncut chip thickness, we assume at first that the work-
piece reference configuration ΩR is given and show how the uncut chip thickness
can be derived from the current cutter position given by the solution of (9.1) and
the displacement field corresponding to (9.2). Note that for presentation purposes
we consider only the special case that the cutter axis is parallel to the z-axis of the
workpiece reference frame. In the general case, additional transformations depend-
ing on the solution of (9.1) have to be applied to compute the vector components in
the reference frame of interest.
As illustrated in Figure 9.8, a point on the cutting edge can be characterized
by the vector rce = r̃ce (t, q, z) = rca + rae . Note that the dashed lines in Figure 9.8
represent the ideal tooth path without machine and workpiece deformations. The
current workpiece domain Ω (t) is given by the reference configuration ΩR and the
displacement field corresponding to (9.2), i.e. x = X + u(t, X) with x ∈ Ω (t) and
X ∈ ΩR . As shown in [10], [23] and indicated by the bar labeled ’h’ in Figure 9.8,
the uncut chip thickness is the distance between a point ’ce’ on the cutting edge and
the workpiece surface measured in the direction of rae . Mathematically, this can be
formulated as follows
216 D. Hömberg et al.
0 / Ω (t),
if rce ∈
h(t, q, z, Ω (t)) = (9.10)
max
∗
h∗ otherwise,
h ∈H
Combining the expression for the uncut chip thickness with the empirical cutting
force model (9.8) divided by a p gives a force per unit length, which possibly assume
different values on each z-level, i.e.
Since the components of the vector R̂ (see Figure 9.8) are given with respect to
the reference frame corresponding to the cutting edge, (9.12) has to be transformed
by means of an orthogonal transformation O(ϕ (z)) in the cutter reference frame to
compute the cutting force per unit length acting on the cutter, i. e.
with ϕ (z) denoting the angle between rce and the x-axis of the cutter reference
frame. Integrating (9.13) along the cutter axis finally provides the resultant force
acting on the cutter.
y y
x z x
z Ω(t) Ω(t)
^
rca ^
Rf Rc
A(t,z)
rce
h yc h
cutting rae Δφ yc
φ(z) Δφ
edge rae
cutter xc cutter xc
axis axis
fz fz
Fig. 9.8 Uncut chip thickness (left) and definition of the volume force (right)
Since the presented model is a macroscopic approach, which cannot reproduce the
chip formation, the relative cutting forces have to be distributed over an area A(t, z)
located between cutting edge and workpiece surface. As illustrated in Figure 9.8,
for a given point on the cutting edge rae and a given level z the area A(t, z) is defined
as the set of all points in Ω (t) enclosed by the disc-ring segment with label ’A(t, z)’.
The segment is defined by the arcs through rae with radius D/2 and length Δ ϕ D
and through (1 − h/rae )rae with radius D/2 − h and length 2Δ ϕ (D/2 − h). Thus,
the volume force reads
0 if (x, y) ∈
/ A(t, z),
sE (t, x, y, z) =
m
R (9.15)
A(t,z) otherwise.
The heat conducted into the workpiece can be computed from the specific cutting
forces and the given cutting conditions (see [26]). If the shear angle appearing in the
corresponding expressions is estimated by an analytical formula as, for example,
proposed in [20], the total heat flux into the workpiece denoted by H(h, K̂) can be
written in explicit form. The corresponding term appearing on the right hand side
of (9.5) reads
0 if (x, y) ∈
/ A(t, z),
sE (t, x, y, z) = H(h,K̂)
e
(9.16)
A(t,z) otherwise.
With solution X∗ce (t, rce ) of the above equation and the cutting edge points we define
a new point y, which corresponds either to the cutting edge or to the new shape of
the reference domain, i. e.
rce / Ω (t),
if rce ∈
y(t) = (9.18)
X∗ce (t, rce ) otherwise.
218 D. Hömberg et al.
ΩRnew = ΩR \ Ωc . (9.19)
0.25
uncut chip thickness / [mm]
0.15 0.25
uncut chip thickness / [mm]
Fig. 9.9 Example for the material removal simulation, i. e. the evolution of the workpiece
shape, and the corresponding uncut chip thickness in case of a stable cut
0.2 0.4
0.1
0.15 0.3
0.1 0.2
0.05
0.05 0.1
0 0 0
1.5 1.6 1.7 1.8 1.9 0.031 0.0315 0.032 0.0325 0.033 0.065 0.066 0.067
−3
time / [sec] x 10 time / [sec] time / [sec]
Fig. 9.10 Example for the material removal simulation, i. e. the evolution of the workpiece
shape, and the corresponding uncut chip thickness in case of an unstable cut
9.3.6 Simulations
The coupled model and the developed simulation algorithm provide a tool to inves-
tigate the characteristics of milling processes involving the dynamics of machine
220 D. Hömberg et al.
and workpiece. Although the temperature effects presented in Section 9.2.5 can
be reproduced with the presented model, an illustration of the related phenomena
is beyond the scope of this work. Therefore, we focued on effects, which can be
simulated with an elastic workpiece model. Due to this restriction the workpiece
equations can be reduced to system (9.2)-(9.4) with a constant temperature, i.e.
Θ = 0. The damping effects in the workpiece were incorporated into the model
on the space discrete level, introducing a so-called Rayleigh damping term. Again,
the unknown damping parameters were determined by comparing simulated and
measured frequency-response functions. Recall that the vector of unknown param-
eters in (9.1) has been fitted so that the multi-body system reproduces the measured
frequency-response functions at the tool centre point (TCP). Moreover, the cutting
force model employed in [22] to compute the stability limit illustrated in Figure
9.3 is similar to (9.8). Consequently, a system composed of the machine model and
a rigid workpiece almost has the same stability limits as the system analyzed to
compute the red line in Figure 9.3.
As observed in the experiments, a supple workpiece structure or the wrong choice
of cutting depth and spindle speed may destabilize the milling process. In order to
display these effects numerically we considered three examples.
In the first simulation, the workpiece structure is rather stiff and the process
parameters have been chosen so that no chatter occurs (a p = 0.5 mm and n =
16, 400 rpm, see Figure 9.3).
50 0.3
rel. force spectrum / [N / mm]
40 0.25
0.2
30
0.15
20
0.1
10 0.05
0 0
0 1000 2000 3000 4000 5000 0 0.05 0.1 0.15
frequency / [Hz] time / [sec]
Fig. 9.11 Relative cutting force spectrum and evolution of the uncut chip thickness corre-
sponding to a stable cut
Figure 9.11 illustrates the spectrum of the cutting force in x-direction divided
by a p and the evolution of the uncut chip thickness. Both diagrams indicate that the
process is stable. No chatter peaks arise in the force spectrum and the uncut chip
thickness converges to the stationary evolution.
For the second example, the spindle speed has been set to n = 17, 800 rpm. In this
case, the process parameters correspond to unstable cutting conditions. Additional
chatter peaks occur in the relative force spectrum (spectrum of the cutting force in
x-direction divided by a p ) depicted in Figure 9.12. According to the considerations
9 Development of a Stability Prediction Tool 221
0.4
rel. force spectrum / [N / mm] 50 Simulation
30
0.2
20
0.1
10
0 0
0 1000 2000 3000 4000 0 0.05 0.1 0.15
frequency / [Hz] time / [sec]
Fig. 9.12 Simulated and measured relative cutting force spectrum and evolution of the uncut
chip thickness corresponding to an unstable cut
60 0.4
rel. force spectrum / [N / mm]
50
0.3
40
30 0.2
20
0.1
10
0 0
0 1000 2000 3000 4000 0 0.05 0.1 0.15
frequency / [Hz] time / [sec]
Fig. 9.13 Relative cutting force spectrum and evolution of the uncut chip thickness corre-
sponding to an unstable cut due to a supple workpiece
outlined in Section 9.2.4.2 and 9.2.3.2 the additional chatter peaks clearly indicate
that the corresponding process is unstable.
The evolution of the uncut chip thickness illustrated in Figure 9.12 confirms this
result. In contrast to the evolution shown in Figure 9.11, the uncut chip thickness
does not converge to the stationary evolution but increases noticeably after a short
decay at the beginning and remains on a high level until the end of the simulation.
In the third example, the process parameters have been set to the same values as
in the first simulation (a p = 0.5 mm and n = 16400 rpm). The stiff workpiece uti-
lized in the first example has been replaced by a more supple structure. Especially
on top, the supple workpiece exhibits a high compliance, which can destabilize the
previously stable milling process.
Analyzing the evolution of the uncut chip thickness reveals that the process does
not converge to a stationary regime. Large workpiece oscillations lead to an increas-
ing uncut chip thickness and thus to increasing cutting forces. The additional chatter
peaks appearing in the relative force spectrum clearly indicate that the correspond-
ing process is unstable. In contrast to the second example, the highest chatter peak
222 D. Hömberg et al.
is not located at 2, 834 Hz at but at 408 Hz. Thus, the weak spot in the structure
leading to chatter vibrations is different in the two examples.
9.4 Conclusions
The goal of this chapter was the development of a complex milling model to in-
vestigate effects of machine and workpiece structure on the stability of milling pro-
cesses. The experimental analysis in Section 9.2 reveals that the stability limits can
be increased by pre-heating the workpiece. Moreover, it illustrates how milling pro-
cesses can become unstable due to a lack of workpiece stiffness. The experiments
conducted for the identification of machine parameters display that the frequency-
response functions at the TCP strongly depend on the excitation method and vary
with the angular spindle position. The simulations in Section 9.3 clearly demonstrate
that the developed model is capable of reproducing the instability effects observed
in the experiments. For the first time a new stability analysis method allows the
determination of stability limits of large DDE-systems with periodic coefficients.
The results are promising and open up various directions for future research.
A challenging task would be to investigate the stability of milling processes with
respect to variations in the machine design.
Finally, from the application point of view, an efficient numerical tool for the
systematic derivation of stability diagrams is most desirable. The developed stability
analysis tool is a first step in this direction. The improvement of the numerics and the
exploitation of model reduction techniques to increase the efficiency of the method
will be subject to further research.
References
1. Altintas, Y.: Manufacturing Automation–Metal Cutting Mechanics, Machine Tool Vibra-
tions, and CNC Design. Cambridge University Press (2000)
2. Altintas, Y., Budak, E.: Analytical prediction of stability lobes in milling. Ann. of the
CIRP 44(1), 357–362 (1995)
3. Altintas, Y., Weck, M.: Chatter stability of metal cutting and grinding. Ann. of the
CIRP 53(2), 619–642 (2004)
4. Amin, A.K.M.N., Abdelgadir, M.: The Effect of Preheating of Work Material on Chatter
During End Milling of Medium Carbon Steel Performed on a Vertical Machining Center
(VMC). J. of Manuf. Sci. and Eng. 125, 674–680 (2003)
5. Budak, E.: Mechanics and dynamics of milling thin walled structures. Ph.D. Thesis,
University of British Columbia, Vancouver, Canada (1994)
6. Bueler, E.: Error Bounds for Approximate Eigenvalues of Periodic-Coefficient Linear
Delay-Differential Equations. SIAM J. Numer. Anal. 45(6), 2510–2536 (2007)
7. Bremer, H., Pfeiffer, F.: Elastische Mehrkörpersysteme. B. G. Teubner, Stuttgart (1992)
8. Campa, F.J., López de Lacalle, L.N., Lamikiz, A., Sánchez, J.A.: Selection of cutting
conditions for a stable milling of flexible parts with bull-nose end mills. J. of Mater.
Proc. Technol. 191, 279–282 (2007)
9 Development of a Stability Prediction Tool 223
9. Campomanes, M.L., Altintas, J.: An improved Time Domain Simulation for Dy-
namic Milling at Small Radial Immersions. Transactions of the ASME (2003),
doi:10.1115/1.1580852
10. Chelminski, K., Hömberg, D., Rott, O.: On a thermomechanical milling model. Nonlin-
ear Analysis Real World Applications 12, 615–632 (2010)
11. Ewins, D.J.: Modal Testing: Theory and Practice. Res. Stud. Press LTD. (1986)
12. Faassen, R.P.H., van de Wouw, N., Oosterling, J.A.J., Nijmeijer, H.: Prediction of regen-
erative chatter by modelling and analysis of high-speed milling. Int. J. of Mach. Tools &
Manuf. 43, 1437–1446 (2003)
13. Faassen, R.P.H.: Chatter Prediction and Control for High-Speed Milling - Modelling and
Experiments. Dissertation, TU Eindhoven, Netherlands (2007)
14. Gatto, A., Iuliano, L.: Chip Formation Analysis in High Speed Machining of a Nickel
Base Superalloy with Silicon Carbide Whisker-Reinforced Alumina. Int. J. of Mach.
Tools & Manuf. 34, 1147–1161 (1994)
15. Haupt, P.: Continuum Mechanics and Theory of Materials. Springer, Berlin (2000)
16. Hughes, T.J.R.: The finite element method. Dover Publications, New York (2000)
17. Insperger, T., Stépán, G.: Stability of the milling process. Period. Polytech. – Mech.
Eng. 44(1), 47–57 (2000)
18. Insperger, T., Stépán, G.: Updated semi-discretization method for periodic delay-
differential equations with discrete delay. Int. J. Numer. Methods Eng. 61(1), 117–141
(2004)
19. Insperger, T., Mann, B.P., Surmann, T., Stépán, G.: On the chatter frequencies of milling
processes with runout. Int. J. of Mach. Tools & Manuf. 48, 1081–1089 (2008)
20. Lee, E.H., Schafer, B.W.: The Theory of plasticity applied to a problem of machining.
Journal of Applied Mechanics 18, 405–413 (1951)
21. Mann, B., Insperger, T., Bayly, P.V., Stépán, G., Schmitz, T.L., Peters, D.A.: Effects of
Radial Immersion and Cutting Direction on Chatter Stability in End-Milling. In: N.N.
(ed.) Proceedings of the International Mechanical Engineering Conference and Exposi-
tion (IMECE), New Orleans, LA (2002)
22. Rasper, P., Rott, O., Hömberg, D., Uhlmann, E.: Analysis of uncertainties in the stability
prediction for milling processes. In: Altintas, Y. (ed.) Proceedings, CIRP 2nd Interna-
tional Conference Process Machine Interactions (2010)
23. Rott, O., Rasper, P., Hömberg, D., Uhlmann, E.: A milling model with thermal effects in-
cluding the dynamics of machine and work piece. In: Denkena, B. (ed.) Proceedings, 1st
International Conference on Process Machine Interactions, PZH Produktionstechnisches
Zentrum GmbH, Hannover-Garbsen (2008)
24. Rott, O., Jarlebring, E.: An iterative method for the multipliers of periodic delay-
differential equations and the analysis of a PDE milling model. In: Vyhlidal, T. (ed.)
Proceedings of the 9th IFAC Workshop on Time Delay Systems, Prague (2010)
25. Schulz, H.: Hochgeschwindigkeitsfräsen metallischer und nichtmetallischer Werkstoffe.
Carl Hanser Verlag, München (1989)
26. Shaw, M.C.: Metal Cutting Principles. Oxford University Press, Oxford (2005)
27. Sims, N.D.: The self-excitation damping ratio: A chatter criterion for time-domain
milling simulations. J. of Manuf. Sci. and Eng., Trans. of the ASME 127(3), 433–445
(2005)
28. Uhlmann, E., Mense, C.: Analysis of the Milling Machine Operation Behaviour for Pro-
cess Stability and Machine Tool Optimisation. In: N.N. (ed.) Proceedings of 10th CIRP
Int. Workshop on Model. of Mach. Oper., Reggio Calabria, Italy (2007)
29. Uhlmann, E., Rasper, P.: Temperaturabhängiges Stabilitätsverhalten–Untersuchung des
Temperatureinflusses auf das Stabilitätsverhalten beim Umfangsstirnfräsen. In: wt Werk-
stattstechnik online, Düsseldorf, Germany, vol. 7(8), pp. 464–469 (2009)
224 D. Hömberg et al.
30. Venuvinod, P.K., Lau, W.S., Narasimha Reddy, P., Rubenstein, C.: On the Formation of a
Fluid Film at the Chip Tool Interface in Rotary Machining. Ann. of the CIRP 32, 59–64
(1982)
31. Weck, M., Teipel, K.: Dynamisches Verhalten spanender Werkzeugmaschinen. Springer,
Berlin (1977)
32. Weinert, K., Surmann, T.: Geometric Simulation of the Milling Process for Free Formed
Surfaces. In: Weinert, K. (ed.) Simulation Aided Offline Process Design and Optimiza-
tion in Manufacturing Sculptured Surfaces, Witten (2003)
33. Zatarain, M., Munoa, J., Peigné, G., Insperger, T.: Analysis of the Influence of Mill Helix
Angle on Chatter Stability. Ann. of the CIRP 55(1) (2006)
34. Zeng, W., Peng, X., Leu, M.C., Zhang, W.: A Novel Contour Generation Algorithm
for Surface Reconstruction From Dexel Data. Journal of Computing and Information
Science in Engineering 7, 203–210 (2007)
Chapter 10
Synthesis of Stability Lobe Diagrams
Abstract. Chatter vibrations during machining lead to poor workpiece surfaces and
increased tool wear. In the worst case, the tools and even the main spindle can be
damaged. Nowadays, the surface regeneration is considered to be the main effect
causing chatter instabilities. Regenerative chatter is initiated by repetitive tooth en-
gagement where the currently engaged tooth cuts the surface produced by the pre-
ceding tooth. In a stability lobe diagram (SLD), the stable and unstable areas are
separated by the graph of a critical cutting parameter plotted against the spindle
speed. Stability lobe diagrams can be used to optimize machining processes in
terms of maximizing material removal rate under stable cutting conditions. These
SLDs are computed by time domain simulations. However, this consumes a lot of
computational time. Thus, several time efficient algorithms in discrete time as well
as frequency domain have been developed in the last decades. This chapter scruti-
nizes under what conditions different algorithms in frequency domain can be
applied. The processes are separated regarding cutting conditions and dynamic be-
havior so that the most time efficient algorithm can be chosen for each class.
10.1 Introduction
To predict the stability boundaries of cutting processes the interactions within the
closed-loop of the coupled sub-systems machine and process is examined,
Fig. 10.1. Regenerative chatter is caused by the repetitive engagement of a tooth
into the surface cut by the preceding tooth. In Fig. 10.1, this is represented by the
time delay T.
Machine
x
y
Fx,y_stat Gxx ( jω) Gxy (j ω)
G ( jω) G (j ω)
yx yy
Fx
F
y aFx ΔFP + -
a
Fy
ap kt, kr T [a ay ]
Δh(t,t-T)
x
Δh(t)
Process
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 225–244.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
226 K. Großmann and M. Löser
The figure shows an example where the process machine interactions in x-y-
plane are investigated and the z direction is neglected. The dynamic behavior of
the machine is represented by the transfer function matrix G(jω), which contains
the direct transfer functions and the cross transfer functions in x and y direction
respectively. However, the process force is a function of chip thickness h. Thus,
the deflections in x-y-plane have to be transformed into the direction of chip
thickness h. This is conducted by means of so-called directional coefficients (ax,
ay). With the dynamic chip thickness Δh the dynamic process forces ΔFp can be
computed by applying an empirical cutting force model. Here, this is represented
by the tangential coefficient kt and the radial coefficient kr. The process forces
now have to be transformed into x-y-plane again to close the loop of the process
machine interaction.
One method to compute stability boundaries of this closed loop is to model and
simulate the interactions in time domain. Modeling non-linearities in the time do-
main is much easier than in the frequency domain; at the same time, however, the
computation time to calculate SLDs increases significantly. One reason for this is
that a certain amount of “real time” has to be simulated to ensure the detection of
unstable cutting using time signals.
It is more time efficient to analyse the process machine interactions in fre-
quency domain. Analytical methods in frequency domain are the oldest algorithms
to predict the stability boundaries and are based on the research works by Tlusty
and Polacek as well as by Tobias and Fishwick [2, 3]. Algorithms in frequency
domain apply stability criteria like the Nyquist criterion on the open loop transfer
function of the process-machine interactions, [1, 4]. In milling operations, the di-
rectional coefficients shown in Fig. 10.1 are time-variant and vary periodically
with the tooth-passing frequency. To apply the methods in frequency domain av-
erage directional coefficients are used. Since the average directional coefficients
equal zeroth order Fourier coefficients this method is called zeroth order approxi-
mation (ZOA-method).
Operations with highly intermittent cutting conditions, such as low immersion
milling, show very strong time variance, which is assumed to affect the process
stability. In these cases, the assumption of constant average directional coeffi-
cients may lead to incorrect predictions of stability boundaries. Various authors
have presented computational time-efficient methods in discrete time domain to
determine stability lobe diagrams for highly intermittent cutting operations. Bayly
et al. have presented a time-finite element analysis [5], Insperger and Stepan have
developed the semi-discretization method [6].
To take the time variant behavior of directional coefficients in frequency do-
main into account Budak and Altintas expanded their single frequency solution to
the multi-frequency solution [7]. This was later on applied to low immersion mil-
ling by Merdol and Altintas [8].
Most of the studies that apply these advanced time efficient methods focus on
the investigation of single influences (for example: single degree of freedom be-
havior, helix angle, low immersion milling). In this chapter, the complex influence
of different effects is discussed for methods working in frequency domain. ZOA
method and two slightly different methods based on the multi-frequency solution
10 Synthesis of Stability Lobe Diagrams 227
1
with T delay T =
n⋅Z
Z number of teeth
n spindle speed
ωT tooth passing frequency ω T = 2π ⋅ n ⋅ Z
Here, Gg(iw) is the directed frequency response function. The Nyquist criterion
can be applied to this transfer function:
< 1 stable
Re(G 0 ( jω )) = = 1 stability boundary
(10.2)
> 1 unstable
Im(G 0 ( jω )) = 0
Simply spoken, the Nyquist criterion checks if an input signal leaves the open loop
without phase shift (imaginary part equals zero) and amplified (real part > 1).
Since the output signal is equivalent to the input signal of the subsequent pass
through of the loop the signal will be more and more amplified and the closed
loop becomes unstable.
In peripheral milling this criterion cannot be applied under any condition since
the coordinates in the x-y-plane are coupled, Fig. 10.2. The simplest approach for
the solution of this problem is by handling the interaction not in the machine coor-
dinates x and y but by investigating the transfer function at the direction of the
chip thickness h. Therefore, the deflections are transformed into the direction of
chip thickness and the chip thickness-dependent process forces are transformed
back into the direction of the machine coordinates. The transformation is carried
out by so-called directional coefficients [1].
Regarding regenerative chatter the static process forces and therefore the static
chip thickness can be neglected, [4]. The static chip thickness is the chip thickness
that would occur without relative displacements between workpiece and tool. The
228 K. Großmann and M. Löser
fz
transformation
Fr
tool
ϕz Fx Δh= sinϕ·Δx + cosϕ·Δy
Ft
Ft= Kt·ap·Δh
Δx Fr= Kr·ap·Δh
Fr Δy
Δh Fx= -cosϕ·Ft - sinϕ·Fr
hstat Fy= sinϕ·Ft - cosϕ·Fr
Fy
Ft workpiece
ϕ
ϕ ϕe
dynamic chip thickness Δh is the change of h caused by the present deflection and
the deflection at the time of the preceding tooth engagement.
Δx x (t ) x ( t − T)
Δh = [sin ϕ cos ϕ ] ⋅ = [sin ϕ cos ϕ ] ⋅ −
y(t ) y(t − T )
(10.3)
Δy
Taking into account that the deflections are the reaction of forces acting on the
flexible machine structure this can be written in frequency domain.
G G xy Fx
Δh = (1 − e − jωT ) ⋅ [sin ϕ cos ϕ ] ⋅ xx ⋅
G yx G yy Fy
(10.4)
[ G
Δh = (1 − e − jωT ) ⋅ a x a y ⋅ xx ] G xy Fx
⋅
yx G yy Fy
G
The dynamic process forces are a function of dynamic chip thickness Δh:
Fx − k ⋅ cos ϕ − k r ⋅ sin ϕ a
F = a p ⋅ t ⋅ Δh = a p ⋅ Fx ⋅ Δh (10.5)
y k t ⋅ sin ϕ − k r ⋅ cos ϕ a Fy
Equalizing the vector of process forces in (10.4) with (10.5) and taking the time-
delayed re-engagement of the subsequent tooth into account leads to the transfer
function of the open loop where Gg denotes the directed frequency response
function:
G 0 = a p ⋅ (1 − e − jωτ ) ⋅ a x [ ] G
a y ⋅ xx
G xy a Fx
⋅
G yx G yy a Fy
(
G 0 = a p ⋅ D ⋅ a x a Fx G xx )
+ a y a Fx G yx + a x a Fy G xy + a y a Fy G yy
(10.6)
G 0 = a p ⋅ D ⋅ (a xx G xx + a xy G yx + a yx G xy + a yy G yy ) = a p ⋅ D ⋅ G g
The directional coefficients depend on the immersion angle ϕ and are therefore
time-dependent. So, to apply (10.6) in frequency domain average directional
coefficients have to be used. Applying the Nyquist criterion and rearranging (10.6)
gives the frequency-dependent critical depth of cut apcrit, [1]
10 Synthesis of Stability Lobe Diagrams 229
−1
a pcrit (ω ) = (10.7)
2 ⋅ Re(G g ( jω ))
Since only positive cutting depths are relevant for cutting operations only frequen-
cies with negative real parts of the directed frequency are considered. Further-
more, the relation between chatter frequency and spindle speed can be derived
from the Nyquist criterion:
60 ⋅ f
n=
1
z m − arctan
{
Re G g ( jω )
} (10.8)
π Im G g ({jω )
}
With (10.7) and (10.8) the relation between cutting depth and spindle speed and,
consequently, the stability lobe diagram (SLD) can be computed. For this purpose,
the dynamic behaviour of the machine has to be assumed as constant within the
observed range of spindle speed.
(10.7) shows that only frequencies are relevant, whose real parts of the directed
frequency response function are negative. For a multi-degree of freedom system
(MDoF system) it is possible to divide the directed transfer function Gg into sec-
tions, which can be assigned to a specified mode respectively. For each of these
sections a mode-dependent stability boundary can be computed. The total stability
lobe diagram can now be determined by choosing the minimum of the mode-
dependent boundaries at every spindle speed. However, it has to mentioned that
this is only valid, if the assumption of averaged directional coefficients can be
made and if the directional coefficients are used to compute the directed transfer
function Gg as shown in (10.6). The directed transfer function Gg takes the cou-
pling of x and y direction into account. Insperger and Stepan have shown that
computing stability lobe diagrams for x and y direction separately by neglecting
the geometrical coupling and superposing these SLDs will lead to an incorrect
prediction of stability boundaries, [9].
∞ T
1
a xx ( t ) =
r = −∞
a xx ,r ⋅ e jrωT t , a xx ,r =
T
a xx ( t ) ⋅ e − jrωT t dt
0
(10.9)
Taking into account that the relation between time and immersion angle ϕ is given
by the tooth-passing frequency and that the directional coefficients are zero, if the
teeth are not engaged, the Fourier coefficients can be written as [8]:
ϕ ex
Z
a
− jrZϕ
a xx ,r = xx (ϕ ) ⋅ e dϕ (10.10)
2π
ϕ st
Where ϕst denotes the angle where the teeth engage the workpiece and ϕex denotes
the angle where the teeth exit the workpiece. Since the Fourier coefficients of ze-
roth order are equivalent to the average directional coefficients the methods using
average directional coefficients are called zeroth order approximation (ZOA).
Fig. 10.3 depicts an example of absolute values of Fourier coefficients.
0.3
up milling
0.2 dtool= 12 mm
0.1 λ= 0°
ae= 2 mm
0
number of teeth Z= 4
r= 0 r= 1 r= 2 r= 3 r= 4
By applying the Fourier series the process forces as a function of the change of
the relative deflection between tool and work piece can be written as follows:
hr
F( t ) = a p ⋅
A r e jrωT t ⋅ Δx( t )
(10.11)
r =− h r
This relation in time domain can be transformed into frequency domain by using
the Laplace transformation. The Laplace transforms are given by:
(
L{Δx (t )} = L{x(t ) − x(t − T )} = 1 − e jωT ⋅ x( jω ) = D ⋅ x( jω ) )
Le{ jrωT t
}
⋅ x(t ) = x( j(ω − rω T ))
(10.12)
hr
F( jω) = a p ⋅ D ⋅
A r ⋅ x( j(ω − rωT ))
r =−h r
r h
F( jω) = a p ⋅ D ⋅
A r ⋅ G( j(ω − rωT )) ⋅ F( j(ω − rωT ))
r =−h r
F( j(ω − pωT )) = a p ⋅ D ⋅ (10.13)
hr
⋅
A r ⋅ G( j(ω − (p + r )ωT )) ⋅ F( j(ω − (p + r )ωT ))
r =−h r
hr
Fp = a p ⋅ D ⋅
A r ⋅ G p+r ⋅ Fp+r
r =− h r
For a Fourier order hr=1 follows:
F−1
F0 = a p ⋅ D ⋅
F1
(10.14)
A −1 ⋅ G −2 ⋅ F− 2 A 0 ⋅ G −1 ⋅ F−1 A1 ⋅ G 0 ⋅ F0
⋅ A −1 ⋅ G −1 ⋅ F−1 + A 0 ⋅ G 0 ⋅ F0 + A1 ⋅ G 1 ⋅ F1
A ⋅ G ⋅ F A ⋅ G ⋅ F A ⋅ G ⋅ F
−1 0 0 0 1 1 1 2 2
The right side of (10.14) shows forces F with indices of r = ±(h r + 1) . Since the
absolute values of Ar converge to zero for increasing order r the terms with indices
of order greater than hr are set to be zero. It follows:
F−1 A 0 A1 0 G −1 0 0 F−1
A
0
F = a ⋅ D ⋅ A A 1⋅ 0 G0 0 ⋅ F0
p −1 0
F1 G 1 F1
0 A −1 A 0 0 0 (10.15)
F out = a p ⋅ D ⋅ A ⋅ G ⋅ F in
h h h h
G0 = ap ⋅ D ⋅ Ah ⋅Gh
This leads to the transfer function of the open loop. Therein, the coordinates are
coupled. To apply the Nyquist criterion the investigated system needs only one
coordinate or – in the case of a multi variable system – the coordinates have to be
decoupled. The coordinates can be decoupled by a modal transformation. The
modal coordinates correspond to the eigenvalues of G0. The Nyquist criterion can
now be applied to every one of the ζ·(2·r+1) eigenvalues. ζ is the number of the
structural coordinates (ζ= 2 for a stability analysis in the x-y-plane), [8].
Since the depth of cut is a scalar value the transfer function can be scaled
with ap.
232 K. Großmann and M. Löser
G0 =
1
ap
(
⋅ G 0 = eig D ⋅ A h ⋅ G h ) (10.16)
Fig. 10.4 shows an example of the real and imaginary parts of decoupled transfer
functions. In this example, the dynamic behaviour has relevant eigenfrequencies of
up to 5 kHz. Since the matrix of transfer functions contains the transfer functions at
frequencies shifted by multiples of the tooth passing frequency at a specified fre-
quency, the superposition of transfer functions is periodic with tooth passing fre-
quency. So the stability analysis need not be carried out for the whole bandwidth of
5 kHz but only for a frequency band of the tooth-passing frequency ft.
ft
100 100
Re(G0) [m-1]
50 0
Im(G0) [m-1]
0 -100
Im(G0) [m-1]
100
-50 0
-100 -100
-100 0 100 0 1000 2000 3000 4000 5000
Re(G0) [m-1] f [Hz]
Fig. 10.4 Decoupled and Scaled Transfer Functions of the Open Loop
These sets of transfer functions have to be computed for every discrete spindle
the stability boundary has to be determined at. The stability boundaries are given
by the reciprocals of the real parts of the intersections between transfer function
and real axis. In [10] Altintas et al. assert that “the most conservative and positive
depth of cut must be considered as a final solution.” So, the critical depth of cut is
given by:
1
a p _ crit =
(
Re G 0 (Im = 0) max ) (10.17)
Fig. 10.5 demonstrates the computation of the stability boundary for the example
mentioned before. The relevant parts of the transfer functions are drawn as bold
lines and the critical depth of cut is ap_crit= 11.2 mm.
However, the assumption that the critical depth of cut is given by the most con-
servative real part, is not correct in every case. However, if all intersections with
the real axis are analyzed, the algorithm can also be used, if multiple stability
boundaries occur at a specified spindle speed. All algorithms in frequency domain
10 Synthesis of Stability Lobe Diagrams 233
100 10
50 5
Re(G0) [m-1]
Re(G0) [m-1]
0 0 1/apcrit= 89.1 m-1
apcrit= 11.2 mm
-50 -5
-100 -10
-100 0 100 80 90 100
Im(G0) [m-1] Im(G0) [m-1]
apply the simplified Nyquist criterion shown in (10.2). The complete criterion
analyzes not only the intersection of the transfer locus with the real axis. It says
the process is stable, if the Nyquist point is encircled counter-clockwise at least
once. An intersection with the real axis where the imaginary part changes from
positive to negative values will not lead to an unstable process.
Fig. 10.6 shows a simple example to demonstrate this. It depicts a section of a
SLD for a one-dimensional process (only x-direction has been taken into account).
The dynamic system has an SDoF behavior (m= 0.572 kg; d= 140 Ns/m;
c= 2.2·107 N/m). The cutting force coefficients are kt= 3.6·109 N/m² and
kr= 2.25·109 N/m². The figure compares the stability boundaries determined by
time domain simulations with the boundaries determined by the multi-frequency
solution. The figure also shows the transfer locus for a spindle speed of nspin-
dle= 10,600 rpm. Point I marks a counter-clockwise intersection and the process
gets unstable at a depth of cut of apcrit= 7.3 mm. The clockwise intersection at
point II compensates the instability and the process becomes stable again up to the
depth of cut marked by point III.
10
Im(G0) [m-1]
Im(G0) [m-1]
III II I
0 0
-10
-20
-500
nspindle = 10600 rpm
20
Fig. 10.7 shows simulated process forces Fy for different helix angles at axial
cutting depths ap= 10 mm and ap= 20 mm but in both cases for the same radial
depth of cut ae= 3mm. The figure also depicts the spectra of the process forces.
For the straight fluted mill there is a sharp change of the forces when the cutting
edge leaves the workpiece. The force at zero frequency (which equals the average
force) is independent of the helix angle. The forces at the tooth-passing frequency
and their harmonics decrease with increasing helix angle. The decreasing effect is
more significant for the axial depth of cut of ap= 20 mm.
The magnitudes of the process forces at tooth-passing frequency and their har-
monics correspond to the Fourier coefficients. Thus, for helicoidal mills the values
of the Fourier coefficients depend on the axial depth of cut ap. In this case, the
value of the Fourier coefficients is given by [11]:
− j⋅r⋅2⋅π ⋅a / p
1− e p
A *r = A r ⋅ (10.18)
j⋅ r ⋅ 2 ⋅π ⋅ a p / p
10 Synthesis of Stability Lobe Diagrams 235
100
λ= 0
250 Fy(f) [N] ap= 10 mm
Fy(t) [N] 80 λ= 10°
200
60 λ= 20°
150
average force
λ= 30°
100 40
50 20
0 0
0 5 10 0 500 1000 1500 2000
t [ms] f [Hz]
200
500
Fy(t) [N]
Fy(f) [N] ap= 20 mm
400 150
300
average force
100
200
50
100
0
0
0 5 10 0 500 1000 1500 2000
t [ms] f [Hz]
Ar represents the Fourier coefficients for the straight-fluted mill and p is the
pitch of a tool with Z teeth.
d ⋅π
p = Tool (10.19)
Z ⋅ tan(λ )
With increasing axial depth of cut the values of the Fourier coefficients decrease.
The Fourier coefficients differ for the different helix angles. The zeroth order
coefficients are independent of the helix angle but the values of the higher order
coefficient decrease with an increasing helix angle λ.
Similar to (10.15), the open loop transfer function can be written as follows:
G 0 = a p ⋅ D ⋅ A *h ⋅ G h (10.20)
Decoupling the coordinates in (10.20) and applying the Nyquist criterion could
still be carried out by computing the eigenvalues, as described in the chapter be-
fore. This can be achieved by computing the eigenvalues for the characteristic eq-
uation
[
det I + ΛA h G h = 0 ] (10.21)
(
with Λ = a p 1 − e jωT , [8]. )
236 K. Großmann and M. Löser
However, if the decoupled system has to be computed for every ap, it is less
time consuming to apply a different approach. For a multi-variable system the sta-
bility analysis can be carried out for the determinant of the matrix of open loop
transfer functions minus identity matrix:
G 0 = det G 0 − I (10.22)
10.2.4 Summary
In the previous three sections, different algorithms for the computation of stability
lobe diagrams have been shown. The first and simplest one is the well-known ze-
roths order approximation (ZOA). The last two approaches are expanded versions
of the multi-frequency solution presented by Merdol and Altintas. One of them
uses the computation of eigenvalues to decouple the system of multiple “frequen-
cy coordinates” (Multifreq_eig) and the other one computes the determinants of
the transfer matrices (Multifreq_det). Which algorithm is best suitable for the de-
termination of a SLD depends on the effects that have to be taken into account.
However, the more complex the algorithm gets, the more computational time is
needed. But even the algorithm for a depth-of-cut-dependent behavior is less time-
consuming than time domain simulations. In the following section, these algo-
rithms are applied to cutting operations with different process and machine
behavior.
10.3 Application
well as the coefficients of the process force model. To describe the process forces
a linear force model was used [12]. The parameters were conducted by means of
cutting tests for aluminium AA7075 workpieces.
Process Up-milling
Number of teeth Z= 4
Tool
Tool diameter dTool= 12 mm
Material AlZn5,5MgCu (AA7075)
Workpiece Tangential force coefficient kt= 830 N/mm²
Radial force coefficient kr= 225 N/mm²
−m
1
G xx =
x ,m ω + jωd x ,m + c x ,m
2
m =1
ny
(10.23)
−m
1
G yy =
y, m ω + jωd y ,m + c y,m
2
m =1
G xy = G yx = 0
3 Gxx
compliance [µm/N]
Gyy
0
400 600 800 1000 1200
f [Hz]
Fig. 10.8 Reference for a SDoF Behavior
Gxx Gyy
Mass m [kg] 0.922 0.925
Damping d [kg/s] 69 69
7
Stiffness c [N/m] 1.93·10 1.95·107
For the example of SDoF, several authors showed systems that may occur at
small radial immersion additional stability lobes. These additional stability lobes
cannot be predicted by the ZOA-method. Zatarain et al. [11] investigated the in-
fluence of the helix angle on chatter stability by using the multi-frequency solution
by Merdol and Altintas as well as the semi-discretization method by Insperger and
Stepan. This work showed that for increasing helix angles the additional lobes
transform into closed instability islands. The same behavior can be shown for this
example here.
30
Multifreq_eig
25 time domain
20
apcrit [mm]
15
10
5
30
Multifreq_eig
2.5
25 λ= 0°
20 Multifreq_det
apcrit [mm]
λ= 30°
15 ZOA
10
5
With increasing helix angle, the additional stability lobes transform into unsta-
ble islands within the stable area. This can be shown with the multi-frequency so-
lution as well as by time domain simulations. Since the directional coefficients
depend on depth of cut the Multifreq_det algorithm has to be used in the case of a
helix angle λ= 30°. For a helix angle of λ= 0 the Multifreq_eig algorithm is used.
For a helix angle of λ= 30° the ZOA algorithm shows nearly the same results as
the multi-frequency solution, except the unstable islands, Fig. 10.10.
30
Multifreq_eig
25 λ= 0°
20 Multifreq_det
apcrit [mm]
λ= 30°
15 ZOA
10
5
For larger radial immersions the time variant behavior of the directional coeffi-
cients gains less impact on the stability boundaries. At a radial depth of cut of
ae= 3 mm, no impact of the helix angle can be shown and the ZOA algorithm
results in practically identical stability boundaries to the multi-frequency solution,
240 K. Großmann and M. Löser
Fig. 10.11. In this case, the fast ZOA method can be applied without loosing accu-
racy of the predicted stability boundaries.
0.6
compliance [µm/N]
0.4
0.2
0
0 1000 2000 3000 4000 5000
f [Hz]
Fig. 10.12 Dynamic Behavior of the Rotational Symmetric MDoF System
30
Multifreq_eig
25 λ= 0°
Multifreq_det
apcrit [mm]
20 λ= 30°
ZOA
15
10
5
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
20
Multifreq_eig
λ= 0°
15
Multifreq_det
apcrit [mm]
λ= 30°
10 ZOA
0
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
0.6 Gxx
compliance [µm/N]
Gyy
0.4
0.2
0
0 1000 2000 3000 4000 5000
f [Hz]
Fig. 10.15 Dynamic Behavior of the Non-Symmetric MDoF System
242 K. Großmann and M. Löser
Fig. 10.16 and Fig. 10.17 show a comparison of the SLDs computed by multi-
frequency solution and ZOA method for radial immersions ae= 3 mm and
ae= 5 mm respectively. In both cases, the ZOA method provides acceptable results
for a helix angle of λ= 30°. However, compared with the results for the symmetric
system the zero helix angle also has an impact at higher radial immersions.
30
Multifreq_eig
25 λ= 0°
Multifreq_det
apcrit [mm]
20 λ= 30°
ZOA
15
10
5
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.16 SLDs of the Non-Symmetric MDoF System, ae= 3 mm
20
Multifreq_eig
λ= 0°
15
Multifreq_det
apcrit [mm]
λ= 30°
10 ZOA
0
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.17 SLDs for the Non-Symmetric MDoF System, ae= 5 mm
10.4 Conclusions
The behavior of the mechanical system in combination with the process behavior
has an impact on the stability. This means that the process and machine behavior
determines which effects have to be taken into account for an accurate prediction
of stability boundaries.
In a case study - presented in this chapter - algorithms for the prediction of sta-
bility boundaries in frequency domain have been applied to different combinations
of process and machine characteristics. A comparison with results in time domain
10 Synthesis of Stability Lobe Diagrams 243
has demonstrated that frequency domain algorithms are applicable, even if the sys-
tem shows a time-variant behavior. It has been demonstrated that in some cases a
simplification can be made so that the time efficient zeroth order approximation is
applicable. Table 10.3 shows a pattern of the computational methods for the inves-
tigated combinations of process and machine behavior.
Table 10.3 Pattern of Applicable Frequency Domain Algorithms for the Investigated Ref-
erence Process
Straight-fluted
a e / d ≈ 0.04 Multifreq eig Multifreq eig Multifreq eig
Helicoidal
Multifreq det Multifreq det Multifreq det
a e / d ≈ 0.04
Straight-fluted
a e / d ≈ 0.25 ZOA Multifreq eig Multifreq eig
Helicoidal
a e / d ≈ 0.25 ZOA ZOA ZOA
Straight-fluted
a e / d ≈ 0.4 ZOA ZOA Multifreq eig
Helicoidal
a e / d ≈ 0.4 ZOA ZOA ZOA
This pattern is valid for the investigated reference process. Ongoing works deal
with the definition of classification numbers to expand this pattern for a general
process and machine behavior. Some investigated issues that have to be quantified
are, for example, the influence of tool diameter and cutting force coefficient as
well as the quantification of the influence of the dynamic behavior, i. e. number of
modes and dynamic stiffness.
References
[1] Weck, M., Brecher, C.: Werkzeugmaschinen Fertigungssysteme Bd.5 Messtechnische
Untersuchung und Beurteilung, dynamische Stabilität. Springer, Heidelberg
(2006)
244 K. Großmann and M. Löser
[2] Tlusty, J., Polacek, A.: Beispiele der Behandlung der selbsterregten Schwingungen
der Werkzeugmaschinen, 3. Forschungs- und Konstruktionskolloquium Werkzeug-
maschinen und Betriebswissenschaft (3. FoKoMa). Vogel-Verlag, Coburg (1957)
[3] Tobias, S.A., Fishwick, W.: A Theory of Regenerative Chatter. The Engineer, Lon-
don (1958)
[4] Altintas, Y., Budak, E.: Analytical Prediction of Stability Lobes in Milling. Annals of
the CIRP 44(1), 357–362 (1995)
[5] Bayly, P.V., Halley, J.E., Mann, B.P., Davies, M.A.: Stability of Interrupted Cutting
by Temporal Finite Element Analysis. Journal of Manufacturing Science and Engi-
neering 125, 220–225 (2003)
[6] Insperger, T., Stepan, G.: Semi-discretization method for delayed systems. Interna-
tional Journal for Numerical Methods in Engineering 55, 503–518 (2002)
[7] Budak, E., Altintas, Y.: Analytical Prediction of Chatter Stability in Milling. Part I:
Modelling, Part II: Applications, Transactions of ASME, Journal of Dynamic Sys-
tems, Measurement and Control 120, 22–36 (1998)
[8] Merdol, S.D., Altintas, Y.: Multi Frequency Solution of Chatter Stability for Low
Immersion Milling. ASME Journal of Manufacturing Science and Engineering 126,
459–466 (2004)
[9] Insperger, T., Stepan, G.: Stability Transition between 1 and 2 Degree-of-Freedom
Models of Milling. Periodica Polytechnica Mechanical Engineering 48(1), 27–39
(2004)
[10] Altintas, Y., Stepan, G., Merdol, D., Dombovari, Z.: Chatter stability of milling in
frequency and discrete time domain. CIRP Journal of Manufacturing Science and
Technology 1, 35–44 (2008)
[11] Zatarain, M., Munoa, J., Peigne, G., Insperger, T.: Analysis of the Influence of Mill
Helix Angle on Chatter Stability. Annals of the CIRP 55(1), 365–368 (2006)
[12] Altintas, Y.: Modeling Approaches and Software for Predicting the Performance of
Milling Operations at MAL-UBC. Machining Science and Technology 4(4), 445–478
Chapter 11
Analysis of Industrial Robot Structure
and Milling Process Interaction for Path
Manipulation
Abstract. Industrial robots are used in a great variety of applications for hand-
ling, welding, assembling and milling operations. Especially for machining opera-
tions, industrial robots represent a cost-saving and flexible alternative compared to
standard machine tools. Reduced pose and path accuracy, especially under process
force load due to the high mechanical compliance, restrict the use of industrial ro-
bots for machining applications with high accuracy requirements. In this chapter, a
method is presented to predict and compensate path deviation of robots resulting
from process forces. A process force simulation based on a material removal cal-
culation is presented. Furthermore, a rigid multi-body dynamic system’s model of
the robot is extended by joint elasticities and tilting effects, which are modeled by
spring-damper-models at actuated and additional virtual axes. By coupling the re-
moval simulation with the robot model the interaction of the milling process with
the robot structure can be analyzed by evaluating the path deviation and surface
structure. With the knowledge of interaction along the milling path a general
model-based path correction strategy is introduced to significantly improve accu-
racy in milling operations.
11.1 Introduction
Major fields of machining applications for industrial robots are automated pre-
machining, deburring and fettling of cast parts or trimming of carbon-fiber-
reinforced laminate. Due to its kinematic structure with 6 axes the robot can cover
a large working space and is able to reach difficult workpiece positions so that it
can be applied to perform complex machining operations. Therefore, compared to
standard machine tools, industrial robots offer an economic and flexible machin-
ing alternative. However, industrial robots do not provide a high absolute and re-
peat accuracy. Current industrial robot systems reach a repeat accuracy of 0.06
mm. Under process load, e. g. in milling operation, an additional deflection of the
tool center point (TCP) occurs. Measured deflections of 0.25 mm under process
loads of 100 N in earlier tests [1] have confirmed the expected compliance. There-
fore, when using an industrial robot for milling applications, inaccuracies of the
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 245–263.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
246 J. Bauer et al.
serial robot kinematic, the low structural stiffness and the effective process forces
lead to path deviations. The results are unwanted trajectory deviations, which lead
to errors in dimension and a reduced surface quality of the workpiece. These devi-
ations mainly consist of a static offset overlaid with a low frequency oscillation of
the tool [2].
In order to increase the milling accuracy with present process forces of the mil-
ling operation a model-based path manipulation module for robots is developed.
This module consists of a robot and a milling force model to predict path errors in
advance of a real milling operation. Additionally, the simulation module allows
the investigation and analysis of the interaction of milling force and motion of the
robot. Finally, the proposed strategy for modeling and simulation serves as the ba-
sis for an efficient offline correction of path deviations.
Overall Objective
The machining operation induces process forces, which act on the machine struc-
ture. The high compliance of the structure causes a deviation of the programmed
path at the TCP. This path displacement causes a new cutting condition of the cut-
ter including a variation of the process force (cf. Fig. 11.1). Thus, milling robots
are characterized by a close interaction of the cutting process with the mechanical
robot structure. An early consideration of the interactions of the milling robot in
the production process can reduce manufacturing failures as well as efforts to ma-
nually re-teach the robot’s path.
The general aim is the development of a strategy to improve the accuracy in ro-
bot based milling operations by prediction and offline compensation of static and
low frequency path deviations. Furthermore, the optimization of cutting parame-
ters and a prediction of milling properties (accuracy in dimension, surface struc-
ture) is in the focus.
In order to simulate the milling force a standard cutting force model presented
in [3] is implemented and adapted for industrial robots. The simulation of the
11 Analysis of Industrial Robot Structure and Milling Process Interaction 247
As elastic motions of the robot may not only appear around the axes of the robots
drives but also around additional axes, an extended kinematical model has been
derived. In this section, the kinematic model is discussed using homogeneous
transformations in a 4x4-matrix representation.
For better readability, the abbreviations
1 0 0 x 1 0 0 0
0 1 0 y 0 cos(α ) − sin(α ) 0
Trans ( x, y , z ) := , Rot ( x; α ) :=
1 z 0
,
0 0 0 sin(α ) cos (α )
0 0 0 1 0 0 0 1
cos(α ) 0 sin(α ) 0 cos(α ) − sin(α ) 0 0
0 1 0 0 sin(α ) cos(α ) 0 0
Rot ( y; α ) := , Rot ( z; α ) :=
− sin(α) 0 cos (α ) 0 0 0 1 0
0 0 0 1 0 0 0 1
248 J. Bauer et al.
with θi being the current position of the joint and the constants di, ai and αi de-
scribing the relative position of the next joint. These joints, which are driven to
move the robot, are referred to as actuated joints. It should be noted that this con-
vention does not contain information on the precise placement of the joint. As
is valid, joints can be anywhere on the z-axis of the previous link's coordinate
frame. To allow for a precise placement of the joint the model is extended by the
joint displacement pi, leading to
This model is extended by two variable rotations around axes, orthogonal to the
joint's axis, leading to the final extended kinematics model
with θx,i and θy,i being the rotation angles caused by elasticities around the addi-
tional axes. These additional revolute axes are referred to as virtual joints (cf. Fig.
11.2). It should be noted that these virtual joints are optional. One or both virtual
joints can be added to each link, if additional elasticities need to be modeled for
this link.
Simulating the motion dynamics of the robot requires additional parameters for
each link of the robot. Namely, these are the mass mi , the inertia tensor Ii and the
center of mass comi for each link. In the implementation used for this work, the
center of mass is described with respect to the coordinate frame defined by linkext,i.
The inertia tensor is resolved at the center of mass in a coordinate frame using the
same orientation as the frame defined by the link.
11 Analysis of Industrial Robot Structure and Milling Process Interaction 249
M ( θ )·
θ + C (θ , θ ) + G ( θ ) = τ + S ( Fxyz , tool , θ )
Implementation
+ + =
During a simulated milling operation the tool moves through the workpiece in dis-
crete time steps ti. The incremental size of the time step is dt.
dttooth 2tooth
dt = .
nint
In every time step ti=1,…, nint the material removal is calculated, where dttooth 2tooth
is the tooth-passing time calculated by
60
dttooth 2tooth =
N z ·n
The material removal simulation calculates a point cloud representing the outer
volume of a single chip, which the discrete chip geometry is extracted from. The
chip geometry is described by the angular discretized chip thickness h(φ,z) with
the entry and exit angles φin, φout. As the chip thickness varies along the cutting
edge the chip gets subdivided into discs of the height dz and in dφ in angular di-
rection (cf. Fig. 11.5).
z radial
h(φ,z)
y
y
ap dz
dz
x dφ
x
The chip thickness is calculated in three steps. In the first step, the set of points
is subdivided into disks of the height dz. As a next step, inside each disc the max-
imum angular points are defined as restricting points by φin and φout (cf. Fig.
11.6a). Due to the complex contact conditions and the complex chip geometries
the entry and exit angles vary over the depth of cut ap. The cross-sectional geome-
try of each disk is calculated based on the cylindrical form of the cutter at the dis-
crete time of ti and ti-1. The area between the input angle φin and exit angle φout is
discretized into multiple sub-areas with the angular distance dφ. The value of the
discrete chip thickness h(φ,z) with respect to φ is calculated by a line intersection
with the two cylinders in the final step (cf. Fig. 11.6b).
11 Analysis of Industrial Robot Structure and Milling Process Interaction 253
Adding all disc levels together, a chart of the chip thickness h(φ,z) is extracted,
which forms the base of the force calculation. This is the most frequently per-
formed arithmetic operation during the removal simulation.
For the prediction of the cutting forces a standard cutting force model based on
Altintas [3] is implemented, where the time delay responsible for the chatter is
neglected. According to the chip discretization the cutter is sectioned into discs of
height dz (cf. Fig. 11.7).
In each disc e, Frta,j represents the force per tooth j in radial, tangential and axi-
al direction,
Frta , j,e = K c ·dz·h j (ϕ , z ) + K e ·dz .
254 J. Bauer et al.
Depending on the angular position of the tooth j of a disc, the corresponding chip
thickness hj(φ,z) is inserted. The cutting force coefficients Kc=[Krc, Ktc, Kac] and
Ke=[Kre, Kte, Kae] need to be identified in advance.
A transformation of Frta,j,e with T(φ) and the subsequent summation over all
teeth Nz and discs Ne results in the process force Fxyz,tool, given in a non-rotating
tool coordinate frame,
Ne N z
Fxyz,tool = T (ϕ )·F
e =1 j =1
j rta, j,e .
The force Fxyz,tool acts at the TCP of the robot and thus represents the interface to
the robot model.
4 4 4
2 2 2
0 0 0
1 1 1
2 2 2 1
1 1
3 2 3 2 3 2
3 3 3
Fig. 11.10 Direct compliance within the working area (z = 900 mm)
In a similar way, the rotational and tilting stiffness are experimentally meas-
ured. While the force load is applied in defined directions at the spindle the dis-
tance sensors measure the tilting and rotation at each axis sequentially. In Table
11.2, the measured stiffnesses are summarized.
rotational and tilting stifffness values in order to compromise the measured stifff-
ness, the model gets em mpirical updated. The simulated and the measured stifff-
nesses within the working g area are compared in Figure 11.11.
0 0 0
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
For the model dynamiccs adaption, the mass and damping matrix have to be de-
termined. The mass and the t location of the centers of mass for each link are exx-
tracted from the technicall documentation. The inertia tensors of all rigid links arre
derived from CAD-representations. By calculating the frequency response funcc-
tions (FRF) of the virtuall robot compliance Hxx, Hyy, and Hzz and comparing them m
to the corresponding robo ot’s FRF, the dynamical correlations of the robot moddel
are evaluated (cf. Fig. 11.12). In order to improve correlation, the mass and stifff-
ness model parameters arre slightly adjusted. Finally, the peak height of selecteed
eigenfrequencies is adjustted by modifying the damping parameters to confirm thhe
measured FRFs.
After updating the mod del parameters, the simulated robot compliance results iin
a reasonable correlation to
t the measured compliance frequency response functioon
in the frequency band of f = 0 to 50 Hz. Especially the frequency and amplitude oof
the resonance peaks correelate well.
100
Fx mea
0 Fx sim
-100
0 5 10 15 20 25 30 35 40 45
Zeit [s]
100
Fy mea
0 Fy sim
-100
0 5 10 15 20 25 30 35 40 45
Zeit [s]
0
Fz mea
-200 Fz sim
-400
0 5 10 15 20 25 30 35 40 45
Zeit [s]
The cutting forces are recorded by a Kistler force dynamometer during the ma-
chining operation. In order to calculate the mean cutting forces, the force signal is
filtered by a low-pass filter with a cut-off frequency of fcorner = 100 Hz. As the
main eigenfrequencies of the robot structure are lower than 60 Hz (cf. Table 11.1)
the consideration of frequency bands above 100 Hz are neglected for coupled ro-
bot milling simulations. The comparison of measured and simulated cutting forces
Fx, Fy, Fz (Fig. 11.13) shows a sufficient congruence of an overall quadratic mean
below 20 % of each force direction.
positions, match the ideal set values ( qreal = qset and qreal = qset ) for virtual and ac-
tuated joints. Thus, a path h through a sequence of points on the workpiece is simuu-
lated using the robots kin nematics. This generates corresponding trajectories of thhe
process forces at the TCP P. As the computation is based on a discretization of thhe
workpiece and an infiniteesimal stepsize Δt is deployed in an accurate simulation,
force signals contain significant noise components. Therefore, low-pass filtering is
applied to get smooth trajectories (cf. Fig. 11.14) for further use.
Using the recursive Neewton-Euler-method [4], the inverse dynamics is solveed
to compute idealized to orques τideal(tk) for a selection of ideal joint positionns
{qideal (tk ) | k = 1, ... , kmax } . Then, with τ ideal and with the assumptioon
qcomp = qideal ,
The proposed model-b based strategy is computed very efficiently as only onne
simulation run with an id dealized, kinematic robot model is performed. The evenn-
tual solution of one lineaar equation system per interpolation point can be carrieed
out in real-time.
11.4.2 Compensatio
on by Deviation Mirroring
The second approach of path
p compensation is based on a geometrical considera-
tion of deviations caused by milling forces acting on the cutter (cf. Fig. 11.15(1.))).
The compensation strateg gy is applied in radial and axial direction of the cutteer.
260 J. Bauer et al.
Figure 11.15(2.) illustrates the idea of calculating a new NC-code, based on mir-
roring path points on the planned NC-path. Thus, when the milling operation is
conducted with the modified NC-code, a coincidence of planned and the resulting
TCP position is expected (cf. Fig. 11.15(3.)).
The experimental millling operation is repeated with the modified robot proo-
gram including the comp pensating path points. The deviation is reduced primarilly
within the strait path segment. The root mean square error of erms = 0.72 mm is
now reduced to erms = 0.15 mm. In the corner, only a moderate error reduction waas
achieved because of the roobot’s internally-controlled path blending.
In z-direction, the profile of the machined part with and without compensation
is measured on a coordinate measuring machine. The compensated path is slightly
closer to the supposed NC-path on the straight path segments. In the corner, the
dipping of the cutting tool is reduced by ∆z = 0.029 mm (cf. Fig. 11.18).
11.5 Conclusion
Industrial robots are a cost-saving and flexible alternative for machining applica-
tions in a low-accuracy area. A reduced position and path accuracy, especially un-
der process force load due to the high mechanical compliance, restricts the use of
industrial robots for further machining applications.
The proposed strategy is built on a coupled simulation module consisting of a
process force model and a robot model. This validated simulation enables the pre-
diction of a force-induced path deviation in robot milling operations. Experiments
have shown that a milling error occurs as a summation of robot control-contouring
error and the process force deviation of up to 0.44 mm. Depending on the control
system and the level of cutting force, the presented approach compensates the er-
ror caused by process force deviation to a large extend. The influence of the con-
trol-contouring error on the overall path deviation is more dominant in the case of
only small process forces. Since this control-contouring error could not be covered
in the simulation by the so-far implemented robot model, the path correction
11 Analysis of Industrial Robot Structure and Milling Process Interaction 263
approach results in only moderate improvements for this case of small process
forces. For further improvements, these effects of the robot’s internal path plan-
ning and control algorithms should be identified and implemented into model.
However, in the case of dominant process forces the path compensation con-
cept is able to reduce the milling error significantly.
Acknowledgments. The authors thank Max Stelzer, who carried out very helpful proofs
of concept in the first phase of the project.
References
[1] Abele, E., Bauer, J., Rothenbücher, S., Stelzer, M., Stryk, O.: Prediction of the Tool
Displacement by Coupled Models of the Compliant Industrial Robot and the Milling
Process. In: Conference on Process Machine Interaction, Hannover (2008)
[2] Weigold, M.: Kompensation der Werkzeugabdrängung bei der spanenden Bearbei-
tung mit Industrierobotern. PhD Thesis TU Darmstadt (2008) ISBN 978-3-8322-
7178-7
[3] Altintas, Y.: Manufacturing Automation: Metal Cutting Mechanics, Machine Tool
Vibrations and CNC Design. Cambridge University Press (2000)
[4] Craig, J.J.: Robotics. Addison-Wesley (1989)
[5] Walker, M.W., Orin, D.E.: Efficient Dynamics Computer Simulation of Robotic Me-
chanisms. Journal of Dynamic Systems, Measurement, and Control 104, 205–211
(1982)
[6] Featherstone, R.: Rigid Body Dynamics Algorithms. Springer (2008)
[7] Höpler, R.: A unifying object-oriented methodology to consolidate multibody dynam-
ics computations in robot control, PhD Thesis TU Darmstadt (2004)
[8] Stelzer, M.: Forward Dynamics Simulation and Optimization of Walking Robots and
Humans, PhD Thesis TU Darmstadt (2007)
[9] Friedmann, M.: Simulation of Autonomous Robot Teams with Adaptable Levels of
Abstraction, PhD Thesis TU Darmstadt (2009)
[10] Walther, A., Griewank, A.: ADOL-C: A Package for the Automatic Differentiation of
Algorithms Written in C/C++. Version 2.1.12, Documentation,
https://projects.coin-or.org/ADOL-C
[11] von Stryk, O.: User’s Guide for DIRCOL (Version 2.1): a direct collocation method
for the numerical solution of optimal control problems, TU Darmstadt (2000),
http://www.sim.informatik.tu-darmstadt.de/sw/dircol
[12] Rehling, S.: Technologische Erweiterung der Simulation von NC-
Fertigungsprozessen, PhD Thesis Universität Hannover (2009)
[13] Surmann, T.: Geometrisch-physikalische Simulation der Prozessdynamik für das
fünfachsige Fräsen von Freiformflächen, PhD Thesis TU Dortmund (2005)
[14] Stautner, M.: Simulation und Optimierung der mehrachsigen Fräsbearbeitung, PhD
Thesis TU Dortmund, München (2002) ISBN 3-8027-8732-3
[15] Selle, J.: Technologiebasierte Fehlerkorrektur für das NC-Schlichtfräsen, PhD Thesis
Universität Hannover (2005)
[16] Weinert, K., Müller, H., Kreis, W., Surmann, T., Ayasse, J., Schüppstuhl, T.,
Kneupner, K.: Diskrete Werkstückmodellierung zur Simulation von Zerspanprozes-
sen. In: Zeitschrift für wirtschaftlichen Fabrikbetrieb: ZWF, München, pp. 385–389
(2002) ISSN 0932-0482
[17] KUKA Roboter GmbH. Data sheet KR 210-2. Gersthofen - Germany (July 2009),
http://www.kuka-robotics.com
Chapter 12
Process Machine Interactions in Micro Milling
Abstract. In this chapter, both analytical and experimental studies on the process
stability of micro milling are presented. The investigations are carried out in order
to improve a comprehensive model, which describes interactions of the dynamic
cutting forces and the dynamic machine tool behavior including the end mill. Do-
minant chatter frequencies at different operating points are determined by analyz-
ing the process forces, acoustic signals as well as optical measurement signals.
Results are documented and discussed by means of stability lobe diagrams. The
findings are confirmed by analyses of the milled surfaces. Finally, some sugges-
tions for improving the parameter identification are given.
12.1 Introduction
Significant Process Machine Interactions (PMI) can be observed in micro milling,
e. g. by comparing the surface roughness of the manufactured workpiece to the vi-
bration-response signals measured on the machine tool housing [18]. Alternating
loads caused by interrupted cutting conditions interact with the flexible end mill
and the machine tool structure. Deviations of the Tool Center Point (TCP) are the
consequences.
In this context, a precise modeling of the interacting elements such as the machine
tool structure, the end mill as well as the cutting conditions is necessary for
process simulation. Stability predictions containing this information help to im-
prove the machining of complex micro-structures.
Based on the experimental modal analysis of the machine tool structure and the
operational vibration analysis of the micro end mill a comprehensive spatial mul-
tiple degrees of freedom (MDOF) model is described in this chapter. It is com-
posed of oscillator chains and a flexible tool, which is modeled as a rotating
Euler-Bernoulli beam using finite elements (FE).
The equations of motion are derived with Hamilton’s principle including gy-
roscopic effects as a result of the rotational motion of the end mill. Besides, a
self-excited part due to the internal material damping of the tool is included. The
dynamic cutting forces, which occur during the process, are coupled to the MDOF
model. Therefore, a geometrical milling force model is developed, which contains
significant characteristics of micro cutting such as ploughing effects below the
critical chip thickness.
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 265–284.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
266 E. Uhlmann et al.
Experiments are carried out in order to verify the process stability behavior.
The simulated and experimentally-determined stability lobe diagrams are then
compared and analyzed in detail.
reference points
spindle shaker
-6
x 10
3
mode shapes
|
N/m xx
|H | (25°)
xx
2
Magnitude
|Hyy|
1.5
0.5
0
500 1000 1500 2000 2500 3000 3500 4000 Hz 5000
Frequency
Fig. 12.1 Experimental modal analysis of the machine tool structure at different reference
points
clamping. The measured FRFs in modal analysis 2 are shifted at different posi-
tions, where |Hxx| (25°) means a 25-degree rotation of the Spindle-Tool Holder-
Tool system (STT) from the initial measure position. All these phenomena indi-
cate that the spindle configuration significantly affects the modal results of the
system at non-rotating state. Considering the measured mode shapes, the first
three natural frequencies of the FRFs can be traced back to the sliding carriage of
the machine tool. The fourth mode is supposed to be caused by the clamping con-
dition at the STT system. Considering the significant redundant mass effect of the
accelerometer, the shaker and other unknown influences, the resonance frequency
position and the damping factor of this mode are not reliable and not directly uti-
lizable.
Due to the small dimensions of the micro end mill it is challenging to carry out
a classical experimental modal analysis directly. Figure 12.2 shows an operational
vibration analysis to study the resonance frequency of the tool.
Fig. 12.2 Operational vibration analysis of the micro end mill with a shaker as excited base
analytical stability analysis. The complex dynamics of the system such as the rota-
tion of the spindle-tool-system can be comprised. In this context, a spatial system
model composed of oscillator chains and finite elements (FE) is utilized in order
to incorporate both the machine tool modes and the flexible tool modes. Mean-
while, it also meets the requirements for process simulation and stability
prediction.
In order to identify the system’s parameters by an experimental modal analysis
the model is firstly treated without rotational effects, i. e. regardless of the spindle
speed and the according rotary inertia of the end mill. Subsequently, the rotational
effects are taken into consideration within the formulation of the equations of mo-
tion for further application.
x1 x2 x3 x4 a) b)
y4
jY q6
mx1 mx2 mx3 Y
Z y3 q2 q7
uY
kx1 , dx1 , ... ,kx4 ,dx4 my3 y2 q3 X
s
q5
my2 Y uZ
y1 L q8
X jZ
z my1 q1
y q4
ky1 ,dy1, ... ,ky4 ,dy4 Z
x
Fig. 12.3 a) Spatial system model; b) Finite element of the micro end mill
The global reference frame (xyz) is inertially fixed on the milling plane. Rotat-
ing this frame by 90° about the y-axis defines the non-rotating reference frame
(XYZ) to simplify the description of the element kinematics while considering the
rotational effects.
When using the time-dependent generalized coordinates
at the two nodes of the element, the translational displacements of the element
ue=(uZ, uY)
can be expressed as
12 Process Machine Interactions in Micro Milling 269
uZ = φ1 −φ2 0 0 φ3 −φ4 0 0 q ,
u 0 0 φ φ 0 0 φ φ e (12.1)
Y 1 2 3 4
where T and U are the kinetic and potential energy of the system respectively, δW
is the virtual work of non-conservative forces and δ represents the variational
operator.
Because of the different geometries of the tool shank, the taper and the cutting
edges at the tool tip, it is not accurate enough to model the micro end mills by a
uniform beam. This can be demonstrated by the first three normalized bending
mode shapes at different boundary conditions as shown in Figure 12.4. Note that
the shank, tip lengths and diameters as well as the taper length considerably affect
the natural frequencies and mode shapes of the tools [6]. In this case, a micro end
mill with a cantilever length of 20 mm, a shank diameter of 3 mm, a nominal di-
ameter of 0.5 mm and the tip length of 1 mm is used as a demonstration. It can be
seen that the first bending mode is dominated by the shank. For clamped-free and
sliding-free conditions they are similar to a uniform beam. The second and third
bending modes are dominated by the taper, especially the tip deflections.
Fig. 12.4 Normalized mode shapes of the tool for different boundary conditions
To combine the modes of the tool and the machine, the parameters (mass, stiff-
ness and damping) of the oscillator chains should be identified. Sub-structure
coupling analysis is a useful method for this purpose. The detailed derivation of
this method can be found in [12, 16]. Generally, the entire machine tool system is
divided into two sub-structures: The complex machine tool structure including the
upper part of the tool shank (FRFs can be measured by experimental modal
270 E. Uhlmann et al.
analysis) and the rest of the tool including the difficult to measure tool tip (FRFs
usually obtained using beam theories or FE software). The identification of the
FRFs at the coupling is challenging due to the rotational degrees of freedom
(DOF). There are several methods to experimentally measure rotational FRFs or
analytically derive them from measured translational FRFs [5]. To simplify the
modeling the following assumptions are made.
For the analyzed micro milling machine tool, the clamping conditions be-
tween the end mill, the tool holder, and the spindle are perfect, i. e. sufficient
clamping length and quasi-isotropic contacts. The rotational DOF can be neg-
lected and the beam can be coupled to the oscillators with a sliding-free boun-
dary condition. Since the measure point is approximately located at the contact
position between the end mill and the tool holder, the effect of rotational DOF is
supposed to be excluded. Moreover, a rigid coupling is adopted here, i. e. the
rest of the end mill is coupled at the last oscillator without spring or damper
elements (see Fig. 12.3).
In order to identify the parameters of the oscillators the system model is sepa-
rately treated in the milling plane. In each direction, the parameters can be identi-
fied with the help of the measured and computed receptances using curve-fitting
methods. A demonstration of two typical measurements in x and y-directions is
shown in Figure 12.5. It should be mentioned that the identified oscillator parame-
ters need to be furthermore optimized for chatter prediction, since the resonance
frequencies and the damping factors are unverified especially at the fourth peak,
which can be traced back to the clamping conditions at the tool shank.
x 10-6 x 10-6
2 2
measured |H | measured |H |
xx yy
N/m identified |Hxx| N/m identified |H |
yy
Magnitude
1 1
0.5 0.5
0 0
0 1000 2000 3000 Hz 5000 0 1000 2000 3000 Hz 5000
Frequency Frequency
The mode shapes of the coupled system can be computed with the identified
parameters of the machine tool modes. Figure 12.6 shows the fourth and fifth
normalized mode shapes in y-direction.
12 Process Machine Interactions in Micro Milling 271
Oscillator chain
th
in y-direction 4 mode shape 5th mode shape
As seen, for this kind of system modeling the fourth mode is dominated by the
oscillator chains including the tool as a rigid body. In contrast, the fifth mode in-
cludes a bending mode of the tool. Torsion, tension, compression effects and cor-
responding modes are not included here.
holds, where ωe is the angular velocity vector of the beam elements with respect to
the fixed frame, eX is the unit vector of coordinate X in the fixed frame, Ω is the
constant spindle’s angular velocity. It should be pointed out that after linearization
about small deformations, the rotations (φ2, φ3) can be substituted approximately
by the rotations (φY, φZ) shown in Figure12.3 b).
The total element’s kinetic energy is composed of translational and rotational
parts and can be expressed as
L L
1 1
Te =
20 µ e vTe v e ds + ωTe Θe ωe ds ,
20
(12.5)
272 E. Uhlmann et al.
where μe is the mass per unit length of the finite element, the absolute veloci-
ty ve is the derivative of ue=(uZ, uY) with respect to time in the non-rotating frame
and Θe is the mass moment of inertia of the element. While neglecting the shear
and torsion deformations as well as the axial loads, the potential energy of the
element can be expressed as
L
1
2 0
Ue = EI e u ′′e T u′′e ds , (12.6)
where E is the Young’s modulus, Ie is the second moment of area of the element
and
ue˝=∂2ue/∂s2. (12.7)
The internal viscous damping of the tool is assumed to be proportional to the ele-
ment bending stiffness EIe with an introduced damping factor di [19, 21]. By var-
iational calculation the linearized element’s equations of motion are
e + ( di K e − ΩG e ) q e + ( K e + di ΩK se ) q e = fe ,
Meq (12.8)
where the element matrices are similar as stated in [14, 21]. It is noteworthy that a
self-excited term arises due to the rotation and internal viscous damping of the
tool, which differs from regenerative effects in cutting processes.
The assembled system’s equations of motion consisting of the finite elements
and oscillator’s equations of motion are accomplished with the compatibility con-
straints at the corresponding element nodes. In total, the comprehensive model has
170 degrees of freedom (DOF). It should be noted that the system’s equations ex-
pressed in the fixed frame have a simple matrix form due to the non-rotating oscil-
lator chains. When they are transformed into the rotating frame, time-dependent
terms (cosΩt and sinΩt) appear in the oscillator-related components [17].
where Θm is the critical angle, which can be determined from the negative effec-
tive rake angle γeff (γeff = π/2 – Θm).
12 Process Machine Interactions in Micro Milling 273
f z2 hcr d ⋅ hcr
Rth = + 1 + . (12.11)
4d 2 2 ⋅ f z2
Figure 12.7 shows the critical chip thickness and the theoretical surface roughness
based on the above-mentioned theories.
2 2
theoretical roughness Rth
critical chip thickness hcr
0.4 0.4
µm µm 0.293
0.293
0.2
0.2
1 1
hcr/rβ = 0.1
hcr/rβ = 0.1
0,5 0,5
0 0
0 2 4 6 µm 10 0 2 4 6 µm 10
cutting edge radius rβ cutting edge radius rβ
Fig. 12.7 Influence of the cutting edge radius r on the critical chip thickness hcr (left) and
on the theoretical surface roughness Rth (right)
Since the geometrical estimation of the surface roughness already indicates sig-
nificant influence of the cutting edge radius, the cutting forces and the process sta-
bility are supposed to be affected as well [8]. Figure 12.8 shows a schematic view
of the geometrical conditions at the cutting edge considering this point.
274 E. Uhlmann et al.
workpiece
+ Δy t effective
Δx t rake angle
γeff
cutting
78,°
Ch
edge
end mill
h(ϕ) wedge angle β
ip
y t-τ
+Δ
cutting edge
τ
Δx t- M clearance
angle α
rβ
h(ϕ)
hcr
s
y α workpiece
x
By multiplying (12.12) with a screen function g(φ) (1 or 0), as described in [4], the
interrupted cutting conditions as they occur in milling can be set according to the
cutting parameters. The volume below the critical chip thickness Vpl is also de-
rived geometrically to take the ploughing effect into consideration.
( )
2
rβ − rβ − hcr
2
d
( ) ⋅(r )
2
V pl = rβ2 ⋅ arcsin 2 − rβ − rβ − hcr
2
β − hcr ⋅ 2 (12.13)
2 ⋅ rβ
The cutting forces in radial and tangential direction can now be calculated by
z =a p
(
K ⋅ a ⋅ hφ ) + ( K ⋅ V )
xf
Fr = r p p ,r pl (12.14)
z =0
cutting ploughing
z =a p
(
K ⋅ a ⋅ hφ ) + ( K ⋅ V ) ⋅ (1 + μ )
xf
Ft = t p p ,t pl f (12.15)
z =0
cutting ploughing friction
12 Process Machine Interactions in Micro Milling 275
specific force Kt
2 2
N/mm N/mm
3 3
10 10
2 2
10 10
0 50 100 µm 200 0 50 100 µm 200
depth of cut ap depth of cut ap
5
10
PM X 190 CrVMo 20 (52HRC)
specific force Ka
AmpColoy944
2
N/mm Cu
PMMA
3 CuZn39Pb1
10
ae = d (full immersion cutting)
d = 1 mm (TiAlN coating)
2 n = 29.000 rpm
10
0 50 100 µm 200
depth of cut ap
Fig. 12.9 Specific forces in radial, tangential and axial direction for different materials
Figure 12.10 shows calculated cutting forces under variation of the helix angle
(left) and the cutting edge radius (right).
λ = 20° rβ = 10 µm
fz = 10 µm fz = 10 µm
ap = 70 µm λ = 30° ap = 70 µm λ = 30°
rβ = 1 µm λ = 40° µf = 0,1
4 4
force in x-direction Fx
N N
0 0
-2 -2
-4 -4
4 4
force in y-direction Fy
N N
0 0
-2 -2
-4 -4
0 90 180 deg 360 0 90 180 deg 360
rotation angle φ rotation angle φ
Fig. 12.10 Influence of the helix angle (left) and the cutting edge radius (right) on the
cutting forces
• The feed per tooth fz is constantly set to 10 µm, which reduces the influences of
the ploughing effect,
• all experiments are carried out in full-immersion cutting to reduce high dynam-
ic excitation frequencies at the entrance and exit of the cutting edge.
The used cutting force model corresponds to the implementations described in [4].
Solely the specific force constants Ki are adapted according to the varying depth of
cut ap and the adjusted spindle speed n (see Fig. 12.9).
12.3.2 Results
The verification of the calculated stability charts requires a reliable criterion for
unstable process conditions. Therefore, it is common practice to divide measured
signals into their spectral components by applying a Fourier transformation. In
case of unstable process conditions chatter frequencies can then be detected. Fig-
ure 12.11 exemplarily shows the spectra of measured forces and microphone sig-
nals at a spindle speed of n = 28,000 rpm and increasing values of the depth of cut
ap from 60 µm to 105 µm.
0.2 0.2
Magnitude
0.1 0.1
110 110
µm µm
p
p
a
a
ut
ut
0 80 0 80
fc
fc
0 0
o
2 4 6 kHz 60 2 4 6 kHz 60
pth
pth
10 10
Frequency
de
Frequency
de
Fig. 12.11 Spectra of the force signals (left) and microphone signals (right)
[1, 4, 7]. Even though the spectrum of the microphone signals shows a lot more
noise, presumably due to the minimal quantity lubrication (MQL), chatter fre-
quencies could reliably be found. More frequency contents can be seen in the
spectrum of the force signals. The most significant is the cutting edge engagement
frequency. Others are the spindle frequency and its multiples. All decisions as to
whether an operation point is stable or unstable could be made from measurements
from both devices yielding the same result.
Considering the large number of measured data a way of clearly representing
the chatter frequencies is needed. For this purpose, the Campbell diagram is pre-
destinated. Here, the chatter frequencies at different spindle speeds are visualized.
Spindle speed dependencies of the chatter frequencies and possible causes for
non-linearities or other phenomena can be analyzed in detail. Figure 12.12 shows
Campbell diagrams for unstable operating points between 26,000 rpm and
33,000 rpm.
Fig. 12.12. Campbell diagrams from the force (left) and microphone spectra (right)
The parallel dashed lines signify the multiples of the spindle frequency. Espe-
cially in the Campbell diagram of the force signal it can be seen that some fre-
quencies are located on those lines. All frequencies between the dashed lines are
possible chatter frequencies. According to the theory of possible Hopf chatter fre-
quencies [7], the chatter frequencies correspond to basic eigenfrequencies between
7 kHz and 7.2 kHz, which is presumably caused by the STT system. Therefore,
the chatter frequencies are compared to calculated spindle-speed-dependent chat-
ter frequencies. A good accordance can be stated for both diagrams. However, the
microphone spectra are considered to be the superior method for chatter detection.
Figure 12.13 shows the analytically computed stability charts, calculated by using
the previously detected chatter frequencies, as well as the experimentally deter-
mined stability charts.
280 E. Uhlmann et al.
250
µm
unstable cutting
depth of cut ap
150
100
stable cutting
50
0
26.000 27.000 28.000 29.000 30.000 31.000 rpm 33.000
Spindle Speed
Fig. 12.13 Experimentally determined stability chart and simulated stability lobes with dif-
ferent chatter frequencies
It can be seen that the fourth mode of the machine tool model is dominant. The
measured basic chatter frequency corresponds to this mode. The asymmetric mod-
eling in x and y-directions also takes the mode interaction into account. Influences
of other modes cannot directly be determined from the chatter frequencies due to
their small significance compared to the fourth mode. However, their effects can
be observed from the milled surfaces as shown in Figure 12.14.
Fig. 12.14 Milled surfaces with stable cutting conditions at 28,000 rpm (left), unstable cut-
ting conditions at 31,000 rpm (middle) and at 29,000 rpm (right)
The milled surface with stable conditions shows steady traces of the cutting
edge trajectories. In contrast to that, unstable cutting conditions lead to different
surfaces. Effects of overlaid vibrations can be seen additionally to the kinematic of
the cutting edge trajectories. At 29,000 rpm, high-frequent deviations of the TCP
12 Process Machine Interactions in Micro Milling 281
are reasons for that. This corresponds to the fourth mode above of 7 kHz. At
31,000 rpm, a waviness caused by lower frequent vibrations can be seen. It is sup-
posed to be caused by the first three modes.
The chatter frequencies as seen in Figure 12.11 and Figure 12.12 vary in their
magnitude as well as in their frequency value. One reason for that are changing
conditions of the cutting process and the machine tool structure. Analyses of the
tool holder and the clamping conditions are carried out in order to point out the
most significant influences. Figure 12.15 shows the run-out as a function of
the spindle speed measured at the tool shank. Also, the chatter frequencies at
different clamping conditions are analyzed.
10
µm
0.02
6
deflection δ
velocity
4 m/s
loose
2
0
0 0 2 4
10.000 20.000 30.000 40.000 rpm 60.000 6 kHz stiff
Frequency 10
Spindle Speed
Fig. 12.15 Run-out as a function of spindle speed (left) and chatter frequency of the end
mill at different clamping conditions (right)
It can be seen that the tool run-out is strongly influenced by the spindle speed.
Therefore, the consideration of this effect within the comprehensive model is ne-
cessary. High magnitudes of frequencies, also at odd multiples of the spindle, and
tool engagement frequency in force can be explained by that.
A significant influence on the dominant chatter frequency depending on the
clamping conditions of the tool can also be stated. The difference of the chatter
frequency can amount up to 2 kHz, which strongly affects the calculation of the
stability chart.
12.4 Optimization
As depicted in the preceding section, chatter frequencies strongly depend on the
variation of tool clamping and process conditions. This complicates the analytical
determination of stability charts. A reliable method to identify chatter frequencies
without extensively repeating the milling operations at different depths of cut ap
would therefore be helpful. For this purpose, an experimental set-up is illustrated
in Figure 12.16.
Piezo-ceramic actuators are clamped between the workpiece and the clamping
vice on the machine tool. The workpiece is excited with different signals during
the milling process in order to evoke chatter.
282 E. Uhlmann et al.
clamping vice
micro end mill -3
x 10
4
velocity
m/s
80
µm
0 60
p
ta
0 50
cu
2 4 6 kHz 40
of
Frequency 10
laser beam
h
piezo-actuators
pt
from Vibrometer
de
experimental setup without excitation
1 1
-1
(m/s )/V 0.5
80 80
µm µm
0 60 0 60
p
p
a
a
50 50
ut
ut
0 2 4 0 2 4
c
c
6 kHz 40 6 kHz 40
of
of
Frequency 10 Frequency 10
h
h
pt
pt
de
de
excitation with chirp signal
During the milling operation, the velocity of the micro end mill is measured us-
ing a laser vibrometer. Chatter frequencies can be determined from the spectra of
the velocity of the end mill as soon as the process reaches the critical depth of cut
ap (in this case ap = 60 µm). When a chirp signal is applied to the piezo-actuators
the chatter frequencies can already be detected from the frequency response func-
tion (velocity/voltage) at a depth of cut of ap = 50 µm. The similar phenomenon is
also observed using noise signals as excitation, with more frequency contents,
however.
Thus, this method allows the location of chatter frequencies at a very early
stage of instability. Furthermore, no measurement equipment has to be applied,
which possibly changes the dynamic behavior. The intended milling parameters
can be used and the tool wear can be reduced compared to conventional methods.
12.5 Conclusion
This chapter has presented both analytical and experimental studies on the process
stability of micro-milling . Experimental modal analyses of the machine tool struc-
ture as well as of the end mill were carried out. Subsequently, a comprehensive
model including the machine tool dynamics as well as the process forces, was de-
veloped in order to calculate stability charts. The model includes rotational and
12 Process Machine Interactions in Micro Milling 283
gyroscopic effects of the spindle-tool holder-tool-system (STT). The model for the
dynamic process forces is based on a geometrical description of the cutting
process. Influences of the helix angle as well as the cutting edge radius were taken
into account. Experimental determinations of the specific cutting forces were then
carried out to complete a preliminary parameter identification.
Using the model, calculated and experimentally determined stability charts
were compared. Measured force, acoustic and optical signals were analyzed and
employed as chatter criterion. The spindle speed dependence of the chatter fre-
quencies were visualized in Campbell diagrams in order to verify the basic chatter
frequencies. It is noticeable that the first bending resonance frequency of the ana-
lyzed end mills under stiff clamping conditions was detected at 8.7 kHz whereas
the measured chatter frequencies during the milling process were between 7 kHz
and 8.5 kHz. It can therefore be concluded that the chatter frequencies are not only
due to the dynamic behavior of the end mill but arise from complex interactions of
the structure eigenfrequencies. Measurements of the milled surfaces confirm this
presumption. Since the chatter frequencies significantly influence the stability of
the system, experiments on the tool run-out and the clamping conditions of the
tool holder were carried out. Divergences of the chatter frequencies of up to 2 kHz
were detected. As a result of the sensitive dynamic behavior an experimental setup
for chatter detection was introduced. The workpiece was excited with different
signals by piezo-electric actuators. Thereby, a reliable in-situ detection of the chat-
ter frequencies without modifying the dynamic system (e. g. by the use of addi-
tional measurement equipment) could be realized.
References
[1] Altintas, Y., Budak, E.: Analytical prediction of stability lobes in milling. CIRP An-
nals 44, 357–362 (1995)
[2] Bayly, P.V., Halley, J.E., Mann, B.P., Davies, M.A.: Stability of interrupted cutting
by temporal finite element analysis. ASME J. Manuf. Sci. Eng. 125, 220–225 (2003)
[3] Brammertz, P.H.: Die Entstehung der Oberflächenrauheit beim Feindrehen- Bericht
aus dem Laboratorium für Werkzeugmaschinen und Betriebslehre der TH Aachen,
Industrie-Anzeiger, Essen, Nr.2, pp. 25–32 (1961)
[4] Budak, E., Altintas, Y.: Analytical Prediction of Chatter Stability Conditions for Mul-
tidegree of Freedom Systems in Milling. Part I: General Formulation, Part II: Appli-
cation of the General formulation to Common Milling Systems. Transactions of the
ASME J. Eng. Ind. 120, 22–36 (1998)
[5] Duarte, M.L.M., Ewins, D.J.: Rotational degrees of freedom for structural coupling
analysis via finite-difference technique with residual compensation. Mech. Syst. Sig-
nal Process 14(2), 205–227 (2000)
[6] Filiz, S., Ozdoganlar, O.B.: A three-dimensional model for the dynamics of micro-
endmills including bending, torsional and axial vibrations. Precis. Eng. 35, 24–37
(2011)
[7] Insperger, T., Stépán, G.: Updated semi-discretization method for periodic delay-
differential equations with discrete delay. Int. J. Numer. Methods Eng. 61, 117–141
(2004)
284 E. Uhlmann et al.
[8] Jun, M.B.G., DeVor, R.E., Kapoor, S.G.: Investigation of the Dynamics of Microend
Milling—Part II: Model Validation and Interpretation. ASME J. Manuf. Sci.
Eng. 128, 901–912 (2006)
[9] Kienzle, O., Victor, H.: Spezifische Schnittkräfte bei der Metallbearbeitung.
Werkstatttechnik und Maschinenbau 47(5), 224–225 (1957)
[10] L’Vov, N.P.: Determining the minimum possible Chip-Thickness. Machines & Tool-
ing 40(4), 45–46 (1968)
[11] Malekian, M., Park, S.S., Jun, M.B.G.: Investigation of Critical Chip Thickness and
Micro Ploughing Forces. In: Proceedings of CIRP 2nd International Conference on
Process Machine Interactions, Vancouver, Canada (2010)
[12] Mascardelli, B.A., Park, S.S., Freiheit, T.: Substructure coupling of microend mills to
aid in the suppression of chatter. ASME J. Manuf. Sci. Eng. 130, 011010 (2008)
[13] Moriwaki, T., Shamoto, E.: Ultraprecision Ductile Cutting of Glass by Applying Ul-
trasonic Vibration. Annals of the CIRP 41(1), 141–144 (1992)
[14] Nelson, H.D., McVaugh, J.M.: The dynamics of rotor-bearing systems using finite
elements. ASME J. Eng. Ind. 98(2), 593–600 (1976)
[15] Schauer, K.: Entwicklung von Hartmetallwerkzeugen für die Mikrozerspanung mit
definierter Schneide, Dissertation, TU Berlin (2006)
[16] Schmitz, T.L., Donaldson, R.R.: Predicting high-speed machining dynamics by sub-
structure analysis. CIRP Ann. - Manuf. Technol. 49(1), 303–308 (2000)
[17] Shi, Y., Mahr, F., von Wagner, U., Uhlmann, E.: A spatial multiple degree of free-
dom machine tool model for micro milling simulation. In: Proc. 2nd CIRP-PMI,
MM06, Vancouver (2010)
[18] Uhlmann, E., Mahr, F., Shi, Y., von Wagner, U., Eßmann, J.: Interactions between
mechanical vibrations and surface roughness during the micro milling process. In:
Proceedings of CIRP 1st International Conference on Process Machine Interactions,
Hannover, Germany, pp. 978–973 (2008)
[19] Wauer, J.: Kontinuumsschwingungen. Vieweg+Teubner Verlag, Wiesbaden (2008)
[20] Yuan, Z., Zhou, M., Dong, S.: Effect of diamond tool sharpness on minimum cutting
thickness and cutting surface integrity in ultraprecission machining. Journal of Mate-
rials Processing Technology 62, S.327–S.330 (1996)
[21] Zorzi, E.S., Nelson, H.D.: Finite element simulation of rotor-bearing systems with in-
ternal damping. ASME J. Eng. Power 99(1), 71–76 (1977)
Chapter 13
Numerical Computation Methods for Modeling
the Phenomenon of Tool Extraction
Abstract. Tool extraction is a phenomenon, where the end mill slips out of the
shrink-fit chuck in axial direction during the cutting process. This leads to severe
damage of the workpiece, the tool and in some cases even the machine spindle. So
far, this is an unexplained problem with no repeatability. In this article, experi-
mental investigations such as scanning electron microscopy (SEM) and residual
stress measurements on the clamping surface of shrink-fit chucks affected by tool
extraction are presented. Furthermore, results from experiments on a special test-
rig and a mathematical approach, which aims at the prediction of failures due to
Process Machine Interaction, are described. Within the mathematical approach, a
finite element model of the tool and the tool holder is linked with a cutting force
simulation. The dynamic behavior of the spindle is determined by frequency re-
sponse function measurements. From these measurements, a modal model is de-
duced and coupled with the finite element model of the tool holder. The presented
mathematical model is used to compute the resulting stresses in the interface of
those components due to process forces.
13.1 Introduction
Thermal shrink-fit tool holders are widely used in high speed (HSC) and high per-
formance cutting (HPC), where high torque transmission and excellent concentric-
ity is required [1]. The clamping mechanism of shrink-fit chucks is based on an
interference fit between the chuck and the tool shank. By heating up the chuck,
typically by a high-frequency inductor, the inner diameter extends and the tool can
be fed into it. At cooling down, the chuck reverts to its original size and thus
clamps the tool. In comparison to other tool holders the clamping forces of shrink-
fit chucks are very high and concentricity typically is below 3 µm at 3 times the
tool diameter. Hence, especially for HSC milling operations, where spindle speeds
exceed 10,000 rpm, shrink-fit tool holders are commonly used due to the simplici-
ty of the tool holder’s design, which contributes to its balance. There are no
clamping mechanisms, which have to be counterbalanced (Fig. 13.1) [2].
Analytical dimensioning of shrink-fits can be found in DIN 7190 [3], which is
adequate for calculating the main parameters such as the joint pressure and the re-
quired interference depending on the intended loads. Rondé [4] performed
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 285–308.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
286 B. Denkena et al.
Tool extraction is a phenomenon, where the end mill slips out of the shrink-fit
chuck during the cutting process, which leads to severe damage of the workpiece,
the tool and in some cases even the spindle [5, 6]. There is also the risk of injury
to the worker by splintering parts. Tool extraction is a problem mainly reported
from high speed or high performance cutting [7].
Figure 13.2 shows a milling cutter, which slipped out of the shrink-fit chuck
and broke off during high speed milling of aluminum. From reported cases it is
known that an extraction of several millimeters sometimes occurs after only a few
minutes of operating time. In other cases, shrink-fit malfunction appeared after
months of operation. Until now, no specific repeated pattern has been recognized.
13 Numerical Computation Methods for Modeling the Phenomenon 287
The appearance of tool extraction can also not be connected to obvious chatter vi-
brations. Furthermore, there are no obvious faults related to the shrinking process.
Hence, the cause of this phenomenon is presently unknown but it is legitimately
assumed that it is an effect from the interaction of the cutting process and the
tool/tool holder/spindle dynamics. Thus, for investigating the phenomenon of tool
extraction the dynamic behavior of the spindle as well as the milling process have
to be considered.
Solution Approach
The next section presents selected damage cases and shows results from ana-
lyses of the accordant shrink-fit holders. Furthermore, the test-rig for reproducing
the tool extraction experimentally is described as well as results from those expe-
riments. Subsequently, the simulation model of the tool and the tool holder is in-
troduced, followed by simulation results and a short summary.
Fig. 13.5 Tool extraction at titanium workpiece and broken tool with tool holder
With an X-ray diffractometer of the type GE XRD 3000 P the stresses were
measured in axial (x) and circumferential (y) direction. For the presented data,
damaged shrink-fit chucks with different tool diameters were analyzed. The re-
sults were compared to new shrink-fit chucks and to those already subjected to
several clamping cycles.
Figure 13.7 shows the mean values of the residual stresses divided into three
groups. New shrink-fit chucks, those subjected to several clamping cycles for ref-
erence and holders from damage cases. Subdivided by circumferential and axial
direction, mean values for each of the three groups were calculated and related to
the mean stresses of the new shrink holders.
The mean stress amplitudes proved that – compared to the new and normal-
lyused – the residual stresses in holders affected by tool extraction are significant-
ly higher than in new or normally used chucks. Especially in the circumferential
direction, the residual stresses increased at a higher percentage than in axial direc-
tion. This indicates that the material properties in the clamping surface are not
only affected in axial direction when tool extraction occurs but rather in circumfe-
rential direction.
13 Numerical Computation Methods for Modeling the Phenomenon 291
Thus, with the help of SEM measurements the clamping surface is further in-
vestigated. In Figure 13.8 a) a section of the clamping surface of a shrink-fit chuck
being used for several clamping cycles without tool extraction is shown. On the
surface, the grinding grooves from the manufacturing process can clearly be iden-
tified in circumferential direction. Moreover, the surface is flattened due to the
previously applied clamping pressure.
The bright marks on the clamping surface of the shrink-fit chuck in Figure 13.8
b), which was subjected to tool extraction, corroborate the theory of the tool shank
revolving and flexing in the clamping bore and piece-wise slipping in the axial di-
rection.
Thus, neglecting thermal effects so far, it seems that not only axial slipping be-
tween tool shank and shrink-fit chuck but rather circumferential slipping is a main
phenomenon when tool extraction occurs. Hence, it is assumed that, due to dy-
namic interaction between the tool and the workpiece, tool extraction is a combi-
nation of axial and circumferential slipping. Once static friction is lost, axial force
induced by the right-handed helix angle of the milling cutter leads to a relative
axial movement.
More than 100,000 tests were performed with diameter 12 and 16 mm tool-
tool-holder combinations at usual interference fits. However, test results showed
that dynamic loads applicable by the test-rig are too low to enforce shrink-fit fail-
ures in terms of relative axial or rotational displacement of tool shaft and tool-
holder. This might not only be explained by the forces but also by differences in
the structural dynamic behavior of the tool/tool holder/spindle assembly in the
test-rig compared with that in a cutting process. Thus, widened shrink-fit chucks
were used for further tests. At these chucks, the clamping bore was previously wi-
dened by honing. With this process, the cylindricity and surface quality of the
clamping bore are only minimally affected as honing is also the finishing process
in the production of shrink-fit chucks.
In the following, results from tests with widened shrink-fit chucks with a no-
minal diameter of 16 mm are presented. In these tests, tool extraction occurred
under several dynamic loads. The parameters defining the dynamic load are ampli-
tudes of radial and axial force, excitation frequency and torque. The loads in radial
(Fr(t)) and axial (Fa(t)) direction are then defined as a harmonic function:
Fr ( t ) = Ax sin ( 2π f x t ) + F0, x + Ay sin ( 2π f y t + ϕ xy ) + F0, y
(13.1)
Fa ( t ) = Az sin ( 2π f z t + ϕ xz ) + F0, z
Figure 13.11 (bottom) shows the distribution of failure occurrence related to exci-
tation frequency. The histogram includes all experiments, where an axial dis-
placement and/or a rotational displacement between tool shank and tool holder
occurred (both referred to as “failure”). Obviously, more than 90 % of the failures
occurred around an excitation frequency of 400 Hz, the third resonance in
y-direction of the rotor shaft/tool/tool holder/spindle system, even though there are
higher amplitudes at other frequencies. Thus, it can be concluded that frequencies
exciting mode shapes with high bending at tool and tool holder promote tool ex-
traction. Furthermore, test analyses revealed that the rotational load also plays an
important role as tool extraction has only happened under significant torque so far.
Of course, other parameters, especially the interference fit, are also responsible for
the combination of load parameters leading to tool extraction.
In Figure 13.12, an example of tool extraction occurring during a test is shown.
The excitation frequency fy is 398 Hz with an amplitude Ay of 1,250 N and a tor-
que load M of 35 Nm. The bottom right plot shows the rotational displacement be-
tween tool shank and tool holder. At the time the torque load is applied, the tool
starts revolving in the shrink fit chuck. At approximately the same time, the tool
moves axially out of the holder, both movements slightly decrease with time. It is
characteristic within this test that an axial displacement can occur even without an
axial load. In most of the tests, the rotational displacement is much higher than the
axial displacement.
13 Numerical Computation Methods for Modeling the Phenomenon 295
Fig. 13.11 Amplitude y2 against frequency (top) and occurrence of failures (bottom)
Fig. 13.13 Clamping surface of shrink-fit chuck without (a) and with tool extraction (b - d)
wave equation. Equipped with appropriate boundary and initial conditions, the
displacement field u(x,t) occurring in the system of tool and tool holder during the
process satisfies the following boundary and initial value problem (BIVP).
( x, t ) − β div σ ( u ( x, t ) ) − div σ ( u ( x, t ) ) = f ( x, t )
ρ ( x)u in Ω × ( 0, T )
u ( x, t ) = U ( x, t ) on Γ D × ( 0, T )
(13.2)
σ ( u ( x, t ) )·n = T ( x, t ) on Γ N × ( 0, T )
u ( x, 0 ) = u ( x, 0 ) = 0.
In the above problem, the domain Ω represents the structure of tool and tool hold-
er and its boundary ∂Ω is decomposed into the two disjoint sets ΓN and ΓD. The
given data is the density function ρ(x), which is assumed to be independent of
time. Moreover, volume forces f(x,t) as well as boundary tractions T(x,t) on the
Neumann boundary part ΓN and inhomogeneous displacements U(x,t) on the Di-
richlet boundary part ΓD are assumed to be given during the whole process. More
precisely, the boundary tractions result from the cutting process and act at the tip
of the tool and the inhomogeneous displacements result from the dynamic beha-
vior of the spindle interacting with the tool holder. Both the cutting forces and the
vibrations of the spindle were modeled separately to save a time-consuming finite
element model consisting of all of the four mentioned structures (spindle, tool,
tool holder and workpiece), see also the upper right image in Figure 13.3. In the
following sub-sections, the finite element model of the tool and the tool holder,
the model of spindle dynamics, the cutting force model and the coupling of the
three models are described in detail.
The shrinking process is approximated by a linear heat strain law. The corres-
ponding contact pressure is modeled by applying a negative change of temperature
to the tool holder. The required temperature change is computed with the follow-
ing formula
Δr
ΔT = (13.3)
αT ri
where ∆r denotes the interference fit, αΤ represents the coefficient of thermal ex-
pansion of the tool holder and ri considers the inner radius of the tool holder. This
approach leads to a volume force denoted by f(x,t) in the BIVP, which is indepen-
dent of time. In the weak formulation, the volume force reads
where ψ(x) and ϕ(x) denote temporal and spatial basis functions respectively and
λ(x) and μ(x) are the Lamé coefficients depending on the materials.
To simulate the process numerically a transient three-dimensional finite element
method is implemented considering the close-to-process structure of tool and tool
holder. Therefore, the elastic wave equation is re-written into a system of first or-
der equations in time by introducing the velocity v as an additional unknown vari-
able such that the DGM in time can be applied
u − v = 0
(13.5)
ρ ( x ) v ( x, t ) − β div σ ( v ( x, t ) ) − div σ ( u ( x, t ) ) = f ( x, t ) .
By applying the DGM in time and standard FEM in space to the above problem,
the corresponding discrete weak formulation results in a linear system of equa-
tions for the coefficients of the finite element basis functions of displacement and
of velocity. To circumvent large displacements due to the rotation of the system
the displacements and the velocities are described in rotational coordinates. By
further applying the approach of Li and Wiberg [9] the system can be decoupled
into a system concerning the velocity and three linear equations for the displace-
ment coefficients. Finally, the stiffness matrix resulting from the divergence term
in the elastic wave equation is used to approximate the material damping with
damping factor β within the process. Let ⊗ be the tensor product, then the two li-
near systems are:
13 Numerical Computation Methods for Modeling the Phenomenon 299
− k −
M ρ v + F1 − 2 Au
v1
Cρ ⊗ M ρ + CC ⊗ M C + C A ⊗ A · v 2 = k −
F2 − Au
(13.6)
v 2
3
k −
F 3 + Au
3
k
Au − + Av 3
15
A 0 0 u1
− k k k
0 A 0 · u =
2 Au + Av 1 + Av 2 − Av 3
(13.7)
0 0 Au 2 2 3
3
− k Av + k Av + k Av
1 2 3
4 4 10
Within the systems the matrix Mρ denotes the mass matrix, MC the mass matrix
resulting from the coriolis force and A = K – ω2MZ the sum of the stiffness matrix
K and the mass matrix MZ resulting from the centrifugal force multiplied by the
square of the revolution velocity ω. The three by three matrices Cρ, CC, and CA
(see 13.8) are the corresponding coefficient matrices resulting from the analytical
time integration of the DGM and the decoupling of the system, where ϕι are the
hat basis functions and the first bubble function in time.
12 1 −1
−32 −1 1
8 β24+ 3 k 4β +k −10 β − 3 k
4β +5k
2 3 3 3 24 60
−1 8 β +3k −10 β − 7 k
Cρ = −21 , CC = kω 3 , C A = k 24
−2
(13.8)
1 1 1
2 3 3 3 24 60
13 −1
0 13 1 −4 −10 β − 7 k −10 β − 3 k 30 β +13 k
3 3 15 60 60 225
The time step is denoted by k and the vectors u– and v– represent the coefficient
vectors of the displacement and the velocity at the end of the previous time inter-
val respectively. Finally, the vectors F1, F2, and F3 correspond to the right hand
side vectors resulting from the volume force and the boundary conditions. The two
linear systems have to be solved in each time step.
A prediction of an extraction of the tool can be made by observing the stresses
occurring at the interface of tool and tool holder during the process simulation.
Using the DIN 7190 norm [3], which concerns the necessary contact pressure of
shrunk tools, an extraction may be possible, if the ratio of contact pressure and
tangential stresses falls below a determined bound. Of course, this may occur only
locally so that experimental tests have to be made to validate the simulations.
Fig. 13.15 Measurement of spindle dynamics with HSK flange from separated tool holder
the interface between tool holder and spindle. The first six components of x(t) de-
note the three translational and rotational degrees of freedom at the center point P.
Figure 13.16 compares the drive point measurements with the estimated main de-
grees of freedom in radial x and y-direction from the transformed modal model.
As can be clearly seen, the FRFs estimated from the transformed modal model re-
produce the measurements well.
Fig. 13.17 Model for determining the cutting edge trajectories and the cutting forces [23]
the corresponding process force at rotation angle φ can be computed as the sum
over all chip numbers and all nz cutting edges to
nz
Fxyz (ϕ ) = {Fxyz } .
n
i
(13.12)
j
j =1 i =1
However, this procedure assumes an ideal stiff tool, which is not the case in reali-
ty. For this reason, the model is extended considering self-excitations of the tool,
which may be caused by the process forces. In this way, the chip thickness hij is
computed by considering the cutting edge trajectory at the rotation angle φ as well
as a given deviation of the cutting edge at the same position and time. This will re-
sult in changes of the undeformed chip cross-section Aij and therefore in different
process forces.
into account. Furthermore, the structure is affected by the vibrations of the spin-
dle, which also result from the process forces. Hence, it is essential to couple the
cutting force model on one side and the model of spindle dynamics on the other
side with the transient finite element model of tool and tool holder, see Figure
13.18.
Fig. 13.19 Coupling of the cutting force model and the FE-simulation of tool and tool
holder [6]
On the other side, the vibrations of the spindle excited by the cutting forces in-
teract with the tool holder at the corresponding interface, see Figure 13.18. When
the dynamic behavior of the spindle is known in terms of the ODE (13.9), it is
coupled via a DEN mapping with the FE model of tool and tool holder by the in-
homogeneous Dirichlet data U(x,t) in the BIVP (13.2). Here, the Dirichlet boun-
dary ΓD is defined as the interface between tool holder and spindle. Starting with
U(x,0) = 0 the displacement U(x,ti) is approximated by the solution x(ti) of the
ODE (13.9). In detail, for every x on ΓD and every time step ti, the corresponding
displacement is the sum of the translation in the first three components of x(ti) and
the rotation of x around the center point P with rotation angles in the fourth to
sixth component of x(ti)
The force vector F(t) is necessary to solve the ODE (13.9). The deflection of
the tool, which is caused by the cutting forces, results in an oscillation of the Di-
richlet boundary ΓD. Hence, the forces and moments in the first six components of
the force vector F(t) on the right hand side of (13.9) are computed by the solution
vectors u1, u2,, and u3 of the linear system (13.7). By computing the stress tensor
σ(u) due to Hooke’s law, the forces and moments are computed to
{F(ti )}13 := σ (u ( x, ti )) ⋅ n dx
ΓD
(13.12)
{F (ti )}46 := ( x − P ) × (σ (u ( x, ti )) ⋅ n)dx
ΓD
where n denotes the normal outward to the Dirichlet boundary ΓD. Since the ei-
genfrequencies expressed by the model of spindle dynamics are not affected by an
excitation the last 28 components of F(ti) are set to zero. To ensure that the correct
data is exchanged within the coupling an iterative loop, as shown in Figure 13.19,
is applied for every time step. This leads to a series of force vectors Fj(ti). Again
the inner iteration stops, if the difference of two consecutive force vectors
F j (ti ) − F j −1 (ti ) falls below a predefined bound. The cycle for a fixed time step
ti is visualized in Figure 13.20.
13 Numerical Computation Methods for Modeling the Phenomenon 305
Fig. 13.20 Coupling of the spindle dynamics and the FE-model of tool and tool holder [6]
Fig. 13.21 Transient simulation of process forces considering dynamic tool deflection
The effect of the coupling, i. e. the consideration of the tool deflection within
the cutting force model is visualized in Figure 13.21. The figure shows the forces
in radial (Fr), axial (Fa) and tangential direction (Ft), which act on the cutting
edges of the tool during one revolution of the cutting process. The continuous
lines represent the forces from the cutting force model considering tool
deflections, whereas the dashed lines are computed by the cutting force model
306 B. Denkena et al.
considering an ideally stiff tool. It turns out that the consideration of the tool dy-
namics has significant influence on the amplitude of the cutting forces, which
should not be neglected. Furthermore, effects such as regenerative chatter can only
be considered, if the models are coupled.
With the help of the coupled models, simulations with various load cases are
performed to investigate the stresses in the contact zone of the tool and the shrink-
fit chuck. As an example, Figure 13.22 shows the change in tangential stress in
axial direction related to the initial stress state of the shrunk tool during a transient
process simulation for a single time step.
Fig. 13.22 Change in tangential stress in tool/tool holder interface due to cutting process
The prediction of relative movement of the tool and the tool holder is made
possible by observing the stresses in the contact zone. If the tangential stress ex-
ceeds a certain limit defined by the normal stress and a friction coefficient tool ex-
traction is assumed, the simulation is stopped.
Summary
The article shows experimental and mathematical approaches for the investigation
of the phenomenon of tool extraction. The analysis of damage cases exhibits that
tool extraction is caused by the dynamic interaction of the tool/tool holder/spindle
structure with the cutting process. SEM and residual stress measurements of the
clamping surface of damaged shrink-fit holders corroborate the thesis that the tool
shank revolves in the clamping bore while simultaneously slipping out in axial di-
rection. With the help of a test-rig the phenomenon of tool extraction is repro-
duced under various load cases and main parameters are analyzed.
Furthermore, simulations with a finite element model of the tool and the tool
holder are performed concentrating on the stresses in the interface between the
tool shank and the tool holder. In order to consider the dynamic behavior of
13 Numerical Computation Methods for Modeling the Phenomenon 307
References
[1] Schulz, H., Rondé, U.: Werkzeuge schrumpfspannen. Werkstatt und Betrieb 127(11),
873–875 (1994)
[2] Eastman, M.: Shrink-Fit Toolholding, Cutting Tool Engineering 49(3) (1997)
[3] Deutsches Institut für Normung e.V., Pressverbände - Berechnungsgrundlagen und
Gestaltungsregeln. Beuth Verlag, Berlin (February 2001) 17.040.10; 21.120.10/
[4] Rondé, U.: Untersuchung von Systemen zum Spannen von Zylinderschaftwerkzeugen
unter besonderer Berücksichtigung ihrer Eignung für die Hochgeschwindigkeitsbear-
beitung, Dissertation, Darmstadt (1994)
[5] Denkena, B., Stephan, E.P., Maischak, M., Heinisch, D., Andres, M.: Numerical
computation methods for process-oriented structures in metal chipping. In: Denkena,
B. (ed.) 1st International Conference on Process Machine Interactions, vol. 1, pp.
247–256. PZH Produktionstechn, Zentrum (2008)
[6] Denkena, B., Stephan, E.P., Maischak, M., Heinisch, D., Andres, M., Krüger, M.:
Investigations on Dynamic Tool, Structure and Process Interaction. In: Altintas, Y.
(ed.) 2nd International Conference on Process Machine Interactions, Vancouver
(2010)
[7] Fladerer, F.: Haltemomente sind beim Schrumpfen ein Maß für die Produktivität. Ma-
schinenmarkt 31(32), 28–29 (2007)
[8] Johnson, C.: Numerical solution of partial differential equations by the finite element
method. Dover Publ., Mineola (2009)
[9] Li, X.D., Wiberg, N.E.: Structural dynamic analysis by a time-discontinuous Galerkin
finite element method. Internat. J. Numer. Methods Engrg. 39(12), 2131–2152 (1996)
[10] Duarte, M.L.M.: Experimentally derived structural models for use in further dynamic
analysis, Dissertation, University of London (1996)
[11] Ewins, D.J.: Modal testing, Theory, practice and application. Research Studies Press,
Baldock (2000)
[12] Allen, M.S., Mayes, R.L., Bergman, E.J.: Experimental modal substructuring to
couple and uncouple substructures with flexible fixtures and multi-point connections.
Journal of Sound and Vibration 329(23), 4891–4906 (2010)
[13] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. CIRP An-
nals - Manufacturing Technology 53(2), 619–642 (2004)
[14] Park, S.S., Altintas, Y., Movahhedy, M.: Receptance coupling for end mills. Interna-
tional Journal of Machine Tools and Manufacture 43(9), 889–896 (2003)
[15] Filiz, S., Cheng, C.H., Powell, K.B., Schmitz, T.L., Ozdoganlar, O.B.: An improved
tool-holder model for RCSA tool-point frequency response prediction. Precision En-
gineering 33(1), 26–36 (2009)
308 B. Denkena et al.
[16] Schmitz, T.L.: Torsional and axial frequency response prediction by RCSA. Precision
Engineering 34(2), 345–356 (2010)
[17] Özsahin, O., Ertürk, A., Özgüven, H.N., Budak, E.: A closed-form approach for iden-
tification of dynamical contact parameters in spindle-holder-tool assemblies. Interna-
tional Journal of Machine Tools and Manufacture 49(1), 25–35 (2009)
[18] Ertürk, A., Özgüven, H.N., Budak, E.: Analytical modeling of spindle-tool dynamics
on machine tools using Timoshenko beam model and receptance coupling for the
prediction of tool point FRF. International Journal of Machine Tools and Manufac-
ture 46(15), 1901–1912 (2006)
[19] Ahmadian, H., Nourmohammadi, M.: Tool point dynamics prediction by a three-
component model utilizing distributed joint interfaces. International Journal of Ma-
chine Tools and Manufacture 50(11), 998–1005 (2010)
[20] Kolar, P., Sulitka, M., Janota, M.: Simulation of dynamic properties of a spindle and
tool system coupled with a machine tool frame. The International Journal of Ad-
vanced Manufacturing Technology, 1–10 (2010)
[21] Clausen, M.: Zerspankraftprognose und –simulation für Dreh- und Fräsprozesse,
Hannover (2005)
[22] Denkena, B., Tracht, K., Schmidt, C.: A flexible force model for predicting cutting
forces in end milling. Production Engineering – Research and Development 13(2),
15–20 (2006)
[23] Denkena, B., Schmidt, C.: Experimental Investigation and Simulation of Machining
Thin-Walled Workpieces. Production Engineering – Research and Development 1(4),
343–350 (2007)
[24] Altintas, Y.: Manufacturing automation. Metal cutting mechanics, machine tool vi-
brations, and CNC design. Cambridge Univ. Press, Cambridge (2000)
Chapter 14
Dynamic and Thermal Interactions
in Metal Cutting
14.1 Introduction
For decades, there has been a steady increase of the quality of machine tools. For
further improvement, a better understanding of the interactions between the ma-
chine tool and the physics of the manufacturing process is necessary. A mapping
of the process to an appropriate physical model is often challenging from a com-
putational mechanics point of view as there are often large deformations as well as
a steady evolution of boundaries and new surfaces. For a mesh-based approach,
this may prove to be awkward due to constant remeshing and a mapping of quanti-
ties between successive meshes. Alternatively, a meshless method like the DEM
naturally includes the separation of material and large deformations. A conven-
tional DEM model consisting of rigid spherical particles for granular material is
enhanced by cohesive visco-elastic-plastic interactions. This generates a ‘granular
solid’, which is subjected to virtual material tests to verify it and to determine pa-
rameter values. The mechanical state of the particles is extended to include ther-
mal properties and heat conduction, too. Thereby, an established workpiece model
is machined in orthogonal cutting processes. The obtained results are compared to
experimentally-obtained results and FEM references. The experiments are made
on a specifically designed test bed, which allows the measurement of forces on the
tool during cutting. To include different cases of cutting, the stiffness of the test
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 309–328.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
310 P. Eberhard et al.
bed as well as the angle of inclination of the tool can be modified so oblique cut-
ting processes can also be considered.
To examine the stiffness of the test bed different measurements are performed
in feed direction (x-axis) and in the direction of depth setting (z-axis). A stiffness
of approximately 490 N/µm is determined in z-direction for the table and tool
14 Dynamic and Thermal Interactions in Metal Cutting 311
a) b)
Fig. 14.2 Stiffness element and its static stiffness
312 P. Eberhard et al.
Fig. 14.3 Performing a free three-dimensional cutting process on the test bed
The investigations of the resultant forces in both orthogonal and free three-
dimensional cutting represent a data basis for the verification of cutting models.
Figure 14.4 shows the course of the cutting and thrust force in the orthogonal ma-
chining of steel C45 with different cutting speeds from 50 to 150 m/min at differ-
ent cutting depths. As expected, the dependencies show a significant rise of the
forces with increasing cutting depths. The increase of the cutting speed, however,
has only little effect on the resultant forces.
In three-dimensional cutting, the components of the resultant force, i. e. the cut-
ting force Fx, the shear force Fy and the thrust force, increase, as expected, with in-
creasing cutting depth, see Figure 14.5. The size of the cutting force, in contrast,
hardly changes with an increasing angle of inclination λS. The course of the thrust
force flattens with a greater angle of inclination and the same maximum values.
The shear force Fy increases with an increasing angle of inclination. This reaction
is intensified with increasing cutting depths ap. The increase of the shear force
with an increasing angle of inclination can be explained with the extension of the
cutting length or width.
3500 1600
N 100 m/min
N
Cutting force
Thrust force
150 m/min
2000
800
0 0
0 0,1 0,2 0,3 0,4 mm 0,6 0 0,1 0,2 0,3 0,4 mm 0,6
Depth of cut Depth of cut
2500
Cutting force
1500
1000
500 S=15°
S=45°
0
0 0,1 0,2 mm 0,4
Depth of cut
a)
800
S=15°
N S=45°
Shear force
400
200
0
0 0,1 0,2 mm 0,4
Depth of cut
b)
800
N
Thrust force
400
200 S=15°
S=45°
0
0 0,1 0,2 mm 0,4
Depth of cut
c)
Fig. 14.5 Dependency of the resultant forces on the angle of inclination λS
The shape of the signal and the amplitude during cutting correspond to the posi-
tion of the temperature sensor and to the constantan thigh in the different layers of
the material respectively. In practice, the method with the welded constantan wire
[8] can only be applied in comparably large cutting depths so as to realize a rea-
sonable resolution and to be able to precisely determine the position of the con-
stantan thigh. Therefore, this method was applied in experimental investigations to
determine the temperature in the primary and tertiary cutting zone and in the base
material. The method on the basis of the constantan foil clamped in-between two
inserts does not exhibit such a restriction and was hence used as the basic method
for recording the cutting temperatures in the secondary cutting zone. The change
in temperature in the secondary cutting zone or in the chip, depending on the
14 Dynamic and Thermal Interactions in Metal Cutting 315
Fig. 14.6 Characteristic course of the signal in the primary cutting zone and in the chip
cutting speed and the distance L of the measuring point and on the position of the
constantan foil relative to the cutting edge of the die plate, is shown in Figure 14.7
for a free three-dimensional cutting process. The figure shows that the change in
temperature at a cutting speed of 50 m/min is extreme. The maximum is reached at
a distance of 1 mm between the measuring point and the cutting edge. This corres-
ponds to the known temperature distribution in the secondary cutting zone. At cut-
ting speeds of 100 m/min and 200 m/min, a temperature maximum is only reached
at greater distances to the cutting edge. The results of the FEM cutting model are
summarized in Table 14.1 for a distance of 0.25 mm to the cutting edge. The max-
imum difference between the simulated and experimentally recorded results
amounts to 13.8 %, which can be regarded as an acceptable error in thermal
investigations.
800 800
°C
Temperature
°C
Temperature
400 400
L=0,25
200 50 m/min 200 L=0,50
100 m/min L=0,70
L=1,00
0 200 m/min 0 L=1,25
Table 14.1 Simulated and experimentally recorded temperatures in the secondary cutting
zone
14.3.1 FEM
The FEM model was developed with the commercial product LS-DYNA for the
orthogonal and three-dimensional cutting process [9], [10]. In setting up the model
it is assumed that the conditions along the cutting edge are the same. A boundary
condition for the workpiece is that the lower workpiece edge is to be fixed in both
directions. Regarding the tool it is assumed that the edges averted from the work-
piece are fixed in the direction of depth setting. The movement of the tool is pre-
determined by a rheonomic constraint in feed direction.
The algorithm for the plastic deformation comprises the following calculation
methods:
• Initialization (initial conditions, boundary conditions, maximum inte-
gration time step size and step size for remeshing)
• Organizing the time cycle
• Calculating the clamping condition
• Evaluating the fraction criterion (Johnson-Cook criterion, geometrical
criterion and criterion of maximum tensile stress. If one criterion is
fulfilled for one element, the element is removed.)
• Controlling the step size to renew the FE mesh
f, Кa Fx, N Fz, N
mm/rev Exp. FEM Err. % Exp. FEM Err. % Exp. FEM Err. %
0.30 2.2 1.9 -14 875±92 832 4.9 430±55 106 75.3
The DEM models the motion of usually free and rigid particles, which interact
with each other, and the system boundaries in case of physical contact. A DEM
algorithm can be sub-divided into three major steps, which are neighborhood
search, force computation and integration of the equations of motion. The neigh-
borhood search delivers the information as to which particles are in physical con-
tact and can therefore interact. Based on this information, the forces acting on the
particles are evaluated. Usually, a force-displacement approach is employed,
which calculates penalty forces based on negligible overlaps. The last step is to in-
tegrate the equations of motion.
Neighborhood search algorithms are not explained here. For information about
those and more detailed information about the method in general refer to [11].
The translational dynamics of a free particle i with position vector ri is readily
described using Newton’s equation, see [12],
miri = f i (1)
with the vector of external forces fi. If rotational dynamics also has to be consi-
dered, Euler’s equation of rigid body motion can be used, which reads
Ii ⋅ ω
i + ωi × I i ⋅ ωi = t i (2)
where Ii is the particle’s inertia tensor, ωi is its rotational velocity and ti the vec-
tor of external torques. The differential equations (1) and (2) can be solved with
several numerical integration schemes. The model setup with dynamically moving
particles and evolving boundaries in conjunction with impacts indicates an inte-
grator, which is not sensitive to jumps in the right hand side of the equation. Fur-
thermore, it should be possible to implement it in an implicit manner for stability
reasons as in a general case the frequencies of the system are not known in ad-
vance for all times. For these reasons, an implicit Newmark scheme is chosen. See
[11] for the details.
The right hand side of equations (1) and (2) contains force laws, which are eva-
luated based on force-displacement relationships. To generate a repulsive poten-
tial, which is used, if two separate bodies collide, either linear springs are used or,
limited to spherical particles, a Hertz contact force, see [13]. To define a realistic
material model of engineering materials, however, cohesive forces are needed.
Depending on particle shape, the physics of the problem and other factors, differ-
ent approaches can be employed. In this approach, spheres are used as the only
particle shape as the contact geometry is then determined fast and computational-
ly-efficient, thus allowing for a finer discretization. The chosen approach is to
connect adjacent spheres with massless, visco-elastic-plastic rods in their centro-
ids, which generates what will be called a ‘granular solid’.
The visco-elastic-plastic rods are represented by a force law, which has been
developed based on a piece-wise linear hardening model in [14]. Subjecting a duc-
14 Dynamic and Thermal Interactions in Metal Cutting 319
σ σ g()
σy
σ0.2 σ0.2
h()
σ0
0 y
σy2
The state of stress σij in a rod connecting two particles i and j results in a force act-
ing on the particles it connects. The elastic plastic force fi,ep acting on particle i is
calculated by
f i, ep = σ ij Aij (4)
j
where Aij is the cross sectional area of the rod. It is calculated as arithmetic mean
of the radius of both particles involved. As each real material sustains loading only
to a certain limit, the rod connection is removed, if
σ ≥ σ max (5)
where σmax is the maximum bearable stress of the rod connection. This can also be
expressed as maximum bearable strain εmax.
The cohesive force interaction is applied to all particles, which mechanically inte-
ract in the initial configuration. The procedure to generate an initial configuration
320 P. Eberhard et al.
hammer
55mm specimen
10mm
2mm
counter bearing
A Charpy impact test is simulated with a regular specimen, i. e. the particles are
arranged in an fcc lattice and an inflated specimen with parameters as given in Ta-
ble 14.3.
The amount of absorbed energy for both types of specimen is shown in Fig-
ures 14.13 and 14.14. Increasing the maximum strain εmax before rupture for a rod
increases the amount of energy it can absorb. This can be seen in Figure 14.13,
where εmax is chosen to be 0.07, 0.08 and 0.09. The best agreement with the expe-
rimental value of 6.99 J is found for εmax = 0.08. In a similar manner, the absorbed
work is determined for a random specimen as seen in Figure 14.14. Here, the best
agreement is found for εmax = 0.09. An alternative setup of the test is given in [18].
0.07 0.08
8 8
0.08 0.09
0.09 0.10
6 experiment 6 experiment
energy [J]
energy [J]
4 4
2 2
0 0
0 0.01 0.02 0.03 0 0.01 0.02 0.03
displacement [m] displacement [m]
Fig. 14.13 Absorbed energy for regular Fig. 14.14 Absorbed energy for inflated
specimen specimen
Apart from reproducing a cutting process on a kinematic and kinetic level, thermal
aspects have to be considered, too. Analogously to the mechanical state and
322 P. Eberhard et al.
interactions, a thermal state and thermal interactions are defined. To each particle i
with mass mi and heat capacity ci a temperature Ti is assigned, whose change is
governed by
ΔQi = ci mi ΔTi (6)
where Qi is the internal energy. A finite change of internal energy ΔQi = ΔQj is in-
duced, if two particles i and j interact mechanically, i. e. are in contact. The ex-
change of energy is governed by Fourier’s law, see [19],
λAij
Q =
δ ij
(
Ti − T j ,) (7)
and depends on the heat conductivity λ of the material, the distance of the particles
δij and the cross-section Aij, which the heat flows through. For a definition of the
flux surface, see [23].
For the orthogonal cutting model, a setup as depicted in Figure 14.15 is chosen.
The tool has no dynamics of its own. Constraints are used for both feed and depth
of cut h. The workpiece has length l and height H.
h
l
z
H
x
To assess the quality of the DEM model, experimentally obtained results for both
cutting forces and temperatures generated during cutting are compared with the
numerical results. For the temperature interaction it is assumed that plastic defor-
mation is dissipated with a Taylor-Quinney coefficient β = 0.4 for aluminum and
β = 0.9 for steel. The parameters of the interaction are chosen as
Erod = 1.8·1011 N/m2, εmax = 0.16, ε0.2 = 0.003, k = 0.01·Erod, and r = 0.04 mm. The
experimental and simulated cutting forces are compared in Figure 14.17 for dif-
ferent depths of cut h ranging from h = 0.2 mm to h = 0.4 mm.
0 250
0.4
0.3 200
temperature [K]
−100
0.2 0.43
force [N]
150
−200 0.36
100 0.23
−300 0.17
50
−400 0
0 1 2 3 0 1 2 3 4
displacement [mm] time [ms]
Fig. 14.17 Cutting force for Aluminum Fig. 14.18 Temperature increase for
AlMgSi1.0 h = 0.2 mm
324 P. Eberhard et al.
The experimental results are shown as constant levels with dashed lines and
with the color-coding indicating the depth of cut. It can be seen that for
h = 0.2 mm there is good agreement in the steady state regime. For larger depths of
cut, agreement is slightly worse but still good. The temperature increases generat-
ed during cutting are shown in Figure 14.18 for different vertical positions meas-
ured from the top of the workpiece indicated in the legend in [mm] for
h = 0.2 mm. In the chip at a depth of 0.17 mm, a rise of 200 K can be reported.
This fits quite well to the results reported by [20]. For Ck45, the forces of the
steady state regime are shown in Figure 14.19. The simulated cutting force agrees
well with the experimental results, which are not available for depths of cut great-
er than h = 0.3 mm. The evolution of the simulated passive force, which decreases
and increases again with increasing depth of cut for 0.1 mm < h < 0.3 mm, is due
to discretization effects. The decrease for h ≥ 0.3 mm is due to the chip pressing
against the rake face of the tool. The temperature increase in the workpiece for
h = 0.5 mm is shown in Figure 14.20 with the vertical position of measurement in-
dicated in the legend in [mm], assuming a Taylor-Quinney coefficient of β = 0.9.
The measured temperature increase for a depth of cut of 0.3 mm is ΔT = 340 K.
This is significantly lower than the simulated temperature increase in the chip,
which is shown for the vertical positions 0.04 mm and 0.43 mm and is found to be
around ΔT = 450 K. An explanation for this behavior is a too high value for β and
too much compressive plastic deformation in the chip. For vertical positions dee-
per within the workpiece, far better agreement can be seen, e. g. at 0.7 mm and
1.0 mm. The temperature curve for 0.56 mm illustrates how the temperature peaks,
when the tool passes, but the heat quickly distributes in the rest of the workpiece.
See [21, [22], [23]for further results.
cut sim
800 passive sim
cut exp
600 passive exp
force [N]
400
200
0
0.1 0.2 0.3 0.4 0.5
depth of cut [mm]
700
0.04 sim.
600 0.43 sim.
500 0.56 sim.
temperature [K]
0.69 sim.
400
1.0 sim
300 0.30 exp.
200 0.70 exp.
1.0 exp.
100
0
0 2 4 6 8
time [ms]
An illustration of the cut off chip is shown in Figure 14.21. The chip is cohe-
sive and clearly separated from the rest of the workpiece.
rotational movement of the overall tool. The tool is coupled with a rotational
spring-damper combination. The dynamics of the tool is integrated into a separate
simulator, a program to model multi-body systems. Both particle and multi-body
simulator are coupled in a co-simulation approach, where state variables as well as
kinetic quantities are exchanged at fixed time steps. See [18] for a more detailed
explanation.
z Ct
mt I t we
x
be
0 150
mm]
mm]
−50 100
−2
−2
force [N] / Δ z [10
−100
50
Fig. 14.23 Cutting force for elastic tool Fig. 14.24 Passive Force for elastic tool
support support
The model parameters are unchanged, except Erod = 5.9 1010 N/m2. The para-
meters determining where the tool is fixed are chosen as we = 0.02 m and
be = 0.04 m, i. e. both passive and cutting force bring about a torque on the tool.
The cutting force for a normalized workpiece width of 1 mm and the depth of cut
are both shown in Figure 14.23, indicated in the legend by ‘cut force’ for the force
and ‘Δztool’ for the change of vertical position of the tool tip. It can be seen that
starting from an initial depth of cut of 0.4 mm the cutting depth continuously in-
creases with continuous tool feed. A similar behaviour can be seen in Figure 14.24
14 Dynamic and Thermal Interactions in Metal Cutting 327
for the passive force. A steep rise in passive force as the tool enters the workpiece
is followed by a continuous increase as the cutting depth increases. The tool tip
moves deeper and deeper into the workpiece due to the moment acting on the tool.
As the length of the workpiece is chosen as l = 4 mm both passive and cutting
force vanish for a tool displacement exceeding 4 mm.
14.4 Conclusion
This contribution illustrates that the proposed material model based on the Dis-
crete Element Method allows to capture essential characteristics of elasto-plastic
solids. The separation of material as shown for a Charpy impact test and in cutting
processes is included without effort. The comparison of simulated and experimen-
tally obtained cutting forces demonstrates that good agreement can be reached,
both on a qualitative as well as a quantitative level. An extension of the model
such that the thermodynamics of heat conduction can be included has been suc-
cessfully implemented. Comparing the temperatures generated during cutting
shows that, for an appropriate choice of the Taylor-Quinney coefficient, good
agreement between simulation and experiment can be reached. A further en-
hancement of the cutting process model, such that dynamics of the tool can be in-
cluded, is made based on a co-simulation approach. There, the tool dynamics is
accounted for in a separate multi-body simulator. The entirely different and physi-
cally sound behavior of the system once the constraints are removed shows the in-
creased applicability of the approach for modeling the interaction between tool,
workpiece and machine.
References
[1] Tsai, L.-W.: Robot Analysis – The Mechanics of Serial and Parallel Manipulators.
John Wiley & Sons, New York (1999)
[2] Merlet, J.-P.: Parallel Robots. Kluwer Academic Publishers, Dordrecht (2000)
[3] Preumont, A.: Mechatronics – Dynamics of Electromechanical and Piezoelectric Sys-
tems. Springer, Dordrecht (2006)
[4] Heisel, U., Storchak, M., Stehle, T.: Einfluss der Umgebungstemperatur beim ortho-
gonalen Zerspanen. wt Werkstattstechnik Online 100, 89–98 (2010) (in German)
[5] Frohmüller, R., Knoche, H.-J., Lierath, F.: Aufbau und Erprobung von Temperatur-
messeinrichtungen durch das IFQ im Rahmen des Schwerpunktprogramms Spanen
metallischer Werkstoffe mit hoher Geschwindigkeit. Spanen metallischer Werkstoffe
mit hohen Geschwindigkeiten Kolloquium des Schwerpunktprogramms der DFG,
108–115 (1999) (in German)
[6] Körtvelyessy, L.V.: Thermoelement Praxis. Vulkan Verlag, Essen (1981) (in Ger-
man)
[7] Müller, B.: Thermische Analyse des Zerspanens metallischer Werkstoffe bei hohen
Schnittgeschwindigkeiten. Dissertation, RWTH Aachen (2004) (in German)
[8] Heisel, U., Storchak, M., Stehle, T., Korotkih, M.: Temperaturbestimmung in den
Zerspanzonen. wt Werkstattstechnik Online 100, 365–370 (2010) (in German)
328 P. Eberhard et al.
[9] Heisel, U., Krivoruchko, D.V., Zaloha, V.A., Storchak, M.: Cause Analysis of Errors
in FE Prediction of Orthogonal Cutting Performances. In: 10th CIRP International
Workshop on Modeling of Machining Operations, pp. 141–148 (2007)
[10] Heisel, U., Kryvoruchko, D.V., Zaloha, V.A., Storchak, M., Emelyanenko, S., Seli-
vonenko, S.N.: Finite Element Analysis of Cutting Force Dynamics. In: Proceedings
of the 11th CIRP Conference on Modeling of Machining Operations, September 16-
18, pp. 163–170 (2008)
[11] Fleissner, F.: Parallel Object Oriented Simulation with Lagrangian Particle Methods.
Schriften aus dem Institut für Technische und Numerische Mechanik der Universität
Stuttgart, Bd. 16. Shaker Verlag, Aachen (2010)
[12] Geradin, M., Rixen, D.: Mechanical Vibrations. John Wiley & Sons, Chichester
(1998)
[13] Lankarani, H.M., Nikravesh, P.E.: A Contact Force Model with Hysteresis Damping
for Impact Analysis of Multibody Systems. Journal of Mechanical Design 112, 369–
376 (1990)
[14] Liu, K., Gao, L., Tanimura, S.: Application of Discrete Element Method in Impact
Problems. JSME International Journal, Series A 47, 138–145 (2004)
[15] Fleissner, F., Gaugele, T., Eberhard, P.: Application of the Discrete Element Method
in Mechanical Engineering. Multibody System Dynamics 18, 81–94 (2007)
[16] Gaugele, T., Fleissner, F., Eberhard, P.: Simulation of Material Tests using Meshfree
Langrangian Particle Methods. Proceedings of the Institution of Mechanical Engi-
neers, Part K: Journal of Multi-body Dynamics 222(K4), 327–338 (2008)
[17] DIN10045, Kerbschlagbiegeversuch nach Charpy. Deutsches Institut für Normung
e.V., Berlin (1991)
[18] Gaugele, T., Eberhard, P., Storchak, M., Heisel, U.: A Discrete Material Model used
in a Co-simulated Charpy Impact Test and for Heat Transfer. In: Proceedings of the 1.
International Conference on Process Machine Interactions, pp. 361–369 (2008)
[19] Tychonoff, A., Samarski, A.: Differentialgleichungen der mathematischen Physik.
Deutscher Verlag der Wissenschaften, Berlin (1959) (in German)
[20] Jaspers, S., Dautzenberg, J.: Material Behavior in Metal Cutting: Strains, Strain
Rates, and Temperatures in Metal Cutting. Journal of Materials Processing Technolo-
gy 121, 123–135 (2002)
[21] Gaugele, T.: Application of the Discrete Element Method to Model Ductile, Heat
Conductive Materials. Dissertation, University of Stuttgart (2011) (submitted)
[22] Eberhard, P., Gaugele, T.: Quasi-static and dynamic properties of separable continua
based on Lagrangian Particle Methods. In: Proceedings of the International Confe-
rence on Particle-Based Methods, Fundamentals and Applications (Particles 2009),
pp. 398–401 (2009)
[23] Gaugele, T., Eberhard, P., Fleissner, F.: Particle Methods Used to Model Cutting
Processes Including Heat Conduction. In: Proceedings of the International Confe-
rence on Particle-Based Methods, Fundamentals and Applications (Particles 2009),
pp. 142–145 (2009)
Chapter 15
Surface Generation Process with Consideration
of the Balancing State in Diamond Machining
15.1 Introduction
Ultraprecision diamond machining is mainly used to manufacture optical parts or
microstructures with form deviations of less then one micron and surface rough-
ness in the range of a few nanometers. To obtain these accuracies the requirements
regarding machine tool accuracy, process control and environmental conditions are
exceptional.
But even if all these requirements are met, unintended vibrations either induced
by the environment or by the process itself can result in insufficient part quality. Usu-
ally external vibrations can be kept low by appropriate foundations of the machine
tool. In contrast, process induced vibrations can occur because of misaligned tools
or workpieces, unsymmetrical workpieces, unsymmetrical workpiece mounting or
inhomogeneous workpiece materials. All these effects lead to non-homogeneous
mass distribution for the rotational axis of the system and therefore to vibrations
when under rotation. If existent these process induced vibrations can lead to an
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 329–360.
springerlink.com c Springer-Verlag Berlin Heidelberg 2013
330 C. Brandt et al.
increased form deviation and an increased surface roughness depending on the ex-
citing frequency. Therefore, precision balancing is crucial for ultraprecision ma-
chining processes [7]. In this chapter only ultraprecision turning will be considered,
nevertheless basic balancing principles will apply for other processes using rotating
spindles as well. The balancing process in ultraprecision machining is done man-
ually and it generally takes more than one iteration step to get the best balancing
condition.
The goal of the project is to get a deeper understanding of the interaction be-
tween balancing and the surface generation process in ultraprecision machining.
Therefore, a process-machine model has been developed which allows to predict the
surface topography based on the balancing state considering the interaction between
the machine tool structure and the machining process. In a first step, we developed
an interaction model for the experimental platform, see Section 15.3. A structure
model was developed to determine the vibrations for given force and moment dis-
tributions induced by unbalances as well as forces from the cutting process. This
model can also be utilized to determine unbalance distributions or balancing weights
from vibration data measured by sensors during idle spindle speed. Additionally, an
analytical process model for describing the cutting forces during the cutting pro-
cess has been developed. The cutting forces were related to the surface structure of
the workpiece. Since the cutting forces influence the vibrations of the machine and
vice versa, both sub-models have to be merged into a mechanical-dynamical model.
The complete model will enable us to determine the effective cutting forces and
displacements related to the workpiece surface. This model also allows to compute
the surface quality and in the future a necessary balancing state to achieve a needed
surface quality.
will move back in cutting direction and towards the surface as well. Depending on
the occurring forces, the damping properties of the tool and the unbalance, the tool
will either start to vibrate itself or it will reach a state of equilibrium.
Due to the interaction between the workpiece and the tool the surface of the
workpiece will be distorted. For an unbalanced and therefore vibrating workpiece
the engaged tool will act as a damper and reduce the induced vibrations. Due to this
process machine interaction it is necessary to take the process forces into account
with regard to the unbalance state and the resulting surface topography . But, be-
cause of size effects process models applied in conventional machining cannot be
applied. Previous investigations showed that in diamond turning the forces increase
for higher cutting velocities [11] whereas in conventional machining the forces de-
crease at higher cutting velocities because of heating and softening of the workpiece
material [14].
To assess the process forces during ultraprecision machining measurements have
been carried out using a triaxial dynamometer for low forces. The dynamometer is
mounted beneath the tool holder, on top of the test stands z-axis.
Fig. 15.7 Balancing experiment platform (test stand), cross sectional view of the main spin-
dle (courtesy: Kugler) [5]
First, we have divided the platform into components which are illustrated in Fig.
15.8, namely the rotating part of the spindle, the spindle casing, the rotating part
of the motor, and the motor casing. The spindle rotor and casing are connected by
an air bearing with two spherical calottes. We have modelled this bearing by two
spring-damper elements although in a first attempt any damping in the springs was
neglected. The motor bearings are also modelled as spring-damper elements. Spin-
dle and motor are connected by a coupling that can compensate misalignments in
axial and radial directions as well as torsion. It is also modelled as a spring element.
Spindle and motor are supported by a granite base that is assumed to be rigid. The
joints to the granite base are modelled as firm spring elements. Additionally, Figure
15.8 shows the coordinate system. In operation, the spindle rotates counterclockwise
around the z-axis.
Secondly, we have developed a vibration model for each part of the machine sep-
arately. If we would consider only unbalances as possible reasons for vibrations it
would be sufficient to allow vibrations in radial directions x and y only. Since unbal-
ances cause harmonic vibrations and we assume isotropic bearings, the vibrations
in x and y direction are the same except for a phase shift of π /2. Nevertheless, the
forces from the cutting process act in all three directions. Thus each point along the
z-axis, i.e. each beam element of infinitesimal length ∂ z, has the following degrees
of freedom (DOF): the displacement u, v, w in x, y, z direction, the torsion angle βz ,
and the rotational angles βx , βy . We collect the DOF in a vector u = u(z,t). The
computation of u is based on an energy formulation, the so called Principle of Vir-
tual Displacements, see [8], Chapter 5. This principle is equivalent to equilibrium
conditions from which a partial differential equation for u(x,t) can be derived. Start-
ing from the energy formulation, we use the Finite Element Method (FEM) for the
discretization in the space variable x. We arrive at a system of ordinary differential
equations (ODE) in time of the form
S for each part of the platform. This procedure is well known and we have mainly
followed [8] in order to derive the matrices. We have used the following partition:
1. Spindle rotor with the coupling: 36 elements.
2. Spindle casing: 30 elements.
3. Rotor part of the motor: 3 elements.
4. Motor casing: 2 elements.
we get a block diagonal structure for the entire mass and stiffness matrix of the
dimension 450 × 450.
Additionally, we have to consider the bearing elements that connect the DOF
of the corresponding nodes in the connected parts via a stiffness parameter C and
damping parameters D. This results in additional diagonal and in off-diagonal ele-
ments of the entire stiffness matrix which are collected in sparse matrices Cair for
the air bearing, Cm for the motor, and Cc for the coupling of motor and spindle:
⎛ ⎞ ⎛ ⎞
Ssp−rot 0 0 0 Cair −Cair 0 0
⎜ 0 Ssp−cas 0 0 ⎟ ⎜ 0 ⎟
S=⎜ ⎟ + ⎜ −Cair Cair + Cc −Cc ⎟.
⎝ 0 0 Sm−rot 0 ⎠ ⎝ 0 −Cc Cc + Cm −Cm ⎠
0 0 0 Sm−cas 0 0 −Cm Cm
(15.3)
The damping matrix D will have the same structure as the additional sparse ma-
trices in the stiffness matrix. Stiffness values for the coupling between motor and
spindle were provided by the manufacturer. For the air bearing between spindle
and spindle casing we had to rely on an inspection record that stated measurement
338 C. Brandt et al.
values for the axial and radial stiffness of the bearing in the steady state. We mention
again, that we have neglected the damping in a first attempt mainly because of the
lack of information. The identification of damping coefficients will be part of future
investigations.
F = ω 2 b, with b := Δ mr.
where ℑ and ℜ denote the imaginary and real part of a complex number. Those
forces only apply to the displacement DOF in x-direction, and y-direction. All the
other DOF are not affected. Therefore, the sub-vector pk of p containing the entries
for the DOF of the k-th node, k =, · · · , N, has the form
⎛ ⎞
ℑ(ω 2 bk eiφk eiω t )
⎜ ℜ(ω 2 bk eiφk eiω t ) ⎟
⎜ ⎟
⎜ 0 ⎟
⎜
pk = ⎜ ⎟
0 ⎟
⎜ ⎟
⎝ 0 ⎠
0
We split punb (t) = (pk )k into part with sin and cos entries only in order to apply the
right hand side ansatz for each of the parts to solve (15.1):
and the mentioned conventional models are not applicable to ultra-precision turning
experiments. Nevertheless, the forces in the experiments show a typical behavior
for micro machining and the used workpiece material (aluminium alloy AlMg3),
i.e. the thrust force is the dominant force component and the forces are in the range
of one Newton or less.
Another class of force models for micro turning are modified Kienzle models,
see [14]. In [10] an ansatz is presented to calculate an undeformed chip thickness h
which permits to consider tools with different tool nose geometries. Another modi-
fied Kienzle ansatz is proposed in [17] where the specific cutting force kc is repre-
sented as a product of functions including cutting velocity vc , friction μ , uncut chip
thickness h and cutting edge radius rβ , i.e.
β −β
with functions f1 (h) = ch−m and f2 = α1 vc 1 + α2 vc 2 . We picked up this idea and
the form of the functions f1 and f2 to model the specific cutting force as product
of functions gi depending on the depth of cut a p , the feed rate f and the cutting
velocity vc , i.e. to model the specific force in the form
Similar equations hold for the specific thrust and feed forces k p and k f . The func-
tions gi (i = 1, 2, 3) and the model constants therein are determined with help of
force measurements with different cutting conditions, see Table 15.1 for details.
The measurements of the specific forces over the depth of cut and feed rate are
illustrated in Fig. 15.11 forces together with the fitted curves g1 and g2 (solid line)
which had been determined to be of the form
−mia
gi1 (a p ) = cia a p (i = t, f , c) (15.8)
and
gi2 ( f ) = cif f −m f
i
(i = t, f , c) . (15.9)
The function g3 has been determined to be of form of the function f2 , i.e.
βi −β2i
gi3 (vc ) = α1i vc 1 + α2i vc (i = t, f , c) . (15.10)
342 C. Brandt et al.
Fig. 15.11 Measured specific forces over depth of cut (left hand side) and feed (right hand
side). The measured thrust force is represented by circles and the measured cutting force by
x-marks. The fitted curves g1 and g2 given by equation (15.8) and (15.9) are plotted as solid
line [3]
The function g3 and the measurements of the specific forces over cutting velocity are
shown in Fig. 15.12. Using the specific cutting forces we are now able to calculate
the specific cutting force components via the usual relationship for the forces by
Kienzle (see [14]):
Fc = kc Ac , Ft = kt Ac , F f = k f Ac , (15.11)
where Ac denotes the cross sectional area of cut which can be approximated by
Ac = a p f .
where lh denotes the length of the tool holder. The angle βy denotes the rotation of
the workpiece around the y-axis. This tilt of the workpiece can influence the actual
depth of the cut a p which is described in equation (15.14). Since the deflections
affect the actual tool tip position and we now define time dependent process param-
eters, namely the actual depth of cut
and
vx (t) = d(t)
˙ , with d(t) = x(t) + r . (15.16)
The actual feed rate f is given by f (t) = n−1 vx (t). In contrast to the given constant
input parameters of the test stand, the time dependant parameters are called “actual
parameters”.
These actual process parameters are plugged into the equation (15.11) for the
force components using the equations for the specific forces (15.7). Using the actual
forces, we are now able to determine the deflections of the tool holder which are
proportional to the forces
Here kei denotes the corresponding stiffness in the direction i (i = x, y, z). Since
all force components have the same structure we get for all three spatial directions
( j = x, y, z) the deflection
cia a p (t)−ma c̃if f (t)−m f
i
−1
α1i vc (t)β1 + α2i vc (t)−β2 a p (t) f (t)
i i i
δ j (t) = kei
(15.18)
with (i = f , c,t) for ( j = x, y, z). Derivatives of the equations for the position
(15.12)-(15.14) and deflections (15.18) combined with equation (15.11),(15.15) and
(15.16) for the actual parameters deliver a system of ordinary differential equations
of the form
1
ȧ p (t) = − δx (t)δ̇x (t) + δy (t)δ̇y (t) − δ̇z (t) + Δ̇x (t) ,
lh
β̇y (t)
+ (r − d(t)) − vx (t) tan (βy (t)) , (15.23)
cos2 (βy (t))
˙ = vx (t) ,
d(t) (15.24)
f f f f
−m
k̇ f (t) = −caf c̃ f a p (t)−ma vx (t) f α1 β1 vc (t)β1 −1 − α2 β2 vc (t)−β2 −1
f f f f f
f f f f
−ma −1 −m f f −m f −1 −ma
+ ma a p (t)
f
ȧ p (t)vx (t) + m f vx (t) v̇x (t)a p (t)
f f
α1f vc (t)β1 + α2f vc (t)−β2 (15.25)
k̇c (t) = −cca c̃cf a p (t)−ma vx (t)−m f α1c β1c vc (t)β1 −1 − α2c β2c vc (t)−β2 −1
c c c c
+ mca a p (t)−ma −1 ȧ p (t)vx (t)−m f + mcf vx (t)−m f −1 v̇x (t)a p (t)−ma
c c c c
α1c vc (t)β1 + α2c vc (t)−β2 ,
c c
(15.26)
k̇t (t) = −cta c̃tf a p (t)−ma vx (t)−m f α1t β1t vc (t)β1 −1 − α2t β2t vc (t)−β2 −1
t t t t
mta a p (t)−ma −1 ȧ p (t)vx (t)−m f + mtf vx (t)−m f −1 v̇x (t)a p (t)−ma
t t t t
α1t vc (t)β1 + α2t vc (t)−β2 .
t t
(15.27)
i
Hereby, we defined c̃if = cif nm f , because of the relation vx (t) = n f (t) and equation
(15.9). The cutting velocity is calculated by vc (t) = 2π n (r − d(t)) and the vibrations
Δi , i ∈ {x, y, z}, as well as the tilt angle βy of the workpiece and their derivatives are
determined by the structural submodel. The system of differential equations will
be solved numerically with the MATLAB solver ”ode15i”. The stiffness values kei
(i = x, y, z) are determined by the geometrical dimensions of the tool holder and its
elasticity module (material: steel, E = 210 kN/mm2). The resulting actual forces Fi
are calculated by the deflections δ using equation (15.17). The numerical results are
presented in Sect. 15.4.5.
The process model computes the actual forces and deflections of the tool which act
at the tool tip position on the workpiece. Therefore, these forces add up to the forces
15 Balancing in Diamond Machining 345
of the unbalances in the right hand side of equation (15.1). The resulting additional
load vector has the form
where ra (t) = r − d(t) denotes the radius of the workpiece minus the already trav-
elled distance of the tool on the workpiece, cf. Fig. 15.10. Thus, a coupled system
of ODEs
Mü(t) + Su(t) = punb + pcut (t, F(t, w), Mt (t, w), Mc (t, w)) ,
ẇ(t) = g (t, w(t), ẇ(t), u(t)) ,
has to be solved, where the function g is given by the system of differential equa-
tions (15.19)-(15.27) and the corresponding variables are collected in the vector
w = (δx , δy , δz , vy , a p , d, kt , k f , kc ). In order to solve the non-linear coupled system
approximately, we have employed a time step algorithm and assumed the forces and
moments from the cutting process to be constant during a small time interval, i.e.
pcut (t) = pcut (ti ) for t ∈ [ti ,ti + Δ t]. Now,
where k2π denotes the section’s angle. An ideal radial section S0 is shown in Fig.
15.13. The surface profile will change in two ways with respect to vibrations of the
workpiece and deflections of the tool. Firstly, the repetitions will not be necessarily
equidistant and the turning grooves will differ in their depth. Secondly, the geome-
try of each channel will change because tool and workpiece will not be positioned
orthogonal to each other. Figure 15.13 shows also a simulation result for a disturbed
process.
In the following, a discrete surface generation model, which is coupled with the
process model, is presented to provide the surface function S depending on vibra-
tions of the workpiece and tool deflection. The central component of the model is
the parameter to state operator
The characteristic motions in turning processes are the rotation of the workpiece
around the z-axis, which is expressed by
⎛ ⎞
cos(φ ) − sin(φ ) 0 0
⎜ sin(φ ) cos(φ ) 0 0 ⎟
ΦW = RZ (φ ) = ⎜⎝ 0
⎟,
0 1 0⎠
0 0 01
and the translation of the tool along the x-axis, which is expressed by
⎛ ⎞
1 0 0 r − f Nt
⎜0 1 0 0 ⎟
ΦT = T ((r − f Nt, 0, lh − a p)) = ⎜ ⎟
⎝ 0 0 1 lh − a p ⎠ .
000 1
These two operators describe the movement of the tool and of the workpiece under
ideal conditions. The positioning and alignment errors are included into the model
by applying further translations and rotations. Figure 15.14 shows the tilt angles βx
348 C. Brandt et al.
and βy of the workpiece. The moment acting around the x-axis has only a marginal
effect to surface generation, and therefore it is neglected in our model. The rotation
rate is assumed to be constant. This leads to the operator for the movement of the
workpiece:
ΦW = T (Δ )RY (βy )RZ (φ ). (15.31)
The tool inclination is changed appropriate to the deflection provided by the force
model by applying multiple rotations, which leads to
ΦT = T ((r − f Nt, 0, lh − a p))R(δx , δy ).
So far, the movement of the tool is described by ΦT and that one of the workpiece
by ΦW . For the material removal algorithm the relative position of the tool to the
workpiece is compulsory. ΦW and ΦT give absolute positions, but we can carry over
the workpiece operation to the tool using
Φ := ΦW−1 ΦT . (15.32)
The diamond tool, or more precisely its edge geometry, can be parameterized as
ε (ϕ ) = (rε cos ϕ , 0, −rε sin ϕ − l + rε , 1)T , ϕ ∈ [0, π ], (15.33)
with tool radius rε and length lh . Thus, the relative tool trajectories are given by
Φε (ϕ ) for ϕ ∈ [0, π ]. In case of ideal conditions, i.e. no vibration, no displacement
and no deflection, (15.32) can be written as
⎛ ⎞
cos(−φ ) − sin(−φ ) 0 (r − f Nt) cos(−φ )
⎜ sin(−φ ) cos(−φ ) 0 (r − f Nt) sin(−φ ) ⎟
Φ =⎜ ⎝
⎟.
⎠ (15.34)
0 0 1 lh − a p
0 0 0 1
In Fig. 15.15 the ideal and disturbed tool tip locus is visualized. For the ideal case the
trajectories of the tool tip are given by applying the tool tip ε (π/2) = (0, 0, −l, 1)T to
(15.34), leading to (x, y, z, 1)T = ((r − f Nt) cos(−φ ), (r − f Nt) sin(−φ ), −a p , 1)T ,
i.e. a spiral with decreasing radius appropriate to the feed rate located parallel to the
workpiece surface at height −a p .
Fig. 15.15 Relative tool tip locus for ideal conditions (left) and for an oscillating surface
(right); a p = 1mm, f =0.5mm
Fig. 15.16 Deflection of the workpiece in z-direction for f1 and f2 ; entire time interval (left),
and detail (right)
Fig. 15.17 Deflection of the workpiece in y-direction for f1 and f2 and depth of cut
The output of the process machine interaction model can also used to visualize
the machined surface. In a first step, surface simulations for simple oscillations of
type
are carried out in order to verify the basic function of the surface model. It is clear
that frequencies pi n in the signal βy can be recovered in the surface structure because
the workpiece is rotating with angular speed ω . In Fig. 15.19 two example surfaces
for different oscillations are shown.
In a second step, the robustness of surface generation is evaluated with the help
of the implemented surface model. This is an important step, because all input os-
cillations are result of numerical solved differential equations. Figure 15.20 shows
exemplarily a simulated surface for an oscillation βy of (50 + ε )Hz, where ε denotes
a small disturbance. When no disturbance exists (ε = 0) and a rotation frequency of
50Hz is assumed, the resulting surface is an inclined plane. With increasing degree
of disturbance the form error grows rapidly, which is demonstrated in Figure 15.20
for ε = 10−4 , 10−3 , 10−2. For ε = 10−4 a first distortion appears, which is growing
intensively for increasing disturbance. Finally, for ε = 10−2 the oscillation is heavily
asynchronous to the rotation leading to a completely wrong visual impression.
Model problems with reduced workpiece diameter and high feed speeds are con-
sidered for the benefit of short calculation time for all sub-models. The process
parameters of setting 2 in Table 15.2 are used in three model problems with dif-
ferent unbalance configurations, see Table 15.3. The related surfaces are plotted in
352 C. Brandt et al.
Fig. 15.20 Simulated surfaces for an oscillation of (50 + ε )Hz; n = 50Hz, r = 30mm
Fig. 15.21 (a)–(c). In case (a), where no unbalances are set, a wavy surface can be
detected. This corresponds to the power spectrum of βy , where three different fre-
quencies can be identified. In case (b) and (c), where unbalances for the workpiece
or balancer planes are set, only the rotation frequency can be identified in the power
spectrum. The result is an inclined surface.
Fig. 15.22 Simulated surfaces for setting 3, Table 15.2, unbalanced (1.823g at the
workpiece)
Fig. 15.23 Simulated surfaces for setting 3, Table 15.2, best possible balancing state
maps the vector p containing the unbalances distribution and the process parameters
to a surface S, i.e. we have the following operator equation
A(p) = S . (15.36)
This equation is called forward formulation of our model. Assuming now that for a
given surface S, we are interested do determine the necessary balancing state in or-
der to obtain the surface S, i.e. to solve equation (15.36) with respect to p. However,
measurement devices typically have limited precision and we assume that unprecise
and noisy measurement data Sδ are available, which fulfill ||S − Sδ || ≤ δ , where
δ denotes the measurement precision. Usually this operator is not continuously in-
vertible, which means that for given noisy data Sδ with a data error of the function
pδ = A−1 (Sδ ) might be an arbitrary bad approximation of the true unbalance dis-
tribution p. Problems with those properties are referred to as being ill-posed. In this
case, least square techniques, where pδ is computed as the minimizer of Ap−Sδ 2 ,
are unstable. The computation of pδ can be stabilized by using the regularization
methods presented in Chap. 3, i.e. by minimizing the so-called Tikhonov-functional
instead. The penalty term Ψ (p) acts as a stabilizer and prevents large values of
Ψ (p).Typical choices of Ψ are, e.g.
1/p
Ψ (p) = p p := ∑ |pi | p
, 0 < p ≤ 2. (15.38)
i
In a first attempt we invert the structural submodel, i.e. given a vibration u we deter-
mine the unbalance distribution p causing the vibrations. This results are presented
in Chap. 3. Future work is dealing with the inversion of the full forward problem
(15.36).
Fig. 15.24 Form deviations for balanced (left) and unbalanced (right) workpiece [11]
Fig. 15.25 Form deviation vs. depth of cut (with and without unbalance) [11]
Fig. 15.26 Form deviation vs. feed (with and without unbalance) [11]
From Fig. 15.27 and Fig. 15.28 it can be seen that the difference between ma-
chining with and without unbalance are only marginal. The roughness values are
slightly higher for machining with unbalance but for the observed cutting parame-
ters no strong influence can be observed.
Possibly the test stand is working at its limit in terms of accuracy as it has a lower
stiffness compared to current ultraprecision machine tools. Another possibility for
the low influence of unbalances is assumed in the process kinematic, as for turning
the surface normal and the direction of the force generated by the un-balance are
perpendicular to each other. For that reason the workpiece will not be deflected di-
rectly in direction of the tool which would cause a periodical change for the depth of
cut. Therefore, the current investigations will be extended by ultraprecision milling
experiments. For these experiments the normal of the generated surface and the di-
rection of the unbalance induced forces and with that the motion of the tool will be
the same.
15 Balancing in Diamond Machining 357
Fig. 15.27 Surface roughness vs. depth of cut (with and without unbalance) [11]
Fig. 15.28 Surface roughness vs. feed (with and without unbalance)
motion additionally to the rotation. The tumbling motion will tilt the workpiece pe-
riodically with respect to the tool. The tilting will not change the friction between
chip and rake face, but the angle between the workpiece surface and clearance face
will change periodically. For balanced machining this angle will be equivalent to
the clearance angle. For unbalanced machining if the workpiece is tilting towards
the clearance face, this angle will decrease and induce a larger frictional load on the
clearance face. With this additional frictional load the increased thrust force for un-
balanced workpieces can be explained. The cutting force would have been expected
to show a significant rise for unbalanced machining, because of the change in the
depth of cut. But at this stage this phenomenon cannot be explained.
A similar behavior for the force components can be recognized for an increasing
feed, again only the thrust force is influenced by the unbalance (see [11]).
Acknowledgements. The presented research has been funded by the German Research
Foundation DFG within the Priority Program 1180.
360 C. Brandt et al.
References
1. DIN ISO 1940-1 mechanical vibration - balance quality requirements for rotors in a
constant (rigid) state - part 1: Specification and verification of balance tolerances
2. Brandt, C., Niebsch, J., Ramlau, R., Maass, P.: Modeling the influence of unbalances
for ultra-precision cutting processes. ZAMM - Journal of Applied Mathematics and Me-
chanics/Zeitschrift für Angewandte Mathematik und Mechanik 91(10), 795–808 (2011),
doi:10.1002/zamm.201000155
3. Brandt, C., Niebsch, J., Vehmeyer, J.: Modelling of ultra-precision turning process in
consideration of unbalances. In: 13th CIRP Conference on Modeling of Machining Op-
erations (2011)
4. Brinksmeier, E., Gläbe, R., Krause, A.: Precision balancing in ultraprecision diamond
machining. In: Laser Metrology and Machine Performance 8, Lamdamap 2007, vol. 8,
pp. 262–269 (2007)
5. Brinksmeier, E., Krause, A.: Dual plane balancing for diamond machining processes.
In: 3rd International Conference High Performance Cutting (HPC), Dublin, pp. 517–528
(2008)
6. Brinksmeier, E., Krause, A.: Surface generation in ultraprecision diamond machining
utilising dual-plane-balancing. In: International Conference on Process Machine Inter-
actions, Hannover, pp. 335–342 (2008)
7. Brinksmeier, E., Riemer, O.: Deterministic prodcution of complex optical elements. In-
ternational Journal of Production Engineering and Computers - Special Issue on CAPP
and Advances in Cutting Technology 4(5), 63–72 (2002)
8. Gasch, R., Knothe, K.: Strukturdynamik Bd. 2: Kontinua und ihre Diskretisierung.
Springer, Berlin (1989); IX, 336 S: graph. Darst
9. Jin, X., Altintas, Y.: Slip-line field model of micro-cutting process with round
tool edge effect. Journal of Materials Processing Technology (2010) (in press),
doi:10.1016/j.jmatprotec.2010.10.006
10. Köhler, J.: Berechnung der Zerspankräfte bei variierenden Spanungsquerschnittsformen.
Ph.D. thesis, Leibniz Universität Hannover (2010)
11. Krause, B.: Process forces in diamond machining with consideration of unbalances. In:
CIRP PMI, Vancouver (2010)
12. Malekian, M., Park, S., Jun, M.: Investigation of critical chip thickness and micro plough-
ing forces. In: Proceedings of the 2nd International Conference on Process Machine In-
teractions, Vancouver, Canada, June 10-11 (2010)
13. Schneider, H.: Auswuchttechnik. Springer (2007)
14. Tönshoff, H.K., Denkena, B.: Spanen: Grundlagen, 2., erw. und neu bearb. aufl edn.
Springer, Berlin (2004); XXIV, 417 S.: Ill., graph. Darst
15. Vollertsen, F., Biermann, D., Hansen, H., Jawahir, I., Kuzman, K.: Size effects in man-
ufacturing of metallic components. CIRP Annals - Manufacturing Technology 58(2),
566–587 (2009), doi:10.1016/j.cirp.2009.09.002
16. Dong, W.P., Blunt, L.: Three-dimensional surface topography, 2nd edn. Penton Press
(2000), http://www.sciencedirect.com/science/book/
9781857180268, XXII, 285 S
17. Weber, M., Autenrieth, H., Kotschenreuther, J., Gumbsch, P., Schulze, V., Lohe, D.,
Fleischer, J.: Influence of friction and process parameters on the specific cutting force and
surface characteristics in micro cutting. Machining Science and Technology 12, 474–497
(2008), doi:10.1080/10910340802518728
18. Zhou, S., Shi, J.: Active balancing and vibration control of rotating machinery: A survey.
The Shock and Vibration Digest 33, 361–371 (2001)
Chapter 16
Modeling and Simulation-Based Optimization
of a Turning Process
16.1 Introduction
Interactions between machine structures and manufacturing processes, such as vi-
brations due to unsteady cutting, limit the process flow. With a detailed prediction
of these interactions it is possible to optimize operation charts, machine parame-
ters, path parameters and further details. The goal of the SPP1180 project "Para-
metric modeling, prediction and optimization of interaction between machining
process and structure using multi-body systems and implicit filtering" is to build a
simulation model of an existing turning lathe and to optimize the manufacturing
process with implicit filtering.
In this work, a facing process is used as test case. The machining operation runs
as follows: The turning tool moves to the point, where the machining starts. At
this point, the exact time of contact between the turning tool and the workpiece is
detected and an additional connection between cutting edge and workpiece is es-
tablished. This process is modeled by a modified unilateral contact augmented by
a cutting model.
In Section 16.2, an overview of the mechanical model of the turning lathe is
given as well as a model of the coupling between process and structure. For opti-
mization it is necessary to develop an objective function, where the quality and the
productivity of the processes have to be represented. The objective function is de-
veloped and presented in Section 16.3. Simulation results are utilized to calculate
the new surface and its characteristics, which criteria for an assessment of the
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 361–379.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
362 R. Britz et al.
machining quality are obtained from. These results, combined with the productivi-
ty of the process, are used as objective function variables. Optimized parameters
for the operation are detected with respect to the dynamic limits of the working
machine. Finally, results of the optimization process are shown and compared to
measurements.
16.2 Modeling
The mass matrix M depends on the vector of generalized position q . The vector
h holds all smooth external and gyroscopic forces depending on q , the genera-
lized velocities u and time t . For constrained systems, (Eq. 16.1) must be ex-
tended by the contact forces:
M ( q )u = h (u , q , t ) + Wλ . (16.2)
The constraint matrix W =[WN , WT ] provides the transformation from the space
of constraints to the configuration space. The vector λ = [λ N , λT ] holds values of
the normal and tangential contact forces.
16 Modeling and Simulation-Based Optimization of a Turning Process 363
rT t
f (q, s )= DT 1 =0 . (16.3)
r t
D 2
The difference vector rD = rC1 − rC 2 between the potential contact points on both
bodies is perpendicular to both tangents.
WN and WT are used for projecting normal and tangential contact reactions
in (16.2).
For the damping value λNd , a linear correlation is applied for g N < 0 :
Fig. 16.2 Linear and non-linear force law [2] Fig. 16.3 Set valued and regularized
form of COULOMB’s law [2]
Equivalent force laws are used for bilateral contacts without the distinction of
case in (Eq. 16.6). Friction is considered in closed unilateral contacts. Using the
normal force λ N and relative tangential velocity g T , the scalar force λT of planar
Coulomb friction is given by:
g T =0 λT ≤ μ λ N , (16.8)
16 Modeling and Simulation-Based Optimization of a Turning Process 365
g T
g T ≠0 λT =− μ λN . (16.9)
g T
This set-valued description is not suited for ordinary differential equation (ODE)
solvers, which are used for time-integration during simulation. Since high uni-
directional tangential velocities without sign-change are expected during the
cutting process, stick-slip transitions do not have to be taken into account, thus
enabling the regularization of (Eq. 16.9) and ODE time integration. One possibili-
ty is to smooth the set-valued force law of Coulomb as follows
−
g T
g T
g T < ε λT =− μ 1 − e λN ,
v ref
(16.10)
g T
g T
g T > ε λT =− μ λN . (16.11)
g T
In contrast to (Eq. 16.8) and (Eq. 16.9), (Eq. 16.10) and (Eq. 16.11) do not permit
real sticking characterized by g T =0 and i. g. λT ≠ 0 . (Eq. 16.10) gives only
λT =0 for vanishing tangential velocities at the contact point. Both Coulomb fric-
tion laws, exact and regularized, are depicted in Figure 16.3.
In order to determine the dynamic cutting forces at the workpiece and the tool cen-
ter point, depending on the process parameters, the applied materials and the tool
geometry, a modular cutting force model was set up. Please note that a detailed
description of the force model can be found in [6] and [7].
It is based on an analytical approach to describe the cutting forces. As there is
no formula for the general description of all turning processes, a modular structure
is chosen. Part formulations for the dynamic material behavior and non-linear fric-
tion at the tool chip interface can be selected, depending on the actual cutting
process. According to [8], the total forces in cutting FC, in feed FF and in passive
direction FP with respect to the side cutting edge angle ψ can be described as
follows:
′ + FCP
FC = FCS ′ , (16.12)
The prime symbol labels forces with respect to the cutting edge plane (see Fig.
16.4), while the indices S and P mark components resulting from shearing and
ploughing. [9] states that the shearing forces can be calculated by (Eq. 16.15) -
(Eq. 16.17) with the shear yield stress τs, the uncut chip thickness h, the width of
cut b, the rake, the chip flow and the inclination angle α, η and i as well as the
mean friction angle β and the shear angle Φ:
τ bh cos(β − α ) + tanitanηsinβ
′ = s ⋅
FCS , (16.15)
sinΦ
cos 2 (Φ + β − α ) + tan 2ηsin 2 β
τ s bh sin (β − α )
′ =
FFS ⋅ , (16.16)
sinΦcosi
cos (Φ + β − α ) + tan 2ηsin 2 β
2
One solution for the determination of the ploughing forces is based on edge coef-
ficients per unit K’, which can be determined by cutting tests and extrapolating the
measured forces to an uncut chip thickness of zero [9]
′ = K CP
FCP ′ ⋅ b , FFP
′ = K FP
′ ⋅ b , FPP
′ = K PP
′ ⋅b . (16.18, 16.19, 16.20)
Furthermore, a strictly analytic approach for the ploughing forces, based on metal
cutting simulations (see Sub-section 16.2.6.2), is described in [10].
16 Modeling and Simulation-Based Optimization of a Turning Process 367
Fig. 16.4 Machining variables and cutting force directions for tools with a side cutting edge
angle [7]
addition to β, the shear yield stress τs can be calculated from the stress distribution.
The shear angle Φ can be determined from the chip formation. In addition, the
point of material separation can be derived from the FEM results and used for the
analytic determination of the ploughing forces [10].
As the temperature in the deformation zone plays a major role for an accurate
simulation of the cutting process, an experimental setup has been developed to de-
termine the temperature of a real cutting process [6]. A high-speed pyrometer
(Kleiber KGA 740-LO) provides the possibility of capturing the temperature while
machining. To get results for both the workpiece and the tool flank face, a special
test piece with milled holes has been developed (see Fig. 16.6). While the temper-
atures of the workpiece in decreasing distance to the tool can be observed at the
beginning, the temperature of the tool is measured at the moment, when the tool
penetrates the milled cavity. Afterwards, these results were used to calibrate the
FEM metal cutting simulation.
Finally, all necessary parameters are transferred to the analytical cutting force
model so that it is possible to calculate the forces on workpiece and tool center
point resulting from the defined cutting process.
The validation of the modular cutting force model was performed by a comparison
of simulated and experimentally-determined cutting forces. Therefore, facing tests
370 R. Britz et al.
were carried out at a turning center while capturing the resulting forces by a 3-
component tool holder dynamometer (Kistler 9121) carrying a tool with a side cut-
ting edge angle ψ of 45° and an inclination angle i of zero. To avoid the influence
of machine vibrations the process was driven within stable cutting conditions.
The results for the machining of AISI 1045 can be found in Figure 16.7. The
simulated forces resulting from a model with the constitutive material law for
body-centered cubic materials and an additional calculation for blue brittleness
(see [7]) and the measured forces show high agreement in the area above 90 %.
An even higher accuracy can be achieved by calibrating the model with meas-
ured temperature and force values. At this, the model updating is performed by
iteratively optimizing the friction parameters. As reliable frictional and normal
stress distributions can be currently found only by experimental determination (see
[15]), it is of high importance to improve the theoretical description of interfacial
friction in future research.
Fig. 16.7 Comparison of measured and simulated forces, exemplarily in cutting direction
(AISI 1045, v = 160 m/min, b = 2.83 mm) [7]
In conclusion, the modular cutting force model is available for the calculation
of the cutting forces on workpiece and tool center point. Hence, the parameterized
analytical force model can be implemented into the Machine Process Interaction
simulation.
The simulation uses the cutting model mentioned above, which reduces the cutting
process to a single force between the chisel and the workpiece. The force applica-
tion point on the chisel is always constant and thus modeled as a point. Since the
16 Modeling and Simulation-Based Optimization of a Turning Process 371
force application point on the workpiece depends on the position of the chisel and
the angular position of the workpiece, it is modeled as a contour with switchable
radius, see Figure 16.9. Furthermore, the orientation between chisel and workpiece
has to be known to calculate the forces with (16.15) - (16.17). The position is af-
fected by the machine vibrations and must be determined at every time step; see
[19]. For that, the cutting edge and workpiece contour are augmented with vectors,
as illustrated in Figure 16.8.
The vectors describe the direction of the blade and the cutting face with respect
to the chisel. Using them together with the frenet trihedron of the contour and the
axial direction of the workpiece, all angles influencing the cutting forces can be
determined. For example, the angle κ is calculated from the bi-normal bw and
the direction of the blade sm , as follows:
π −bw sm
κ = − . (16.21)
2 bw sm
To figure out the sign of the angle we evaluated the scalar product with the normal
of the frenet trihedron nw :
Additional angles are calculated accordingly. Note that these operations are made
with respect to an absolute coordinate system.
Up to now, equations have been formed to compute the radial penetration and
all cutting angles. Penetration in axial direction and the discretization of the work-
piece, which yield an elliptical contact contour of the workpiece, are determined
by imperfections of the workpiece or clamping of the workpiece during the cutting
process. To this end, the oval contour of the workpiece is discretized during the in-
itialization, as shown in Figure 16.9.
372 R. Britz et al.
The radius of the ellipse is described in polar coordinates and is added to the
face side of the workpiece. The origin of the contour corresponds to the axis of ro-
tation. Before the cutting process starts, κ is set to 45°. During the machining, the
cut width b depends on time and increases. The cutting forces are a function of
chip width b and depth h , so b and h have to be evaluated in each time step.
For this purpose the initialized, previous and current contours and the length of the
workpiece are stored. The gap between workpiece and chisel is calculated from
the contact kinematics between the previous contour and the position of the cut-
ting edge. Using the contour parameter s , the current chip depth is estimated and
the current contour radius is stored. Also, the current position of the cutting edge,
the angle κ and the chip width in radial direction brad are calculated using the ra-
dius of the initialized contour. The chip width in axial direction, bax results from
the length of the workpiece minus the cutting edge position. The values brad and
bax are compared and the smaller one is used to calculate the actual chip width b
in the current state.
The model for the contact between turning tool and workpiece introduces the inte-
raction between the cutting process and the machine: one of the important pheno-
mena in process-structure-interaction is the separation of turning tool and
workpiece during the cutting process. The theory presented in the preceding sec-
tions allows the description of this behavior. The contours of the bodies are de-
scribed by a point (turning tool) and an oval contour (workpiece). For modeling
the cutting process, the gap between turning tool and workpiece is needed, which
is calculated from (16.3). With the achieved contour parameter s , the distance or
penetration results from g N = rSWS n1 . In addition, if penetration is detected with the
known position vector r , the new surface is calculated and recorded at every time
step. The values of the radius are recorded in a table and interpolated during the
next revolution. The solver needs to detect transitions and distinguish between the
possible configurations; see [19]. For this purpose, three operation states are de-
fined according to Figure 16.10:
16 Modeling and Simulation-Based Optimization of a Turning Process 373
a) Open contact: turning tool and workpiece are separated and the gap is positive.
b) Closed contact, no cutting: the turning tool contacts the workpiece without cut-
ting. Thereby, no chip is taken from the workpiece. At the point of contact,
normal and tangential forces are exchanged as for a standard frictional con-
tact. The penetration is smaller than a defined minimal penetration ε.
c) Closed contact, cutting: the penetration exceeds ε. Cutting starts and the contact
forces are overlaid by the cutting forces. During the cutting process, the new
surface resulting from the process needs to be recorded.
The numerical simulation accurately detects the transition times between the
operation states by event-based integration. It uses a root function to monitor the
contact states of the cutting model. This includes any overruns and drops below
the minimal penetration level, which describes the beginning and end of cutting.
16.3 Optimization
The aim of the project is to determine the optimal parameter set for an operation,
i. e. detect a set, which satisfies the economic aspect of the production. The com-
prehensive model of process and structure is used to predict the result of the ma-
chining; see [20]. With this information, the optimization algorithm is able to
determine the best parameter set.
The work of [16] shows that even optimization problems concerning rather simple
mechanical systems can result in target functions, which are non-smooth, have
many local minima and may chaotically depend on the optimization parameters.
Based on these experiences, the optimization algorithm IFFCO [17] is chosen, an
implementation of the implicit filtering method [18] for problems with bound
constraints.
On the other hand, the assessment of quality is not straight forward since it de-
pends on forces, deflections, vibrations, geometry of the tool, etc. In this work, the
objective function considers only deflections and the tool geometry to assess the
new surface and forces to value the tool wear.
The simulation records contact and cutting forces as well as the trace of the
turning tool on the workpiece. The objective function evaluates some specific val-
ues: first, the mean offset Δz of the turning tool path from the desired path,
representing the static deflection during the cut. Another important factor is the
theoretical roughness of the produced surface. Using the trace upon the workpiece
and the radius of corner it can be calculated as follows: a set of points in radial di-
rection is chosen within a small region around the x axis and used as supporting
points, see Figure 16.11a. Using the radius R and feed per revolution f , the
piecewise surface contour in radial direction is calculated, see Figure 16.11b:
2
r f f
zi ( x) =zi 0 + R1 − 1 − ; for − <r< . (16.26)
R 2 2
16 Modeling and Simulation-Based Optimization of a Turning Process 375
Next, we calculate the commonly-known center line average Ra and average sur-
face roughness Rz along this contour. Note that these values are only theoretical
estimates of surface quality. In real processes, surface roughness is also influenced
by other effects like abrasion thermal stress and so on. However, the trend of qual-
ity is determinable and adequate for optimization.
The cutting force indicates the stress on the turning tool tip and the workpiece
surface. The ratio of the mean values of cutting force FC and normal force FN on
the cutting edge gives an advice of the abrasion. The objective function is eva-
luated using the simulation results. Using factor a we are able to weight produc-
tivity and quality:
R R Δz FN Q0
g = (1 − a) a + z + + + a . (16.27)
a0
R R z0 Δ z 0 FC Q
While the chosen objective function was successfully applied to find parameter
settings, which guarantee a good compromise, the real target is different: we want
to determine a parameter setting, which realizes the maximum productivity at a
minimum quality necessary to satisfy our technical requirements. Hence, factor a
would have to be known a priori. However, by introducing penalty terms, if the
surface quality is below a certain boarder, the optimizer can be utilized in a differ-
ent way: We want to get a minimum quality, therefore, the best parameters for the
maximum productivity have to be generated by the optimizer. At first, the virtual-
ly-produced surface is interpreted. If the minimum quality boarder is crossed, a
penalty is added to the objective function. If not, the objective function is calcu-
lated as given in (16.27).
376 R. Britz et al.
Fig. 16.12 Change of quality, force ratio and productivity during the optimization process
These results show that the simulation tool is able to find an optimal setup for
smooth finishing, regarding the used model.
These results are compared with measurements on the machine unit. Since the
number of experiments is limited only certain machine settings could be realized.
Therefore, some parameters, which strongly affect surface roughness, tool stress or
productivity, were chosen, e. g. the settings for the cut at the beginning and at the
end of the optimization. Table 16.1 lists parameter values used in the experiments.
In Figure 16.13, the theoretical roughness is compared to the measured rough-
ness during the optimization process. The difference between simulation and ex-
periment increases with a decreasing feed per revolution because the influence of
the chisel geometry decreases and the influence of the abrasion on the workpiece
increases. The abrasion effect is not modeled; the lower boarder of the box con-
straint must be modified. The failure increases rapidly, if both cutting parameters
go to zero.
16 Modeling and Simulation-Based Optimization of a Turning Process 377
Step 1 2 5 7 8 9 11 30 40 52
Infeed 2.15 0.2 0.2 0.68 0.2 0.34 0.34 0.49 0.43 0.46
[mm]
Feed 0.1 0.57 0.34 0.10 0.21 0.06 0.18 0.07 0.07 0.08
[mm/rev]
Fig. 16.13 Comparison of the measured roughness and the theoretical roughness. The trend
of the produced roughness is predictable.
The predicted and the measured forces are shown in Figure 16.14. The mean
value of the predicted force is also close to the measured force.
Fig. 16.14 Comparison of the mean values of the measured forces and the predicted forces
The comparison shows that the simulation sufficiently predicts product quality
and tool stress. This allows the determination of an optimal cut setting, which sa-
tisfies the demands on productivity and quality.
378 R. Britz et al.
In general, the result is influenced by the machine. The developed tool is able
to optimize the process with respect to the machine. This is figured out in [20]. In
this article, two virtual machines with different stiffnesses are used. A rough ma-
chining process is optimized for both machines. The optimization tool finds dif-
ferent parameter settings on the machines, indicating that the optimization tool
figures out a parameter setting with respect to the process-structure interaction.
16.5 Conclusion
A tool for the optimization of turning processes has been introduced, which in-
cludes the model of the turning machine, the process and the target function. The
capability of the optimization environment is illustrated by means of examples. The
tool obtains the optimal parameter settings for a smooth finishing process with the
results of the simulation having been successfully verified by experiments.
References
[1] Pfeiffer, F., Glocker, C.: Multi-body Dynamics with Unilateral Contacts. John Wiley
& Sons, New York (1996)
[2] Britz, R., Ulbrich, H.: Lathe: Modeling and Coupling of process and structure. In:
Proccedings of the 1st International Conference on Process Machine Interactions, pp.
231–238 (2008)
[3] Förg, M., Zander, R., Ulbrich, H.: A framework for efficient simulation of spatial
contact problems. In: ECCOMAS (2007)
[4] Shabana, A.A.: Dynamics of Multi-body Systems. Cambridge University Press,
Cambridge (1998)
[5] Ulbrich, H.: Maschinendynamik. Teubner Verlag, Wiesbaden (1996)
[6] Zaeh, M.F., Schwarz, F.: Consideration of Tool and Workpiece Temperatures in a
Modular Cutting Force Model. In: Proceedings of the 1st International Conference on
Process Machine Interactions, pp. 353–360 (2008)
[7] Zaeh, M.F., Schwarz, F.: Modeling and Simulation of Process and Structure Interac-
tions Considering Turning Operations. In: Proceedings of the ASME International
Manufacturing Science and Engineering Conference (2009)
[8] Oxley, P.L.: The mechanics of machining: an analytical approach to assessing machi-
nability, Chichester u.a., Horwood (1989)
[9] Altintas, Y.: Manufacturing Automation. Cambridge University Press, Cambridge
(2000)
[10] Schwarz, F.: Simulation der Wechselwirkungen zwischen Prozess und Struktur bei
der Drehbearbeitung. iwb-Forschungsberichte Nr. 242. Herbert Utz Verlag, München
(2010)
[11] Yen, Y., Jain, A., Altan, T.: A finite element analysis of orthogonal machining using
different tool edge geometries. J. Mat. Proc. Technol. 146, 72–81 (2004)
[12] Zerilli, F., Armstrong, R.: Dislocation-mechanics-based constitutive relations for ma-
terial dynamics calculations. J. Appl. Phys. 61, 1816–1825 (1987)
[13] Zerilli, F.: Dislocation-mechanics-based constitutive equations. Metallurgical and
Material Transactions 35A, 2547–2555 (2004)
16 Modeling and Simulation-Based Optimization of a Turning Process 379
[14] Sartkulvanich, P., Koppka, F., Altan, T.: Determination of flow stress for metal cut-
ting simulation – a progress report. J. Mat. Proc. Technol. 146, 61–71 (2004)
[15] Özel, T.: The influence of friction models on finite element simulations of machining.
Int. J. Machine Tools and Manufacture 46, 518–530
[16] Neumann, L., Ulbirch, H.: Optimisation of a CVT-Chain. In: Proceedings of the 6th
World Congresses of Sturctural and Multidisciplinary (2005)
[17] Choi, T.D., Eslinger, O.J., Gilmore, P.A., Kelly, C.T.: User’s Guide to IFFCO. Center
for Research in Scientific Computation, North Carolina State University (2001)
[18] Kelley, T.C.: Iterative Methods for Optimization. Frontiers in Applied Mathematics.
SIAM, Philadelphia (1999)
[19] Britz, R., Ulbrich, H.: Simulation of Facing Processes of Profiles with Hexagon Cross
Section. In: Proceedings of the ASME International Manufacturing Science and En-
gineering Conference (2010)
[20] Britz, R., Ulbrich, H.: Modeling and Simulation-based Optimization of a Facing
Process. In: Proceedings of the 2ndt International Conference on Process Machine
Interactions (2010)
Part IV
Forming
Chapter 17
Advanced Forming Process Model - AFPM
Abstract. This chapter discusses methods of modeling and simulating metal form-
ing processes and explains their application in product design, production and
process planning. In today’s Finite Element (FE)-based forming analysis, major
effects on the forming process are being neglected. Based on the analysis of the
elastostatic press and tool properties, a conventional FE process model was ex-
tended with the most dominant elastostatic influences. It is shown that complex
elastic systems, such as die cushions and tool guidance, are quite easily imple-
mented in FE process simulations by using discrete elements and other reduction
methods, recently introduced in commercial simulation software. The benefit of
the Advanced Forming Process Model (AFPM) is demonstrated by an experimen-
tal verification. Servo mechanical presses enable the manufacturers to establish
high-speed processes in sheet metal forming. There, the dynamic press behavior
has a much larger influence on the forming process than it has in the relatively
static conventional deep drawing. As an example, a highly dynamic forming
process is simulated and explained in the following.
17.1 Introduction
Established as a powerful tool for designing sheet metal processes, the Finite Ele-
ment Method (FEM) supports product development as well as the planning of
production process and tool manufacturing.
Current process models merely contain the blank and the interface between
blank and tool, the so-called die-workpiece interface [1]. Commonly, the blank is
meshed with shell elements with simple constitutive equations. The die-workpiece
interface is usually represented by the Coulomb friction law and a contact algo-
rithm with constant coefficients. In present FE-simulations, the forming tool and
machine are modeled as rigid, i. e. the effects of their elastic properties on the
forming process are ignored. In reality, both forming tool and press clearly affect
the forming process as the requirement of die spotting during the tool try-out de-
monstrates. One way to minimize time and effort in tool try-outs is using more ac-
curate simulation methods during the tool planning stage.
New press technologies, such as servo-mechanical presses, enable faster
processes with remarkable accelerations. Higher accelerations require light parts
in motion. Lighter tools and press parts are less stiff and would therefore deform
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 383–401.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
384 K. Großmann et al.
stronger under process load. Conversely, these elastic deformations influence the
process conditions and therefore the final part shape significantly. Conventional
FE process simulations do not show the impact of elastic deformation on the
process.
Therefore, process simulations are required, which take elastic tool and press
properties into account and enable manufacturers and press designers to compute
the process-machine interactions. A better quality and efficiency control of metal-
forming processes already during the tool’s planning stage require interactions be-
tween the subsystems machine, tool and workpiece to be included in a complete
forming model.
The objectives of this chapter are
• introducing a holistic yet simple approach to address elastostatic deformations
of press and tool in FE process simulation and
• showing a way to simulate highly dynamic forming processes.
Holistic process modeling of forming operations entails the coupling of the three
subsystems, press, tool and workpiece. Concepts as to how to connect the “work-
piece” model (the common metal forming model) to the press model are presented
in [4, 14]. The approaches are differentiated according to their way of integration:
• Offline coupling (non-reactive) is process characterization within the machine
model based on process force progressions. The force curve is calculated by
means of an FE-workpiece model. The data are transferred to the press model
via an ASCII-file [5]. Among others, this coupling method is applicable to ana-
lyze the operating performance of the machine and to detect the load on assem-
blies. In [5, 6], the ram deflections of a multi ram press were estimated. In [7],
the effects of ram tilting on the deep drawing process were analyzed.
• Integrated coupling stands for the extension of the workpiece model with a
press model in the FEA-environment. The integrated FE-model realizes direct
interaction between the process load and tool position as a result of the press
behavior. The concept describing the press by a reduced structure representing
process relevant effects is exemplified in [2, 3].
• By coupling discrete models (co-simulation) the machine effects are depicted
in independent press models. Advantageously, the complex influences of the
machine behavior (drive, guidance system, frame, etc. [9]) are modeled in de-
tail by Multi-Body-Simulation (MBS) while the workpiece description is made
by FEM. Thus, two simulation tools (MBS and FEM) connected by simulator
coupling are to be applied. The simulator coupling organizes the exchange of
data and synchronizes the different solution algorithms. A complete forming
model by coupling discrete simulation models is exemplified in [8]
17 Advanced Forming Process Model - AFPM 385
To design the forming process only the press behavior directly affecting the
process is relevant. Interactions between assemblies inside the press are less im-
portant. Hence, the integrated solution realized in [2, 3] is adequate. Furthermore,
applying the integrated approach avoids the independent major problem of coupl-
ing different simulation software.
Fig. 17.1 Advanced Forming Process Model (AFPM) a) extended by a press model consi-
dering elastic mounting of the rigid ram in the COG, b) extended by a press model sepa-
rated in drive and guidance springs for the application of an elastic ram (example: O-frame
press with 4-point configuration)
In the next step, the forming process model is enhanced by modeling the tools
as elastic. This approach demands the tool to be elastically embedded into the ma-
chine model. The press ram and the table of the press have to be modeled as elas-
tic. As in real press structures, the elastic ram model’s joints to the press structure
are positioned at the locations of press drive and ram guidance.
Figure 17.1b illustrates the AFPM upgraded with spring elements, which act in
horizontal (X and Y) direction and in vertical (Z) direction. According to the real
press drive system, they are connected to the corner nodes of the ram model. The
following properties, which influence the press behavior, define the position and
parameterization of the spring elements:
Models of different press structures are shown in Figure 17.2. The stiffness val-
ues of the horizontal and vertical springs are derived from force-displacement
curves measured as stated by standard DIN 55189 [10]. The measuring proce-
dure to determine the deflection of the press under static load is explained in
Section 1.2.1.
17 Advanced Forming Process Model - AFPM 387
The principle of moments enables the conversion of the measured values of tilt-
ing stiffnesses cα, cβ and total vertical stiffness ctotZ to spring constants cX, cY and
cZ. For this purpose, geometrical parameters as ram size zram and the distances xdr
and ydr between force transmitting drive elements are necessary.
Following this principle, the tilting stiffnesses and the vertical total stiffness
were converted to stiffness parameters of the horizontal and vertical spring ele-
ments for all model structures in Figure 17.2. The transformation functions have
been summarized in Figure 17.3.
In case there are no measured values for the total stiffnesses available, refer-
ence values from [11] could be used alternatively. There typical ranges of vertical
total stiffnesses and tilting stiffnesses, classified by the press’ nominal force, can
be found. The ram height zram can be deduced from characteristic dimensions of
the press’ workspace. In the following example, the parameters for a press with a
nominal force of 1000 kN are used.
388 K. Großmann et al.
Fig. 17.4 Model structure of the die cushion for applying an elastic blank holder
The die cushion model for an elastic blank holder is shown in Figure 17.4. The
stiffnesses of die cushion drive, guidance, pressure pin and pin guidance are
represented by substitutional springs. These spring elements are arranged on the
coupling nodes of the elastic blank holder as they are in the real structure. The
stiffness of the spring element was parameterized according to [13].
17 Advanced Forming Process Model - AFPM 389
Fig. 17.5 Sheet thickness reduction and material flow, O-frame press with 1-point
configuration a) Rigid press and tool, die cushion supported rigidly, b) Elastic press, die cu-
shion supported rigidly, rigid tool, c) Elastic press, die cushion supported elastically, rigid
tool
the simulation results. Without considering the reinforcing effect of the tool guid-
ance the production of sheet metal parts cannot be guaranteed simulatively so that
the design of the parts would have to be changed unnecessarily at this.
The tool guidance can also be modeled with discrete elements. Figure 17.6
shows the reduction of the tool guidance (a) to a beam model (b) and further into a
discrete model with spring elements (c).
The stiffness of the tool guidance in x and y-direction has the characteristic of a
beam and is dependent on the free length of the tool guidance post (lTG). Since lTG
depends on the current press stroke the stiffness of the spring element is coupled
with the z-coordinate of the ram.
During the entire process, nodes N1, N2, N3 and N4, used to couple ram, BH
and bolster plate on the tool guidance, stay aligned along an axis perpendicular to
the surface of the bolster plate. The change of the ram’s instantaneous center of
rotation with stroke is automatically implemented in the FE simulation.
Fig. 17.6 Implementation the elastic behavior of tool guidance into FE process simulation:
a) section view press, b) tool guidance as beam model, c) tool guidance modeled with
spring elements
Fig. 17.7 Tilting behavior and press model with non-linear compression springs
Since the approach to using linear-elastic springs did not correspond to the real
tilting and stiffness behavior of the examined press, the press model is refined
with non-linear spring characteristics. In order to correctly consider the tilting be-
havior, which is dependent on the direction of the rotation, only non-linear com-
pression springs were attached to the corner nodes of the ram, see Figure 17.7.
The stiffness properties of the experimental tool were modified by implement-
ing disc springs in the die to be able to vary the total stiffnesses in the press-tool
system (Fig. 17.8).
In order to determine the relevant process characteristics it is necessary to in-
stall appropriate sensors. For recording the deflection, behavior incremental dis-
placement encoders with a measuring resolution of 1µm were attached to the ram
guidance. Furthermore, inductive displacement sensors were attached to the tool to
measure the distance between die and die bearing plate. Thus, the tilting of the die
provoked by using different disc spring configurations was measurable. Magneto-
strictive displacement sensors were used to determine the tilting and the vertical
resilience of the blank holder.
392 K. Großmann et al.
Fig. 17.8 Modification of the total stiffness and the tilting stiffnesses in the experimental
tool by means of variable disc springs
The punch force was measured with four load-cells placed under the punch.
The press control systems delivered blank holder and ram forces. All input signals
were recorded with PC measuring boards. The measuring results were visualized
using the software Diadem 8.0 by National Instruments. An overview of the mea-
suring system is shown in Figure 17.9.
influence on the total stiffness of the experimental set-up. The elastostatic proper-
ties of both die cushion and tool guidance were included by modifying the spring
characteristics of the blank holder model (cx±B,NH; cy±A,NH; cz,NH)
Fig. 17.10 Advanced Forming Process Model considering the elastic properties of the expe-
rimental equipment.
The blank was meshed with shell elements with 5 integration-points over the
thickness. Adaptive meshing was applied. The elastic-plastic material behavior of
DC04 and H260B was modeled using Reihle’s law and parameters from [15] to
describe the flow stress curve. The friction between sheet metal and tool compo-
nents was modeled with the classical Coulomb-Friction-Model and a constant fric-
tion coefficient of 0.1. In order to improve performance and stability of the
simulation the relative tool travel was realized by moving the punch, which is
fixed in reality, instead of the ram.
Fig. 17.11 Comparison of the simulation results with the experimental results - remaining
flange widths for different blank holder forces
Comparing the flange widths MP1 and MP3 on the long sides of the rectangular
cup indicates an asymmetric material draw-in. Due to ram tilting about the x-axis
the material draw-in at MP1 is more strongly restrained than at MP3. Figure 17.11
shows a good agreement between simulation and experiment for these particular
measuring points. This demonstrates that the advanced process model obtains
closer-to-reality results than the classical deep-drawing simulation.
The sheet metal thickness distribution, for example, at bottom corners of the
rectangular cup, reflects the different conditions in the contact surfaces between
sheet metal and tool. There, the substantially larger blank holder pressure (here at
the left flange side) causes a larger thickness reduction.
In order to avoid critical states in workpiece and tool as well as to achieve the
desired workpiece quality, it is important to be aware of the accurate state of de-
formation and stress during the forming process. The state of deformation and
tensile stress was experimentally determined for selected specimens in order to
evaluate the FE simulation results. Therefore, a square line grid was applied on the
blank surface and then scanned by means of photogrammetry. The image
processing returns the xyz-coordinates of the grid nodes.
The deformation and stress analysis as well as the data visualization and the
date storage were realized with the PC software AutoGrid® [16].
The simulation results can be examined based on the determined total or equiv-
alent strains. Computed results and experimentally determined results show a
good agreement. The state of stress for this particular case can be modeled with
high accuracy. This is the pre-condition for an identification of deep-drawing fail-
ures by analyzing the plane strain state with a Forming Limit Diagram (FLD)
Furthermore, the prediction of the wrinkles and cracks using an FLD was con-
firmed by the experimental results. The Forming Limit Curve (FLC) was generat-
ed according to the European standard ISO 12004 [17].
17 Advanced Forming Process Model - AFPM 395
Fig. 17.12 Comparison between the results of the simulation and the experiment - analysis of the
drawing failures using a FLD
The results obtained with the advanced forming process model (Fig. 17.12)
show that the computation of wrinkling and the prediction of cracks match the ex-
perimental test results. Comparing these findings to the results of a classical deep-
drawing simulation clearly shows that experimentally observed deep-drawing
failures could not be predicted without including elastic properties of machine and
cushion into the FE-process model.
396 K. Großmann et al.
Since the simulation of the elastostatic machine and tool behavior is feasible,
error compensation for these effects seems to be the next logical step. Therefore, a
methodology was developed counteracting elastostatic press and tool deformations
by rectifying the tool’s surface before it is built. The surfaces of the tooling after
springback compensation are fed into the FE simulation. The FE model, enhanced
with important elastic properties, calculates the elastic deformation of the tool’s
surface. In the next step, the deformed surface is extracted and the deviation is
added as offset to the original surface (Fig. 17.14)
The experiments were conducted on an experimental set-up with allows for ex-
tensive die deformation [19]. Figure 17.15 displays experimental and simulation
results for deep drawing with and without compensated die surface. The flange
outline (dashed line) shows that the material draw-in with compensated die sur-
face is very close to the draw-in predicted for deep drawing with an ideally stiff
ram. This indicates that the compensation for elastic press and tool deformations is
feasible.
quasi-static process simulations are obviously not suitable to calculate process pa-
rameters and reaction forces under dynamic conditions. Therefore, new methods
to simulate and predict process conditions for dynamic operations need to be es-
tablished [9].
The principle of “Synchro“ deep drawing was developed at the IWU-Chemnitz-
Dresden and is explained in [20]. In this process, the blank holder (BH) is sub-
jected to a blank holder force (BHF) oscillating between zero and the maximum
value. The die is attached to the ram, which follows a pre-defined displacement
curve in such a way that the part is drawn, when no BHF is applied. When the die
stops, the BHF rises and the wrinkles in the flange area of the part are flattened
out. Preliminary experiments have shown that the process has potential to achieve
larger drawing depths. The synchronization and optimization of the individual
displacement and force curves is a task for the FE process simulation.
Figure 17.16 shows the FE simulation of “Synchro” deep drawing. The die dis-
placement curve and the BHF oscillation serve as input parameters. Since it is not
possible to apply both displacement and force boundary condition to one rigid
body simultaneously, the BHF movement away from the sheet is realized by ap-
plying a negative BHF. The resulting BH displacement is limited to the desired
maximum distance between BH and sheet by using the option of a rigid body
stopper [21]. The rigid body stopper is synchronized with the die displacement.
Mass scaling is common practice in quasi-static FE simulations to achieve reason-
able CPU times [21].
When simulating “Synchro” deep drawing with explicit FE codes, mass scaling
is not an appropriate measure. Since the dynamic behavior is dependent on the
moved masses and their accelerations, the parts in motion must be modeled with
real masses, which are exposed to the real accelerations. Even selective mass scal-
ing of the sheet metal, which is very light compared with the press ram, is
not feasible since the oscillation of the BH excites the blank to vibrate. The
400 K. Großmann et al.
References
[1] Roll, K., Wiegand, K.: Eine Möglichkeit zur Berücksichtigung der Werkzeugelasti-
zität bei der Blechformsimulation. In: Proceedings of Neuere Entwicklungen in der
Blechumformung Leinefelden-Echterdingen (Stuttgart), MAT INFO Werkstoff-
Informationsgesellschaft mbH, Frankfurt (2006)
[2] Großmann, K., Hardtmann, A., Wiemer, H.: Simulation des Blechumformprozesses
unter Berücksichtigung des statischen Verhaltens der Pressmaschine. ZWF Zeitschrift
für wirtschaftlichen Fabrikbetrieb 101(10), 600–605 (2006)
[3] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Stand der Simulation von
Umformprozessen mit den elastischen Einflüssen aus Maschine und Werkzeug. In:
Proceedings of EFB-Colloquia Neue Wege zum wirtschaftlichen Leichtbau, pp. 185–
198. EFB, Hannover (2007) (in Fellbach)
17 Advanced Forming Process Model - AFPM 401
[4] Großmann, K., Wiemer, H.: Maschine, Werkzeug und Prozess – ein ganzheitliches
System. In: Proceedings of 6th Sächsische Fachtagung Umformtechnik, TU Dresden
(1999)
[5] Großmann, K., Wiemer, H., Thoms, V., Schirmacher, F.: Modellierung des Blechum-
formprozesses in der Anlagensimulation auf Grundlage von Ergebnissen der FEM-
Prozeßsimulation. EFB-Forschungsbericht No. 178. EFB, Hannover (2001)
[6] Großmann, K., Wiemer, H.: Simulation of the Process Influencing Behaviour of
Forming Machines. In: Steel Research International, vol. 76(2/3), pp. 205–210. Ver-
lag Stahleisen GmbH, Düsseldorf (2005)
[7] Schirmacher, F., Wiemer, H.: Kopplung von Prozess- und Anlagensimulation – ein
wesentlicher Teilschritt auf dem Weg zum Virtuellen Umformen. In: Proceedings of
EFB-Colloquia Prozessoptimierung in der Blechverarbeitung, pp. 11.1–11.18. EFB,
Hannover (2001)
[8] Brecher, C., Schapp, L., Paepenmüller, F.: Gekoppelte Simulation von Presse und
Massivumformprozess. Werkstattstechnik Online 96(7/8), 526–529 (2006)
[9] Wiemer, H.: Stand und Möglichkeiten der Systemsimulation von mechanischen
Pressmaschinen. State and Potentials of Forming Press Simulation. Dissertation, TU
Dresden (2004)
[10] German Standard DIN 55189, Ermittlung von Kennwerten für Pressen der Blechve-
rarbeitung bei statischer Belastung. Beuth Verlag, Berlin (1988)
[11] VDI-Guideline VDI 3145, Pressen zum Kaltmassivumformen. Blatt 1: Mechanische
und hydraulische Pressen. VDI-Verlag, Düsseldorf (1985)
[12] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Faster to Sound Parts by Ad-
vanced Forming Process Simulation - Advanced Forming Process Model Including
the Elastic Effects of the Forming Press and Tool. In: Steel Research International,
vol. 78(10/11), pp. 825–830. Verlag Stahleisen GmbH, Düsseldorf (2007)
[13] Pahl, K.J.: Elastische Wechselwirkungen im Ziehapparat einfach wirkender Pressen.
VDI Verlag, Düsseldorf (1994)
[14] Großmann, K., Wiemer, H.: Möglichkeiten der erweiterten Modellierung von Um-
formprozessen zur Berücksichtigung der Wechselwirkungen mit Werkzeugen und
Maschine. In: Proceedings of Neuere Entwicklungen in der Blechumformung. MAT
INFO Werkstoff-Informationsgesellschaft mbH, Frankfurt (2008)
[15] Thoms, V., Voelkner, W., Süße, D.: Methoden zur Kennwertermittlung für Blech-
werkstoffe. EFB-Forschungsbericht No. 187. EFB, Hannover (2002)
[16] Feldmann, P.: Application of strain analysis system AutoGrid® for evaluation of
formability tests and for strain analysis on deformed parts. In: Proceedings of the
IDDRG Conference Györ, Hungary, pp. 21–23 (2007)
[17] European Standard, EI 12004, Metallic materials - Sheet and strip - Determination of
forming limit curves. CEN European Committee for Standardization, Brussels (2007)
[18] Haufe, A., Roll, K., Bogon, P.: Sheet metal forming simulation with elastic tools in
Ls-Dyna. In: Proceedings of Numisheet, pp. 743–748 (2008)
[19] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L., Kriechenbauer, S.: Static
Compensation for Elastic Tool and Press Deformations during Deep Drawing. Prod.
Eng. Res. Devel. 4(2-3), 157–164 (2010), doi:10.1007/s11740-010-0216-7
[20] Großmann, K., Wiemer, H., Neugebauer, R., Mauermann, R.: Simulationsun-
terstützung für das Tiefziehen auf Servospindelpressen. In: Proceedings of EFB-
Colloquia: Bauteile der Zukunft. EFB, Hannover (2010)
[21] Hallquist, J.O., et al.: LS-DYNA Keyword User’s Manual, vol. I, Version 971, Li-
vermore Software Technology Corporation, Livermore California (2007)
Chapter 18
Consideration of the Machine Influence
on Multistage Sheet Metal Forming Processes
Abstract. Metal forming processes are highly affected by the properties of the
forming machine. In multistage processes, the force path curve of each stage is in-
fluenced by the surrounding stages. This research focuses on the interaction be-
tween the forming machine and multistage processes with respect to the quality of
the workpiece. The accuracy of the numerical results could be increased by coupl-
ing the process simulation with the machine simulation. Further influencing fac-
tors, such as different friction conditions during forming and the elasticity of the
tool, were considered in the simulation. In addition, the thermal properties of the
press were investigated and considered in the machine simulation. Hence, a better
correlation between the numerical results and the experiment could be achieved.
18.1 Introduction
Metal forming processes, such as deep drawing, place high demands on the prop-
erties of the forming machine. Even slight scatters or inaccuracies in the ram mo-
tion can lead to imperfections of the workpiece. The characteristic deformation of
the forming machine and the tool system under load has a crucial influence on the
result. To simulate forming processes, the use of the finite elements method
(FEM), where local influences on the process (e. g. the lubrication or inhomoge-
neous material properties) can be included in the simulation, is state of the art. In-
fluences between the machine and the forming process have not been taken into
account yet. In metal forming simulations, an ideal rigid forming machine is as-
sumed. Thus, differences between simulation and reality occur. Hence, for many
applications it is not possible to manufacture parts with the required accuracy right
from the beginning so tools have to be set up for each machine in a try-out phase.
In order to reduce the start-up time for new metal forming tools it is necessary
to consider the properties of the forming machine in the FE process simulation.
Different approaches for a coupled simulation to consider the interaction between
forming machine and forming process are known. A classification is drawn be-
tween offline-coupling, model integration and co-simulation.
In offline-coupling, the process simulation and the machine simulation are cal-
culated separately. After calculating the process, the resulting force path curve is
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 403–417.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
404 B.-A. Behrens et al.
transferred into the machine simulation as input data for a new calculation. Subse-
quently, the correction of the ram movement as a result of the machine simulation
serves as a boundary condition for the FEM simulation, if a new calculation by the
two models is performed by means of an iterative loop. In several loops, an ap-
proximation to the real workpiece dimensions can be achieved [1]. The most im-
portant advantage of offline coupled simulations is that both simulations are
implemented in their common software environment. Due to this, both simulations
can benefit from the tools, which are available in the typical software environ-
ment. The disadvantage of this approach is the fact that no dynamic interaction
can be simulated and that the calculation time increases due to the repeated
calculation.
An approach, where the machine and the forming process are simulated in a
common simulation environment, is called model integration [2, 3]. In [4], the
machine structure was reduced and integrated into the FEM process model by
means of elasticity conditions. The tilt and the vertical deflection as well as the de-
formation of the die cushion and the ram were taken into consideration by means
of spring elements. This approach resulted in an improvement of the accuracy of
the forming simulation. However, if the machine properties and the drive control
are examined, this method is limited.
In co-simulation two independent, sufficiently detailed models for the forming
machine and the forming process, which exist in different simulation environ-
ments, are used [5, 6]. These models exchange data after each time-step. Thus, it
is possible to consider dynamic influences within the simulation, which occur dur-
ing the forming process. In [7], a coupled simulation for bulk metal forming
processes was realized. Due to the consideration of the machine properties the ac-
curacy of the forming simulation could be increased. Furthermore, the results of
the forming simulation were compared using machine models with different de-
grees of detail. Even with the least detailed machine model the simulation results
of the workpiece dimensions could be approximated to reality by means of a co-
simulation in comparison to a separate process simulation.
Within this project, an offline-coupling was implemented. The machine model
is based on a phenomenological approach. The process simulation is modeled in a
classic FEM environment. The simulation of single processes is state of the art.
This project deals with the simulation of multistage processes, where interactions
between each stage have to be considered additionally.
18.2 Machine
Experimental investigations were carried out on a high-speed press with a space
rod drive (fig. 18.1). The advantage of a space rod drive is the independent ad-
justment of the ram position and the length of the stroke. The connection rods of a
conventional mechanical press can only be moved in two directions. The connec-
tion rods of a space rod drive have one more degree of freedom. Thus, the kine-
matics of the press result in a slower course of motion shortly before reaching the
bottom dead centre (bdc). This is especially advantageous for deep drawing
processes. The bdc marks the reversal point of the ram.
18 Consideration of the Machine Influence 405
18.2.1 Approach
The method to describe the elastic properties of forming machines is based on the
results of experimental tests using a phenomenological approach. The deflections
of the forming machine under process load are measured and reproduced using an
elasticity matrix with variable elements as machine model. The required informa-
tion for the model is provided by means of a single measurement of the press,
based on a static press measurement similar to the standard DIN 55189 [8]. Due to
this fact the model is applicable for many metal forming machines after carrying
out this measurement.
The relative total shift between the ram and the bolster plate results from de-
formations of the connection rods, the ram guide, the press frame, the bolster plate
and the ram as well as the partially non-linear elastical characteristics of bearings
and guidances. This leads to a shift of the position of the bottom dead centre. The
deformations of the press components, except the ram and the bolster plate, can be
summed up as deflection of the ram in relation to the bolster plate including the
tilting. The deflection and the tilting are first calculated in the machine simulation
assuming a rigid bolster plate and ram. The relative total shift is calculated in a
further step by superposing the deflection with the deformation of the bolster plate
and the ram (Fig. 18.2). For employing the principle of superposition it has to be
assumed that superposed loads cause summable deformations. This principle is
applicable as long as there is a linear correlation between the loads and the result-
ing deformations [9]. This approval is valid for very stiff bodies as the ram and the
bolster plate.
406 B.-A. Behrens et al.
Unloaded Loaded
Ram
Modelling:
1. Load in stage 1
Hydr. cylinder
Pressure
measurement Hand pump
0.9 Stage 1
0.8 Stage 2
Total shift at stage 1 [mm]
F
0.7
F
0.6
0.5
0.4 Stage 3
0.3
F
0.2
0.1
0.0
0 50 100 150 200 250 300
Load [kN]
Fig. 18.4 Total shift at stage 1 dependent on a load at stage 1, stage 2 and stage 3
the properties of the press and therefore on the quality of the manufactured prod-
ucts. To predict such effects by means of a coupled simulation it is essential to ex-
tend the machine model with respect to temperature effects. Hence, the influence
on the static properties of the press due to changes of the temperature was ex-
amined. It could be shown that the position of the bdc is affected by temperature
changes.
To determine the temperature curve of the press, temperature measurements
were executed during the operating time of the press up to a nearly steady thermal
condition. Apart from the temperature curve of the press, its influence on process
parameters, such as forming force and the position of the bottom dead centre, was
investigated. In order to quantify and consider the thermal influence in the ma-
chine model, parts were formed after different operating intervals and the force
path curve of these metal forming processes was captured for the different thermal
conditions at each operating interval. Therefore, piezo-electric load cells were in-
tegrated into the tool stages and thermocouples were applied to relevant press
components (Fig. 18.5). According to Fig. 18.5, for a permanent number of
strokes of 90 rpm the system reaches a nearly steady thermal condition after eight
hours of operation.
35.3
Temperature [°C]
30.0
27.3
24,7
Reference
22.0
0 1 2 3 4 5 6 7 8 9
1.Stage 2.Stage 3.Stage Press operating time [h]
Deep drawing Contouring Embossing
Con-rod left Oil tank left Guidance left Ram
Con-rod right Oil tank right Guidance right
In order to provide a general comparison between the force-time curve and the
temperature-time curve, the temperature of the right connection rod is shown in
Figure 18.6. The right connection rod has a direct influence on the embossing
stage as the process force at the embossing stage is mainly transferred by the right
connection rod. Between the temperature-time curve at the right connection rod
and the curve of the embossing force at different operating times correlations can
18 Consideration of the Machine Influence 409
Deep-drawing
Deep-drawing Contouring
Contouring Embossing
Embossing
35.0
Embossing force [kN]
Contouring force [kN]
61.75 36.75
Temperature [°C]
77.4 32.8
61.00 77.2 35.50 30.7
60.25 77.0 34.25 28.5
59.50 76.8 33.00 26.3
58.75 76.6 31.75 24.2
58.00 76.4 30.50 22.0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Press operating time [h] Press operating time [h]
Fig. 18.6 Maximum forces at the forming stages with increasing operating time
The differences of the position of the bdc due to the increase of the temperature
of the press can be measured by means of displacement transducers. The changes
of the bdc are affected by the expansion of the press components due to the in-
creased temperatures. The initial condition of the “cold” operating state is taken as
a reference. The difference between the bdc and the reference bdc defines the total
shift caused by changes in the temperature profile of the machine. Fig. 18.7 illu-
strates the position of the bdc with respect to the operating time for the stages deep
drawing, contouring and embossing. It was measured by incremental displacement
transducers.
410 B.-A. Behrens et al.
a 76.30 b
Fig. 18.7 Development of the shift of the bottom dead centre (bdc) related to the operating
time of the press
Fig. 18.7 shows that the shift of the bdc of all three stages is similar and con-
verges exponentially with time. The highest gradients of the shift curve in position
of the bdc occur during the start-up period of the press. As the operation time in-
creases the gradients of the shift curve in position of the bdc decrease until a near-
ly steady thermal condition is reached. This result corresponds to the temperature
measurements at the press components (Fig. 18.5). Thus, for each stage a shift of
the position of the bdc in forming direction occurs and converges towards a cer-
tain value with increased operating time.
Differences in the position of the bdc between initial condition at the start-up
and the steady thermal condition of the press after eight hours occur at all three
stages. At the deep drawing stage, the difference is 0.08 mm, 0.05 mm at the con-
touring stage and 0.03 mm at the embossing stage. Although these differences
mainly result from the temperature increase they also depend on the process. The
differences in the shift of the position of each bdc in relation to the position of the
bdc in “cold condition” are 0.1 % (deep drawing), 0.06 % (contouring), 0.04 %
(embossing). Thus, for the machine model a temperature coefficient of 0.04 % for
the steady thermal condition of the press was implemented. This value improves
the accuracy of the simulation of the press for all investigated processes by use of
the minimal requirements for the embossing stage. Therefore, no further settings
are required by the user.
Fig. 18.8 Design of the examined forming stages deep drawing, contouring and embossing
forces of other stages, which are applied simultaneously. The stages can be
mounted into the press and operated individually as well as together.
The design of the three-stage tool system was supported by a process simula-
tion. Therefore, simulation models based on the CAD data of the tools were gen-
erated and numerical experiments by means of the finite elements method were
executed. Fig. 18.9 illustrates the three simulation models for the forming stages
deep drawing, contouring and embossing.
The tools are designed to form sheet metals with a thickness of 1.0 mm. The re-
sults of the simulation show that this design leads to the required utilization of 80
% of the nominal press force, if a DC 04 deep drawing steel is used. Within the
simulation, the friction was set to a constant value of μ = 0.16. In the progression
of the project, a new friction law to consider the dependency of the friction on the
smoothing pressure and the contact pressure was applied.
In a further step, the drawing die was modeled elastically. The deep drawing
stage is illustrated in Fig. 18.10. The punch and the blank holder were still mod-
eled with rigid shell elements. Only the die was modeled with elastically-solid
elements to prevent an increasing simulation time. The elastic properties of the die
were taken into account in order to simulate the varying local contact conditions.
This allows an accurate calculation of the contact pressure.
Forming
direction Punch
(rigid)
Blank holder
(rigid)
z Die
x (elastic)
y
Fig. 18.10 Finite elements model of the deep drawing stage with the elastic die
For an improved description of the friction between the tool and the workpiece
during a sheet metal forming process the evolution of the surface conditions of the
workpiece has to be considered. The influence of the changing surface topology
on friction conditions has to be included in the FE simulation. The contact pres-
sure leads to a smoothing of the surface topology. The smoothed surface results in
a reduced friction coefficient. For this reason, a concept for the description of a
technical surface structure was used [11]. A workpiece with different smoothed
regions is shown in Fig. 18.11.
18 Consideration of the Machine Influence 413
Strong smoothing
Weak smoothing
A friction law µ = (PE, PK), where the friction coefficient µ is a function of the
smoothing pressure PE and the contact pressure PK, is implemented in
ABAQUS/Explicit via a user subroutine. This friction law considers the local sur-
face pressure as well as the transformation history of the sheet metal surface.
The difference in contact pressure at the contact surface leads to a different
component smoothing. The high pressures at the die radius (Fig. 18.12) contribute
to different friction conditions during the forming process.
35.0
Contact pressure [MPa]
15.0
0.0
sequentially and the entire cycle iterates. Fig. 18.13 illustrates the progression of
the coupled simulation between the process simulation in ABAQUS and the ma-
chine simulation in Matlab.
2. Step
Time-depending
Initial condition
course of the
Approximated
forming force
time-depending
course of motion 1. Step 3. Step
of the ram Process Machine
(forming machine
simulation simulation
ideal rigid)
Abaqus
Matlab
4. Step
New
cycle Convergence
criterion
reached? Correction of
Break the ram stroke
A cycle of the coupled simulation contains four sequentially running steps. The
ram motion at each stage, which is necessary for the required part depth assuming
an ideal rigid press, serves at first as boundary condition for the process simula-
tion. The process simulation calculates each forming stage one by one and then
provides the related coordinates for the nodes of the mesh of the tool punch, the
applied node forces and the related forming times in a text file (ASCII format).
The text file with the information about the force-time curves of the punch is sent
to the machine simulation in Matlab in a second step. By means of this input data
the relative shift between the bolster plate and the ram can be determined by the
machine simulation. The output of the machine simulation consists of way-time
curves for the process simulation, considering the deflection and the deformation
of the machine components. This result is then sent back to the process simulation
and serves as new input data for the process simulation. This process is repeated
until a certain convergence criterion is reached. The convergence criterion consists
of the comparison of the difference between the forming paths of two succeeded
iterations. This value was set to 0.01 mm. Fig. 18.14 illustrates a simulation,
where the convergence criterion was reached after three iteration cycles.
18 Consideration of the Machine Influence 415
Deep drawing
26,01 mm
26,01 mm
25,99 mm
26,75 mm
26,00 mm
25,00 mm
26,85 mm
24,61 mm
26,20 mm
24,62 mm
24,61 mm
26,23 mm
Contouring
Embossing
0 1 2 3 Iterations
Fig. 18.15 Experimental and simulated height of the part after the deep drawing stage
The height of the formed parts was measured with the optical measurement sys-
tem GOM ATOS II. In Table I, the results are compared and it is shown that the
results of the coupled enhanced simulation could be improved in comparison with
the enhanced decoupled and the previous coupled simulation. Using an elastic tool
416 B.-A. Behrens et al.
model, an improved friction law and the temperature factor leads to a more realis-
tic simulation but the simulation time is more than six times higher than the pre-
vious coupled simulation.
18.6 Conclusion
The numerical simulation of forming processes is an established method for the
analysis of common metal forming processes. By this method, the interactions be-
tween the process and the machine are not taken into consideration. Within the
experiments in the frameworks of this project it was shown that the machine has
an impact on the process and that this impact also affects the geometry and the
quality of the manufactured parts.
By means of the described coupling of the process simulation with a machine
simulation it became possible to take into consideration the influences of the ma-
chine. These influences mainly result in the shift of the bottom dead centre caused
by the deformation and the deflection of the ram as well as the tilting of the ram.
To improve the accuracy of the dimensions of the simulated workpieces these
influences have to be considered in the boundary conditions of the process simula-
tion. Therefore, an offline-coupling was used and validated by the results of expe-
rimental data. It was shown that by means of this offline-coupling the simulation
results approached the real geometrical dimensions of the workpieces.
With an increasing degree of detail of the models and an enhancement by fur-
ther influencing parameters, more realistic results with a higher compliance were
reached. However, these lead to higher calculation times and to the necessity of
further computing resources. It could be shown that the total shift of the bottom
dead centre for the analyzed workpiece is less than 1 mm. Particularly the thermal
influence of the machine on the process is very small. The differences of the posi-
tion of the bottom dead centre concerning a change in the temperature profile of
the machine range around a few hundredths of a millimeter.
Due to the fact that the influence of a forming machine on a forming process is
nominal, the coupled simulation has especially advantages for the tool set-up in
18 Consideration of the Machine Influence 417
forming machines for tools including embossing processes. The tolerances of the
height of an embossing punch are often limited to hundredths of a millimeter.
References
[1] Bäcker, V., Klocke, F., Wegner, H., Timmer, A., Grzhibovskis, R., Rjasanow, S.:
Analysis of the Deep Rolling Process on Turbine Blades using the FEM/BEM-
Coupling. In: IOP Conf. Series: Materials Science and Engineering, vol. 10 (2010)
[2] Großmann, K., Hardtmann, A., Wiemer, H., Penter, L.: An Advanced Forming
Process Model including the Interactions between Machine, Tool and Process. In:
Proceedings of 1st International Conference on PMI, Hannover, Germany, pp. 125–
132 (2008)
[3] Bogon, P., Roll, K.: Ein Ansatz zur Berechnung und Kompensation der elastischen
Werkzeugdeformation bei Ziehwerkzeugen. In: 30. EFB-Kolloquium Blechverarbei-
tung 02./03. März, Bad Boll (2010)
[4] Großmann, K., Wiemer, H.: Feasibilities of Advanced Forming Process Modelling
Considering the Interactions with Dies and Machines. In: Proceedings of Conference
New Developments in Sheet Metal Forming Technology, Fellbach bei Stuttgart,
Germany, pp. 311–350 (2008)
[5] Lohse, H., Marthiens, O., Helduser, S., Matthias, T., Behrens, B.-A.: Reglerauslegung
für hydraulische Tiefziehpressen. In: Ölhydraulik und Pneumatik. 3/2010/März, pp.
S.2–9. Vereinigte Fachverlage GmbH, Mainz (2010)
[6] Behrens, B.-A., Marthiens, O., Matthias, T., Helduser, S., Lohse, H.: Antriebs- und
Prozessoptimierung hydraulischer Tiefziehpressen mit Hilfe der gekoppelten Simula-
tion. EFB-Forschungsbericht NR. 291, Europäische Forschungsgesellschaft für
Blechverarbeitung (2009) ISBN 978-3-86776-325-7
[7] Schapp, O.: Gekoppelte Simulation von Maschine und Prozess in der Massivumfor-
mung. Dissertation, RWTH Aachen (2008)
[8] NN, DIN 55189, Ermittlung von Kennwerten für Pressen der Blechverarbeitung bei
statischer Belastung. Beuth-Verlag, Berlin (1988)
[9] Pestel, E.: Technische Mechanik. Band 1. BI Wissenschaftsverlag (1988) ISBN: 3-
411-03153-0
[10] Krimm, R.: Berechnung der lastabhängigen Maschinenauffederung zur Verkürzung
der Anlaufzeit neuer Transferwerkzeugsätze. Dissertation, Leibniz Universität Han-
nover (2006) ISBN: 3-939026-07-7
[11] Behrens, B.-A., Sabitovic, A.: Modelling of friction in deep drawing considering re-
versible sheet changes. In: Proceedings of the IDDRG Conference, Olofström, Swe-
den, pp. 125–132 (2008)
[12] Behrens, B.-A., Matthias, T., Czora, M., Poelmeyer, J., Ahrens, M.: Improving the
accuracy of numerical investigations of multistage sheet metal processes by coupling
a process FE analysis with the machine simulation. In: Proceedings of 1st Interna-
tional Conference on Process Machine Interaction, Hannover, Germany, pp. 133–138
(2008)
Chapter 19
Optimization of Tool and Process Design for the
Cold Forging of Net-Shape Parts by Simulation
19.1 Introduction
Cold forging offers the potential to manufacture, in large quantities, parts with a
high dimensional accuracy and surface quality as well as high strength due to
work hardening. Thus, the industry increasingly demands high strength cold
forged parts, which need no machining after the forming process. This kind of net-
shape production can help to save machining resources and material costs. For this
reason, achieving Net-shape accuracy has been an important target in recent re-
search activities on cold forging [1].
However, the high flow stress of the workpiece materials results in high form-
ing loads. As a consequence, the whole system, which consists of press and tool-
ing system, is subjected to considerable displacements and elastic deflections,
which can cause deviations in the workpiece dimensions. These interactions be-
tween process, press and tooling system make it difficult to achieve a high work-
piece accuracy solely based on process simulation [2]. Hence, achieving high
workpiece accuracy without extensive adjustment at the production start requires
the consideration of the elastic behavior of the whole system at an early stage, i. e.
within the phase of laying out the process in the FE simulation.
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 419–437.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
420 T. Kroiß and U. Engel
This can be done with a combined FE modeling of the cold forging process and
the characteristic of press and tooling system. Based on such a comprehensive, in-
tegrated simulation model, it would be possible to determine an adjusted process
and tool design so that the displacements and deflections of press and tooling sys-
tem are compensated. However, to permit a systematical determination of the ap-
propriate influencing parameters without try-out on the level of FE simulation, the
simulation model should be transformed into a parametric model of the entire
process behavior. This can be the base for a computer-aided optimization of the
values of parameters with influence on the workpiece dimensions so that high
workpiece accuracy can be achieved considering the interactions.
In the following, an approach is presented for the consideration of the men-
tioned interactions in the tool and process design by FE simulation through a
combined modeling of process, press and tooling system. To obtain a quick and, at
the same time, exactly enough combined modeling, tailored methods for the de-
termination and modeling of the entire deflection characteristic were developed in
a comprehensive approach. Built up on a set of simulations with the integrated
model under variation of several influencing parameters, a parametric process
model including an optimization algorithm was developed. To reduce the neces-
sary number of simulations for the generation of the parametric process model,
process knowledge was acquired with regard to the specific behavior of common-
ly used cold forging processes. The approach was developed for cold forging
processes in combination with stroke-controlled and force-controlled presses
respectively.
behavior were developed [7, 8]. Stroke-controlled presses usually show a non-
linear behavior with a clearance due to initial effects of the presses’ drive system
and a nearly constant stiffness at higher load. However, the deflection not only
affects the press itself but also the tooling system up to the punch and the pre-
stressed die. The deflection of these components leads to a change of the punch
position influencing the workpiece dimensions. To represent this behavior in the
FE process simulation, the deflection of all these mentioned press and tool com-
ponents has to be modeled.
Fig. 19.1 Tooling system with steel end stops on force-controlled press [9]
Furthermore, the tool components affected by the forming load are split up into
the top and the bottom tool plate. Again, modeling the mentioned components in
the FE simulation and loading them according to the real forming process leads to
a constant stiffness for each assembly. Both a reduced 2D axially-symmetric mod-
el and two 3D models were built in the case study in Section 19.4.2 using the FE
19 Optimization of Tool and Process Design 423
simulation program simufact.forming. The stiffness of the end stops can be direct-
ly incorporated into the FE process simulation as a spring element. For the tool
components in the flow of the forming force, an equivalent value as substitute
quantity is calculated based on the stiffnesses of the top and bottom assemblies.
This substitute stiffness is incorporated into the FE process simulation as a spring
element, too. In Section 19.4.2, the approach is carried out for a backward can ex-
trusion process on a force-controlled press.
Experimental FE simulation
setup models
2.0
top tool plate
mm
1.6 load cell
1.4
Vertical deflection
Fig. 19.2 Deflection characteristic of Komatsu-Maypress MKN 63/6A (a); Tooling system
– Experimental setup / FE simulation models of top and bottom assembly (b) [10]
The tooling system for the full forward extrusion process in consideration is illu-
strated in Figure 19.2 b. The left side of the figure shows the experimental setup.
424 T. Kroiß and U. Engel
The tooling system is supported by two bars and a ring at the center. The support
ring avoids deflection of the bottom tool plate. The pillar die-set contains a load
cell to get the forming loads in the experimental investigations. The die for the full
forward extrusion process, pre-stressed by two shrink rings, has a nominal initial
diameter of 12 mm, which is reduced to 6 mm after the shoulder. The die shoulder
has an angle of 45°. To determine the load-deflection curves in simulation, the
system is divided into two assemblies, which include the parts mounted on the top
and the bottom plate of the tooling system. Figure 19.2 b right shows the elements
of the simulation models. The two FE models contain all parts of the tooling sys-
tem in the flow of the forming force.
The simulation model of the top includes the tool plate, the load cell and the
adapter for the load cell. The model at the bottom contains the tool plate, the sup-
port ring, the support for the ejector and an insulating plate (E = 2,550 N/mm2).
The simulation models are 2D axially-symmetric. This simplification is feasible as
the plates are solely compressed due to contact with axially-symmetric parts with-
out bending. The two assemblies are loaded through rigid bodies, which have the
shape of the punch head (top) and the pre-stressed die (bottom).
The stiffnesses of the punch and the pre-stressed die are not included in this
calculation as they are considered as elastic bodies in the FE process simulation.
The gradient of the displacement of the rigid contact bodies vs. the load represents
the stiffness of the respective assembly. The resulting linear load-displacement
curves are shown in Figure 19.3. The assembly of the top tool plate has
a stiffness of Ctool_top = 1,851 kN/mm, the bottom plate a stiffness of
Ctool_bottom = 5,336 kN/mm. As the two stiffnesses behave like two serially ar-
ranged spring elements, their reciprocals add up. This leads to a stiffness of
Ctool = 1,374 kN/mm for the whole tooling system.
600
Ctool_bottom = 5336 kN/mm
kN
400 Ctool_top = 1851 kN/mm
300
Force
200
C tool = 1374 kN/mm
100
0
0 0.05 0.10 0.15 0.20 mm 0.30
Deflection
The stiffnesses of the tooling system and the press are serially arranged as well.
Hence, the reciprocals of Ctool = 1,374 kN/mm and Cpress = 265 kN/mm add up to
the reciprocal of the whole system. This leads to a stiffness for the whole system
of Cpress_tool = 222 kN/mm [10].
19 Optimization of Tool and Process Design 425
The deflection characteristic was incorporated into the FE simulation of the full
forward extrusion process (Fig. 19.4 a). The full forward extrusion process was
modeled 2D axially-symmetric in the FE simulation program simufact.forming. It
is a combined simulation of workpiece, die and the pre-stressing system of the die.
Thus, the model includes the influence of the radial deflection of the die on the
workpiece dimensions. The punch is also an elastic body. Its stroke (60 mm) is
applied by a rigid body at the top of the elastic punch.
Experimental Validation
The workpiece lengths for the experiments and simulations are shown in
Figure 19.5. The average lengths of the experiments are 22.81 mm (+0.11 mm,
-0.15 mm) for 20MnB4 and 21.73 mm (+0.18 mm, -0.21 mm) for C35. The differ-
ence in length between 20MnB4 and C35 is 1.08 mm. The workpiece lengths re-
sulting from the integrated FE simulation model are 22.83 mm for 20MnB4 and
21.75 mm for C35. Hence, the difference in length between the two workpiece
materials is also 1.08 mm. The deviation compared to the experiments is only 0.02
mm for both materials. Only a small difference in average forming loads in the
BDC of 3 kN (20MnB4) and 6 kN (C35) can be observed compared to the expe-
riments. The process simulations with the rigid model lead to similar forming
loads. However, the resulting workpiece lengths of 25.90 mm (20MnB4) and
25.57 mm (C35) differ considerably from the experiments. The respective devia-
tions in length are 3.09 mm (20MnB4) and 3.84 mm (C35). The pure process si-
mulation (0.33 mm) cannot sufficiently represent the difference in length between
the workpiece materials compared to the real process (1.08 mm). The integrated
FE simulation model is able to accurately represent the interactions between forg-
ing process, tool and machine with influence on the workpiece dimensions. The
modeling of the whole characteristic with a shift and a substitute stiffness leads to
a reduction of the deviation between experiment and simulation from up to 3.84
mm (17.7 %) to 0.02 mm (0.1 %) compared to the rigid simulation [10].
27
20MnB4 25.90 C35 Experiment
mm 25.57
Integrated FE
25
Workpiece length
simulation model
24 FE process simulation
23 22.81 22.83
22 21.73 21.75
21
0˜
149 kN 146 kN 152 kN 205 kN 199 kN 208 kN
Fig. 19.5 Workpiece lengths depending on workpiece materials and forming force in BDC;
comparison of experiments to Integrated FE model and pure process simulation [11]
Fig. 19.6 Backward can extrusion process (a); Integrated FE simulation model of backward
can extrusion process on force-controlled press (b) [9]
Fig. 19.7 FE models of top tool plate (a), bottom tool plate (b) and end stops (c) [9]
1400
kN
C tool_bottom = 4218 kN/mm
1000
800 Ctool_top = 2189 kN/mm
Force
600
400
200
C tool = 1441 kN/mm
0
0 0.05 0.10 0.15 0.20 0.25 mm 0.35
Deflection
Experimental Validation
The results from the integrated FE simulation model were compared to backward
can extrusion experiments. To be able to check the correctness of both stiffnesses
Cstop and Ctool of the FE simulation model, the influencing parameters press force
(force on end stops) and forming force were varied in the experiment as well as in
the simulation. In the experiment, the press force is varied by the press control.
The forming force is varied by using two different workpiece materials 100Cr6
(1.3505) and 16MnCr5 (1.7131). Both materials noticeably differ in their flow
stresses, which leads to different punch forces. In the simulation, the forming
force is varied using flow stress data of the two workpiece materials. The punch
force in experiments is measured with a piezo-electric load cell Kistler 9091A.
The experimental investigations on the backward can extrusion process were
carried out on the hydraulic press Lasco TSP 100. Five workpieces were formed at
each parameter setting. The press was used in force-controlled mode at a punch
velocity of 50 mm/s. The press forces were set to 391, 612 and 867 kN for
16MnCr5 (potentiometer settings of 50, 75 and 100 %). The experiments with
100Cr6 were carried out at a press force of 612 kN. In the simulation, the press
forces were set as termination condition, too. The workpieces in the experiments
were coated with a phosphate conversion layer as lubricant carrier, which soap
was applied on for lubrication. The friction factor in the FE simulation was identi-
fied numerically. The experiments were carried out with tool and workpieces at
room temperature. The four parameter settings were also assigned to the integrated
FE simulation model. To show the improvement in simulation accuracy, addition-
al FE simulations with rigid tool (Ctool → ∞) and end stops (Cstop → ∞) / elastic
punch and rigid tool and end stops / rigid punch, respectively, were run. The pre-
stressed die remained elastic in all simulations. The accuracy of the simulation
models is evaluated using the workpiece dimension thickness at bottom
(Fig. 19.9). It is a characteristic for the actual punch position including the
deflections.
Figure 19.9 shows the workpiece dimensions resulting from the experiments
and the FE simulations. The integrated FE simulation model, considering the
whole deflection characteristic, achieves the highest accuracy compared to the ex-
periments for almost all parameter settings. However, at a press force of 867 kN
(16MnCr5), the simulation model with rigid tool and elastic punch shows a similar
deviation from the experiments (0.03 mm) compared to the entirely elastic model
(0.04 mm). The reason for this is the compensation of the tool deflection in the
flow of punch force by the increasing compression of the end stops at high press
forces. At low forces on the end stops due to low press forces, mainly the elastici-
ty of the tool in the flow of the forming force is crucial for the total deflection.
Hence, the deviation of the rigid models compared to the experiments is consider-
able at a low press force of 391 kN (0.20 and 0.31 mm). The entirely elastic model
shows a significantly lower deviation in this case (0.03 mm). The simulation mod-
els with rigid tool or additionally rigid punch lead to the same workpiece dimen-
sions at different press forces because the increasing compression of the end stops
at higher press forces is not represented.
430 T. Kroiß and U. Engel
4.2
Simulation
Experiment Tool elastic, punch elastic
mm
Tool rigid, punch elastic
4.0
Thickness at bottom
Tool rigid, punch rigid
3.8
3.6
0
16MnCr5 16MnCr5 16MnCr5 100Cr6
391 kN 612 kN 867 kN 612 kN
Fig. 19.9 Thickness at bottom of backward can extruded part – comparison between expe-
riment and simulation models depending on workpiece material and press force [9]
The high flow stress of 100Cr6 leads to a higher forming force, a higher punch
deflection and thus to a higher remaining thickness at bottom and a smaller depth
of can. The integrated FE simulation model shows the highest accuracy for this pa-
rameter setting, too (0.05 mm of deviation). However, even the simulation with ri-
gid tool and punch results in a bigger thickness at the bottom for 100Cr6 compared
to 16MnCr5 due to the radial elastic deformation of the pre-stressed die. An elastic
modeling of the punch already helps to increase the simulation accuracy compared
to a rigid punch as its long and thin shape contains a high compliance [9].
Table 19.1 Variation levels of influencing parameters related to center point; linear primary
effects on target values, calculated by parameter variation from low to high level
Variation levels Effect on Effect on Effect on
Influencing
related to DT1 DT2 LT
parameters
center point in mm in mm in mm
DI1 ± 0.30 mm 0.60 0.00 -2.55
DI2 ± 0.30 mm 0.00 0.60 -2.27
II ± 50 % -0.12 -0.08 0.96
SI ± 0.50 mm 0.00 0.00 2.97
19 Optimization of Tool and Process Design 431
The following three target values were defined to specify the dimensions of the
full forward extruded part (Fig. 19.4 b): workpiece length LT = 22 mm, starting di-
ameter DT1 = 12 mm and final diameter DT2 = 6 mm. Four influencing parameters
with considerable influence on these target values were identified. These are
available for the following optimization of the target values: The die diameters
above (DI1) and below (DI2) the die shoulder, the punch position SI and the relative
interference fit II of the pre-stress rings of the die. The dimensions of the billet are
constant. To be able to consider quadratic process behavior each of the influencing
parameters was varied on three levels based on a center point. To detect all poten-
tial interactions a full factorial experimental design was applied. This leads
to an amount of 34 = 81 different parameter combinations and, thus, 81 simula-
tions with the integrated FE model. The spring stiffness in the model is
Cpress_tool = 222 kN/mm. The press clearance of 0.4 mm is included as a constant
shift of the punch. The friction factor is m = 0.08. Based on a sensitivity analysis,
the variation range for the influencing parameters was determined as shown in
Table 19.1.
The data points for the three target values permit the calculation of the linear
primary effects and interactions. Only the low and high variation levels of the pa-
rameters are required for these calculations. The primary effects (Table 19.1)
show a dependency of the final workpiece length LT on all influencing parameters.
A change of the die diameters affects LT due to the constancy of volume. II also
contributes to all target values by influencing the effective die diameters. The pri-
mary effects of the die diameters (DI1,2) on the workpiece diameters (DT1,2) nearly
agree with the variation range of the influencing parameters from the low to the
high level (0.6 mm). Varying one die diameter has no visible influence on the oth-
er one. SI shows no noticeable influence on the workpiece diameters in the inves-
tigated variation range.
0.5
0.42
0.33
mm
0.25
0.3 0.20
Interaction
0.2 0.09
0.1
0.01
0 LT
DI1*DI2 D *I DT2
I1 I DI1*SI D *I DT1
I2 I DI2*SI
I*S
I I
Fig. 19.10 Absolute values of the interactions between two influencing parameters related
to the target values; calculation based on parameter variation on low and high level [10]
Sequential Approach
Standardized Length
The approach uses the specific process characteristics found above. In addition,
process knowledge concerning the geometric relations of the workpiece dimen-
sions is applied. The objective of this analytic approach is the elimination of the
influence of the die diameters (DI1,2) and the interference fit II on the workpiece
length LT, related to a reference point. This corresponds to the calculation of a
standardized workpiece length, which only depends on the punch position SI. Af-
ter this, according to the independency of the workpiece diameters of the punch
position, the required punch position to achieve the specified workpiece length can
be determined by linear interpolation. In Figure 19.5.2, the resulting lengths of the
81 simulation runs are plotted. The systematic permutation of the influencing pa-
rameters in this full factorial experimental design leads to a periodic behavior of
the workpiece lengths. In the full factorial investigation, the punch position is va-
ried every 27 simulations. Hence, the 81 single standardized lengths should form a
step function with the steps at the changes in the punch position.
According to the observed linear behavior without interactions concerning the
workpiece diameters, the required die diameters DI1,2 and interference fit II to
achieve the specified workpiece diameters can be determined by means of their
434 T. Kroiß and U. Engel
FE simulation results
mm
22 Ls_low
Length
Ls_high
20 Ls
18
Calculation of standardized
length Ls with 7 data points
0
1 9 18 27 36 45 54 63 72 81
Consecutive numbering of FE simulations
Fig. 19.11 Workpiece lengths resulting from FE simulations based on full factorial experi-
mental design; standardized length Ls calculated based on 7 data points
Seven data points marked with grey dots in Figure 19.11 are required as the ba-
sis for the analytic approach: Based on a reference point at the low punch level
(No. 9), the interference fit II and the two die diameters DI1,2 are varied separately
between the minimum and the maximum variation level. The influenced work-
piece volume is calculated using the change in workpiece length and the actual
workpiece diameter below the die shoulder. The die diameters are then also varied
on the high punch level (reference point No. 63) because the workpiece volume
affected by the diameters depends on the proportionate length above and below
the die shoulder and, with that, on the punch position.
19 Optimization of Tool and Process Design 435
In the present case study, the 81 workpiece lengths resulting from the simula-
tion runs with of the full factorial design are the specified target values for LT. De-
pending on the respective values of die diameters and interference fit, a back
calculation from the influenced workpiece volume to the change in workpiece
length is made for each of the 81 lengths. The change is subtracted from the target
length for the high and low punch level. This leads to Ls_high and Ls_low respec-
tively (Fig. 19.11). To consider the dependency of the influenced workpiece vo-
lume on the punch position the two curves are weighted depending on the punch
position. The resulting standardized length Ls in Figure 19.11 clearly shows the
required step characteristic. The die diameters and the interference fit to reach the
specified workpiece diameters are determined by adding up the effects, since no
relevant interactions were observed concerning the diameters.
To assess the accuracy of the presented approach it is compared to the simula-
tion results of the full factorial experimental design. The workpiece lengths
represent the target values. The application of the standardized length leads to the
corresponding approximated values for the punch position. Ideally, the resulting
punch position should match the values set in the FE model. However, there are
some deviations. The workpiece lengths corresponding to the respective punch
position were calculated in a reverse calculation for comparison with the 81 simu-
lation results. In addition, for a comprehensive evaluation, the full factorial
approximation and a linear process model neglecting the interactions are also
compared. The linear model is based on a single variation of each influencing pa-
rameter on a low and a high level from a center point. It requires eight simulation
runs compared to seven needed for the standardized length. The function values of
the process models are compared to the workpiece lengths resulting from the si-
mulation. Accuracy is evaluated using the absolute value of the maximum error
and the sum of squared errors.
0.4
of max. error
Standardized length
0.3 1.5 (7 simulations)
Fig. 19.12 Errors of the three process models compared to the FE simulation results based
on the full factorial experimental design for target value workpiece length
Figure 19.12 shows the results of this comparison. As expected, the full factori-
al approach (81 points) has the lowest deviations from the original data points.
The maximum error is 0.06 mm, the sum of squared errors is 0.03 mm2. The linear
process model based on the addition of effects has considerably higher error
436 T. Kroiß and U. Engel
values of 0.50 mm and 2.96 mm2. Obviously, neglecting the interactions is not va-
lid. The standardized length represents the real behavior in a much better way.
Here, the maximum error is 0.19 mm, the sum of squared errors is 0.26 mm2. This
means a reduction by 62 % and 91 %, respectively, compared to the linear process
model, which even requires one more simulation run. Thus, the presented ap-
proach includes a comprehensive modeling of the entire process behavior with an
extremely reduced number of required simulation runs. It has the potential to be
applied to further quasi-stationary cold forging processes.
References
[1] Balendra, R.: Net-Shape-Forming: State of the Art. J. Mater. Process Technol. 115,
172–179 (2001), doi:10.1016/S0924-0136(01)00812-3
[2] Kroiss, T., Völkl, R., Engel, U., Geiger, M.: Modeling of Process-Tool-Machine Inte-
ractions in Cold Forging. In: Proc. 1st Int. Conf. on PMI, Hannover, Germany, pp.
125–132 (2008)
[3] Behrens, B.A., Matthias, T., Czora, M., et al.: Improving the accuracy of numerical
investigations of multi-stage sheet metal processes by coupling a process FE analysis
with the machine simulation. In: Proc. 1st Int. Conf. on PMI, Hannover, Germany,
pp. 133–139 (2008)
[4] Brecher, C., Schapp, L., Paepenmüller, F.: Gekoppelte Simulation von Presse und
Massivumformprozess. Berücksichtigung (nicht-)linearen Maschinenverhaltens in der
Umformsimulation. wt Werkstattstechnik Online 96, 526–529 (2006)
[5] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Faster to sound parts by ad-
vanced forming process simulation – advanced forming process model including the
elastic effects of the forming press and tool. Steel Res. Int. 78, 825–830 (2007)
[6] DIN 55189, Machine tools – determination of the ratings of presses for sheet metal
working under static load. Beuth, Berlin (1988)
[7] Chodnikiewicz, K., Balendra, R., Wanheim, T.: A new concept for the measurement
of press stiffness. J. Mater. Process Technol. 44, 293–299 (1994), doi:10.1016/0924-
0136(94)90442-1
[8] Arentoft, M., Wanheim, T.: A new Approach to determine Press Stiffness. CIRP
Ann. 54(1), 265–268 (2005), doi:10.1016/S0007-8506(07)60099-7
[9] Kroiß, T., Engel, U., Merklein, M.: Simulation-Based Determination of Deflection
Characteristic of Tooling System and its Modeling in FE Simulation of Cold Forging
Process. Steel Res. Int (special ed. Met. Form 2010) 81, 318–321 (2010)
[10] Kroiß, T., Engel, U., Merklein, M.: Process Modeling In Cold Forging Considering
The Process-Tool-Machine Interactions. In: Proc. 10th Int. Conf. Numer. Methods
Ind. Form Process, AIP Conf. Proc., vol. 1252, pp. 312–319 (2010)
[11] Kroiß, T., Engel, U., Merklein, M.: Modeling of the Behavior of a Cold Forging
Process Considering the Deflection of Tooling System and Press. In: Proc. 2nd Int.
Conf. on PMI (2010) ISBN: 978-0-9866331-0-2
[12] Spellucci, P.: Numerische Verfahren der nichtlinearen Optimierung. Birkhäuser, Ba-
sel (1993)
Chapter 20
Interaction Effects between Strip and Work
Roll during Flat Rolling Process
Abstract. During the flat rolling process (cold or hot), the strip flatness and
thickness profile are highly influenced by the interaction effects between strip and
work rolls. To understand and analyze these effects a new modeling concept was
developed. Within this concept, the tool simulations are separated from the
process simulation. With the help of an automatic coupling module, the influences
of the tool effects are realized within the process simulation. With this modeling
concept, three types of interaction phenomena are studied and validated using ex-
periments: elastic roll effects during the cold rolling process, work roll thermal ef-
fects during the hot rolling process and tribological effects (abrasive wear) on the
process simulation. It was also shown that, compared to the single FE model,
this modeling concept is relatively faster and suitable for large 3D models without
losing the quality of the predicted results.
20.1 Introduction
During the flat rolling process, the dimensional tolerances and the mechanical-
metallurgical properties of the rolled products become more stringent. Therefore,
computational models are being used to investigate the physical aspects, such as
thermo-mechanical events of the strip, mechanical deformation and thermal evolu-
tion of the work rolls, and tribological effects, such as wear. Conventional numer-
ical modeling of such processes assuming plane strain conditions (2D) may not be
sufficient since the contact pressure distribution along the strip width may have a
non-uniform profile due to the elastic deformation of the work rolls. Depending on
the roll elastic deformation, strip flatness profile changes, i. e. crowns, will origi-
nate (see Fig. 20.1). The modeling of such processes using a conventional single
FE model and considering all the above-mentioned effects, will not only compli-
cate the computation (deformable – deformable contact situation has to be
handled) but also affect the computation time. This situation is even more compli-
cated when contact between elastic tools (in the case of backup rolls) has to be
considered.
Significant effort was devoted to the study of mechanical and thermo-
mechanical effects in the strip and work rolls using the numerical methods.
Particularly, A. Kainz [1] simulated a skin pass cold rolling process (in 2D) consi-
dering elastic effects of the work rolls using the commercial FE program
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 439–458.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
440 S. Puchhala, M. Franzke, and G. Hirt
Fig. 20. 1 Strip profile quality problems during cold rolling process
For the modeling of a hot rolling process, Tseng [6] simulated steady state heat
transfer in the work roll using a two-dimensional Finite Difference Method. Using
an empirical approach, he included the heat generation term due to plastic defor-
mation and friction between roll and strip. On the other hand, Hsu and Evans [7]
conducted a two-dimensional FE analysis of the rolling process considering a
steady state heat transfer condition. C.G.SUN and S.M.HWANG [8,9] presented
integrated models considering the thermo-mechanical effects both in roll and strip
for 2D and 3D. They presented a coupled model for the hot rolling simulation,
where they considered the steady state heat transfer using 2D models for the rolls.
They also considered the elastic deformation of the work rolls. LEE and Hwang
[10] presented a similar model for the computation of the thermal effects for the
work rolls. These models have been validated with the experimental results and
are good for the estimation of the thermal behavior of the rolls during a hot rolling
process. The only untouched area in those models is the analysis of bi-directional
interaction effects between strip and rolls.
At the same time, during cold and hot rolling processes, the tribological events,
particularly abrasive wear of the work rolls, directly affect the rolled strip surface
quality and may hinder the flatness profile [11, 12]. Many investigations have
been carried out to understand the wear mechanism of the rolls during rolling
processes [13,14,15,16]. All of these studies are related to the modeling of wear
using analytical methods or empirical models. As far as we know, offline calcula-
tions of precise local work roll wear during hot or cold rolling have not yet been
noticed.
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 441
The objective of this work is to develop a validated simulation tool for the
analysis of the interaction effects between the work rolls and the strip during a flat
rolling process. For that, a novel FE modeling concept was developed considering
three different aspects, which influence the strip quality, i. e. elastic effects of the
work rolls during a cold rolling process, thermo-mechanical effects of the work
rolls during a hot rolling process and work roll tribological effects (mainly abra-
sive wear) during cold and hot rolling processes. In the following sections, each of
these aspects is discussed in detail. In the next section, a detailed description of the
de-coupled modeling concept (applicable for the modeling of all three aspects
mentioned above) is explained. The details of abrasive wear models and their im-
plementation are also discussed within the next section. In the section “modeling
and analysis of a cold rolling process with elastic rolls”, the cold rolling experi-
mental setup and the optical deformation measuring method is explained. Later,
the FE modeling of the cold rolling process using the new modeling concept is de-
scribed; and the simulation and experimental results are given in a sub-section. In
the section “modeling and analysis of hot rolling process”, an identical structure as
mentioned above is maintained. In the following section, the modeling of work
roll wear is described. At the end, a detailed discussion on the overall work (mod-
eling concept and the results) is given.
FE-Model for
FE-Model for
process simulation
elastic tool simulation
(With rigid tool)
Deformation*
damping
Increment
Contact-pressure
Within this concept, the work roll is fixed in all DOF (Degrees Of Freedom)
and the boundary conditions, such as heat flux at roll-strip contact, cooling
zones are defined as rotating (with the rolling velocity) boundary conditions (see
Fig. 20.3).
The conductive heat flux (at roll - strip contact zone) is calculated using the
state variables (contact pressure and strip temperature) provided by process simu-
lation (Eq. 20.1).
q = hgap (Tstrip − Troll ) (20.1)
The gap conductance hgap is defined as a function of contact pressure, roll rough-
ness and roll hardness. Detailed explanation of this model is given in the section
“modeling and analysis of hot rolling”. In the same way, the convective heat
fluxes (due to water cooling and air cooling) are calculated using Equations 20.2
and 20.3.
q = hwater (Troll − Twater ) (20.2)
pv dt = H μτ v dt
k k
w= t t t (20.7)
H
– D = τ t vt Z = μ D (20.8)
k (Z acc,e ) n
– we (r ) =
H pe (r , t )vslid (r , t )dt where Z acc = Z
r =1
e (20.9)
Table 20.1 Process and geometric parameters for the cold rolling process
Parameter Value
Initial strip thickness 1.67 mm
Exit strip thickness 1.65, 1.48, 1.22, 1.20 mm
Work roll radius 300 mm
Work roll velocity 200 mm/sec
Friction coefficient 0.3
Strip tension (front and back) 30 kN
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 445
To obtain a deep understanding of the plastic flow of the strip and elastic de-
formation of the work rolls, implicit FE simulations were performed utilizing the
newly-developed concept. As described in the implementation section, the process
simulation model consists of a strip with elastic-plastic material description; and
the rolls are considered as rigid bodies. Four different thickness reduction cases
were simulated. Since the contact length for each case is not higher than 10 mm, at
least 20 elements (with 0.5 mm length for both strip and roll) are necessary for the
detailed analysis of plastic flow within the roll gap (Fig. 20.4).
Fig. 20.4 FE mesh of the strip and work rolls for cold rolling simulations
To validate the simulation results obtained by the new modeling concept and to
understand the underlying elastic-plastic strip material flow and elastic work roll
deformation, cold rolling experiments were conducted on the rolling mill installed
at IBF (Fig. 20.5 left). High-strength steel strips (ZStE340) with initial thicknesses
of 1.67 mm were rolled to four different thickness reductions (Tab. 1).
Fig. 20.5 Rolling mill installed at IBF (left), rolling mill deformation in rolling direction
(top right) and in thickness direction (bottom right)
446 S. Puchhala, M. Franzke, and G. Hirt
The rolling was conducted at a relatively low velocity 500 mm/s (conventional
cold rolling velocity ranges 5 m/s up to 10 m/s) as the rolling mill limit is reached.
For each case, the rolling force and the exit-strip thickness were measured. Fur-
thermore, the elastic deformation of the whole mill was also measured using an
optical measuring method TRITOP [38]. As shown in Figure 20.5, coded (fixed
reference marks) and uncoded marks were placed on the rolling mill. At static
loading conditions, free-form photographs were taken using a high resolution
camera. With the help of the TRITOP software, the deformation of the uncoded
marks was measured (see Fig. 20.5). As expected, the maximum deformation
(0.12 mm) was observed in thickness direction whereas in the rolling direction the
maximum deformation was observed as 0.08 mm.
strain along the strip width is not uniform (for high thickness reduction case),
which confirms the non-uniform exit strip thickness for a 28.2 % thickness reduc-
tion case. Whereas for a low thickness reduction case, the plastic strain along the
strip width is relatively uniform, this will lead to a uniform thickness reduction.
For the analysis of the plastic flow of the strip the comprehension of the shear
stress distribution is very important. As shown in Figure 20.7 (left), there is a small
“no-slip-zone” in the 28.2 % thickness reduction case. This means that there will be
no significant thickness reduction in that region, whereas in a 14 % thickness re-
duction case (Fig. 20.7 - right) there is no such zone. From the Figures, it is clear
that the plastic flow using the new modeling concept is identical to the single FE
model. It can also be observed that the rolling force in the new model is smaller
than in the single FE model, which may be due to a lower contact length in the new
model. This directly depends on the frequency of the activation of the elastic roll
model. This situation can be improved by increasing the activation of the tool simu-
lation model, which may lead to a higher computation time. A compromise has to
be found between these two phenomena to achieve reliable results. Detailed expla-
nation regarding the frequency of the activation of tool simulation is given in [20].
Any change in the work roll profile will directly affect the end strip quality. In
order to control the required output strip specifications a careful analysis of the
roll deformation is the key. Figure 20.8 shows the stress distribution in the strip
thickness direction for a 23 % thickness reduction case predicted using
LARSTRAN and ABAQUS [36].
In both cases, the maximum compressive stress is about 360 N/mm², which is
considerably lower than the yield stress of 860 MPa of the roll material. For the
de-coupled simulation, a total of 200 coupled cycles were used and the computa-
tion time is up to 2 times faster than the single FE model. The computation time
for the single FE model was about 614 minutes, whereas with the new concept
(using DOELTOSIM module), it required only 316 minutes. From Figure 20.8 it
is clear that the developed modeling concept is reliable and faster. Further details
of the elastic roll influence on cold rolling process are explained in [19, 20].
The quality of the measured temperature using the pyrometer lies in the defini-
tion of the emission coefficient. The pyrometer was trained to obtain the precise
coefficient using an additional experiment, where a specimen connected with a
thermo-couple and the temperature was measured using both systems. Once both
systems were measuring the identical temperature, the pyrometer was used in the
real experiments. For the analysis of the strip temperature within the roll gap ex-
tensive numerical simulations were conducted.
In order to study the effects of the transient heat transfer phenomenon at the roll
gap a good understanding of the process parameters is critical. For that purpose,
numerical simulations were conducted using the newly-developed modeling con-
cept. Within this concept, the process simulation was modeled using thermo-
mechanical strip and the work rolls were modeled as rigid bodies. Although the
roll was modeled as a rigid body the solver will consider the temperature distribu-
tion of the roll for solving the heat transfer at the roll gap. The work roll tempera-
ture will be updated using an external work roll model.
For the detailed analysis of the heat transfer phenomenon at the roll gap, the de-
finition of gap conductance is most critical. Tseng [21, 22, 23] reviewed various
analytical models for the definition of gap conductance and mentioned two sophis-
ticated relations. The first is based on the analysis by Mikic [24] (Eq. 20.10), the
second was semi-empirically derived by Yovanovich et. al. [25] (Eq. 20.11). In
both Equations, kh is the harmonic mean thermal conductivity in W/mm K, Ra is
the combined arithmetic surface roughness in mm, P is the contact pressure and M
is the hardness of the softer material. For hot rolling, the analysis of thermal con-
tact conductance of the interface should include the conductance of oxide scale.
To serially link the oxide layer with interface contact conductance, the conduc-
tance in Equations 20.11 and 20.12 can be modified as shown in Equation 20.13,
where the underline signifies the contact conductance including the oxide scale
conductance, which equals k0/δ0. Here, k0 and δ0 are the thermal conductivity and
450 S. Puchhala, M. Franzke, and G. Hirt
thickness of oxide scale respectively. The properties of the oxide layer during the
hot rolling of various grades of steels were studied in detail by the large number of
researchers [26, 27, 28]. Conventionally, high pressure jets are used to desiccate
between the stands so that the scale layer is kept relatively thin. A 10 μm oxide
layer is consistent with most measurements [27, 28]; and the hardness of the oxide
layer was observed as 980 MPa (HV100). A user sub-routine was developed (and
implemented in a model program for PEP) to define the gap conductivity accord-
ing to Equation 20.13, where the local contact pressure is extracted from the
process simulation. With this, a detailed analysis of heat transfer between work
roll and strip was conducted.
Fig. 20.10 Boundary conditions for work rolls during hot rolling process
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 451
Fig. 20.11 Strip shear stress and temperature distribution in hot rolling
The first one is the influence of elastic tool effects and the second the definition
of gap conductance. The shear stress as shown in Figure 20.11 (top left) is pre-
dicted for two different rolling velocities, both using rigid and elastic rolls for 21
% thickness reduction case. For a high velocity, low shear stress is predicted and
vice versa. This may directly depend on the strip temperature distribution. At
the same time, the influence of the elastic work roll on the plastic deformation of
the strip is considerably low. However, the influence of the definition of gap
452 S. Puchhala, M. Franzke, and G. Hirt
conductance on the plastic deformation has a considerable effect (Fig. 20.11, top
right). On the entry side, the shear stress distribution has identical behavior in both
models whereas on the exit side, the pressure-dependent model predicts high shear
stresses. This may be due to the tendency in the strip temperature in both models.
Fig. 20.12 Surface temperature distribution of the work roll for first rotation (left) and after
100 sec (right)
For the clear understanding of the heat transfer between strip and work roll, a
pressure-dependent gap conductance was defined according to Equation 10. Fig-
ure 20.11 (bottom left) shows the predicted gap conductance (for two different
friction coefficients) for a node as it passes through the roll gap. From the Figure,
it is clear that the gap conductance has a profile identical to the contact pressure
distribution and reaches its maximum at the exit side of the roll gap, where the
normal compressive contact pressure has its maxima. Figure 20.11 (bottom right)
makes it clear that the strip temperature predicted using both models before and
after the roll gap matches well with the measured temperature (755 °C). An impor-
tant phenomenon can be observed here: the strip temperature along the roll gap
has a large deviation (pressure-dependent model predicts up to 230 °C lower than
the constant gap conductance model- see Fig. 20.11) in both models. Neglecting
this phenomenon during the modeling of a hot rolling process may lead to serious
problems since the metallurgical changes directly depend on the forming tempera-
ture and others [30, 31].
Fig. 20.13 Temperature distribution at surface and inside the roll (left - conventional cool-
ing, right - double cooling)
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 453
Fig. 20.14 Temperature distribution at the surface and inside the roll
A detailed analysis of the thermal evolution of the work rolls is very important
since the life and the wear of the work rolls can be controlled by improving the
cooling system [32, 33]. The proper cooling may reduce the induced thermal
stresses and, from that, the abatement of the thermal fatigue may be achieved. To
understand this phenomenon, a new rotating boundary model was developed and
implemented within the FE program PEP [39] and LARSTRAN. For the qualita-
tive numerical analysis, the hot rolling lab experiments were modeled assuming
different cooling strategies. Figure 20.12 shows the temperature evolution at the
surface of the roll for the first rotation (left) and after 10 seconds of rolling (rolling
velocity 1 m/sec) for three different cooling strategies (conventional, double cool-
ing and late cooling [33, 34]). The schematic diagram of the first two strategies is
shown in Fig. 20.13 and the late cooling means that the cooling starts at 130°).
From the first rotation curves for all cooling methods it can be observed that due
to insufficient cooling a small amount of heat (about 15°C) is retained in the work
rolls. After 100 seconds of rolling, the temperature distribution has a clear differ-
ence (up to 30 °C) between double and conventional cooling methods. The steady
state thermal evolution (Fig. 20.13) confirms that the thermal gradient produced
during one cycle is restricted to the surface of the roll. One more important phe-
nomenon can be observed in these figures, i. e. the temperature on the surface at
some angular positions is cooler than the core and in other locations hotter than
the core. This leads to tensile stresses at cooler locations and compressive stresses
at the hotter locations, which may cause the fire-cracks on the surface of the rolls.
To understand the cooling efficiency and the resulting thermal effects the histo-
ry of the temperature penetration for various cooling strategies has to be ex-
amined. Figure 20.14 and 20.15 show the thermal evolution of the work rolls for
four different cooling methods at two different locations (9 and 16 mm inside).
From the Figures it is clear that the double cooling phenomenon slows down the
penetration of the thermal gradient and less heat is retained in the rolls after each
rotation.
The most important tribological effect, which influences the exit strip thickness
profile is work roll wear. During a cold rolling process, due to very high quality
tolerances, macroscopic wear of the work rolls may not be observed since the rolls
are changed frequently after a defined roll surface roughness. During hot rolling
process (particularly the continuous strip casting mill), however, the work roll
wear is more significant and mostly not uniform along the strip width [12,13].
Figure 20.16 shows a typical wear profile of a work roll [35], where the wear at
the edges of the strip is considerably higher than the middle. This might be be-
cause of the non-uniform temperature (particularly cooled edges) of the incoming
strip [34], which results in high contact pressure at the edges. The middle wear,
however, is relatively homogeneous. For the analysis of the wear profile, various
analytical wear models have been developed and are also being used for on-line
observations [11, 12, 13]. The major drawback of most of the models is that they
are not suitable for the prediction of local effects but useful to get the average val-
ue (along the strip width direction) of the wear [13, 14]. It is also necessary to un-
derstand the phenomenon of local wear for the better planning of the scheduling
and roll change.
Fig. 20.16 Typical work roll wear profile during hot rolling process (Stand 5)
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 455
Within this work, a decoupled modeling concept was introduced for the predic-
tion of local work roll wear during cold and hot rolling process. As described in
the concept and implementation section, three different wear models were imple-
mented to analyze the work roll wear effects on the rolling process. The models
were validated by the published data [13, 18]. In particular, Sikdar [18] published
hot rolling process information and the measured work roll wear for the finishing
stands. Within their work, the roll surface profile was measured before and after a
rolling schedule using a contact type profile gauge with a measuring accuracy of ±
1 μm. After rolling a selected number of strips (for the published case, 56 strips),
the worn work rolls were removed from the finishing mill and subjected to 2h of
water cooling followed by 3h of air cooling. This cooling schedule eliminated any
effects of thermal expansion on the work rolls. The other source used for the vali-
dation was published by Mohammed Tahir [13], where the wear of the work rolls
was evaluated for different roll materials for hot rolling finishing mills. According
to this input, various simulations were conducted to understand the work roll wear
using three different wear models and compared with the published data. The
work roll wear of the finishing stand (stand 4, top roll) of a tandem rolling mill
was further analyzed. The process conditions at this stand are provided in [18].
Three different strip widths (900, 700 and 500 mm) were simulated for the strip
length of 100 km each. The average simulated rolling force (for three strips) was
about 13.5 MN whereas the measured force ranges at about 12.5 MN [18].
For the analysis of wear using the empirical model (Eq. 6), the coefficients α, β
and γ were 1.0E-10, 0.188 and 0.11 respectively. To predict wear the normal pres-
sure is obtained from the process simulation and Equation 6 solved for each
surface element. As shown in Figure 20.17, for the work roll wear profile at the
centre, i. e. up to 500 mm, the wear is higher and uniform. The reason is that, for
all strip widths, the work roll is loaded at the centre. When using an empirical
wear model, the wear along the strip width is uniform; the reason for this could be
that the model considers only the normal contact pressure. Another cause may be
the inlet strip temperature, which is assumed to be uniform over the width, and
there is no considerable elastic deformation of the work rolls, which influence the
contact pressure profile. Figure 20.17 (right) shows the contact pressure distribu-
tion (for 700 mm strip width) before and after updating the worn roll profile with-
in the process simulation.
Since the stresses due to friction are relatively high during the rolling process
any prediction neglecting these may lead to undesired results [17]. For that, an
energy-based wear model was implemented, which considers the frictional dissi-
pation rate. For each element, the accumulated wear work is estimated. The
required shear stress and the normal stresses are extracted from the process simu-
lation. The relative velocity in rolling direction is the difference of the roll velocity
and the strip velocity within the roll gap. Using this information, wear Equation 10
is solved for each surface element. As shown in Figure 20.17, the wear predicted
by the energy model is relatively local and higher than empirically estimated at the
middle, where high shear stresses may occur. The order of magnitude from both
models is very close to the measured wear profile. Further analysis and the valida-
tion of these models have to be conducted with real experimental data.
A new FE modeling concept was developed for the analysis of interaction effects
during the flat rolling process. Within this concept, the tool simulation is separated
from the process simulation and, using an automatic coupling module, the com-
munication between these two models was realized. The advantages of this model-
ing concept includes a short computation time (up to 2 times faster than a single
FE model in the case of an elastic tool model) compared to the single FE model,
suitable for the analysis of flatness problems (modeling of 3D), easy to handle
thermo-mechanical effects during the modeling of a hot rolling process and the in-
fluence of tribological effects (particularly abrasive wear of the work rolls) on the
strip quality can be analyzed. The simulation results of the cold rolling process
(with the careful selection of frequency of activation of the tool computation)
show good agreement with the single FE model and also the experiments. At the
same time, the modeling of the hot rolling process was also validated with the help
of lab experiments. The rolling force was clearly underestimated using a constant
gap conductance model, even with a high friction coefficient. When using the
pressure-dependent model, however, the predicted rolling force is very close to the
measured value. At the same time, the exit strip temperature and strip exit thick-
ness also matches well with the measurements. For the analysis of roll thermal
events for different cooling strategies, extensive numerical analysis was conducted
using the rotating boundary model. It was shown that the model is suitable and
easy to handle for the prediction of steady state thermal gradients within the work
rolls.
At the end, wear models were validated using published data. Since the history
of the rolled strip width is not mentioned within the publication, further analysis
and validation of the simulated results are necessary. Future work will consider
obtaining precise wear data from the industry to validate the developed models.
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 457
References
[1] Kainz, A., Krimplelstätter, K., Zemen, K.: FE Simulation of thin strip and temper
rolling processes. In: ABAQUS Austria User Conference (2003)
[2] Kim, T.H., Lee, W.H.: An integrated FE process model for the prediction of strip pro-
file in flat rolling. ISIJ 43, 1947–1956 (2003)
[3] Ohe, K., Kajiura, S., Simada, S., Mizuta, A., Morimoto, Y., Fujino, T., Anraku, K.:
Development of shape control in plate rolling. In: METEC Conference proceedings,
vol. 2, pp. 78–85 (1994)
[4] Buessler, P., Montmitonnent, P.: A review on theoretical analysis of rolling in Eu-
rope. ISIJ 31, 525–538 (1991)
[5] Guo, R.-M.: Development, verification and application of an optimal crown shape
control model for rolling mills with multiple control devices. In: AISE Conference
(1995)
[6] Tseng, A.A.: A numerical heat transfer analysis of strip rolling. ASME 106, 512–517
(1984)
[7] Hsu, C.T., Evans, R.W.: Finite element analysis on the hot rolling of steel. Adv. Tech.
Plasticity 2, 587–593 (1990)
[8] Sun, C.G., Hwang, S.M.: Prediction of roll thermal profile in hot strip rolling by the
Finite element method. ISIJ 40, 794–801 (2000)
[9] Sun, C.G., Hwang, S.M., Ryoo, S.R., Kwak, W.J.: An integrated FE process model
for precision analysis of thermo-mechanical behaviour of the rolls and strip in hot
strip rolling. Comp. Methods. Appl. Mech. Engg. 191, 4015–4033 (2002)
[10] Lee, J.H., Hwang, S.M., Park, H.D.: FE-based on-line model for the prediction of roll
force and roll power in hot strip rolling. ISIJ 40, 1013–1018 (2000)
[11] Jiang, Z.Y., Tieu, A.K.: Contact mechanics and work roll wear in cold rolling of thin
strip. Wear Journal 263, 1447–1453 (2007)
[12] Magne, A., Gaspard, C., Gabriel, M.: Wear behaviour of steels for hot working roll-
ing-mill rolls. CRM 57, 25–39 (1980)
[13] Mohammed, T., Widell, B.: Roll wear evaluation of HSS, HiCr and IC work rolls in
hot strip mill. Steel Research 74, 624–630 (2003)
[14] Kivilcim, E.N., Nürnberg, G., Golle, M., Hoffmann, H.: Simulation of wear on sheet
metal forming tools—An energy approach. Journal of Wear, 357–363 (2008)
[15] Archard, J.F., Hirst, W.: Wear of metals under un-lubricated conditions. Proceedings
of the Royal Society of London 236, 397–410 (1956)
[16] Byon, S.M., Kim, S.I., Lee, Y.: A semi-analytical model for predicting the wear con-
tour in rod rolling process. Journal of Materials Processing Technology 191, 306–309
(2007)
[17] Bowden, F.P., Tabor, D.: Friction, lubrication and wear: a survey of work during the
last decade. British Journal of App. Physics 17, 1521–1544 (1966)
[18] John, S., Sikdar, S., Mukhopadhyay, A.: Roll wear prediction model for finishing
stands of hot strip mill. Iron and Steel Making 33, 169–175 (2006)
[19] Franzke, M., Puchhala, S., Dackweiler, H.: Modeling of interaction effects between
strip and roll during flat rolling process. In: NUMIFORM Conference Proceedings,
vol. 908, pp. 1489–1494 (2007)
[20] Franzke, M., Puchhala, S., Dackweiler, H.: Modeling of interaction effects between
process and machine during flat rolling process. In: PMI Conference Proceedings
(2008)
458 S. Puchhala, M. Franzke, and G. Hirt
[21] Tseng, A.A., Huang, C.H.: The estimation of surface thermal behavior of the working
roll in hot rolling process. Heat and Mass Transfer 18, 1019–1031 (1995)
[22] Tseng, A.: Thermal modeling of roll and strip interface in rolling process: part 1 – re-
view. Numerical Heat Transfer 35, 115–135 (1999)
[23] Tseng, A.: Thermal modeling of roll and strip interface in rolling process: part 2 – re-
view. Numerical Heat Transfer 35, 135–154 (1999)
[24] Mikic, B.B.: Thermal contact conductance: Theoretical considerations. International
Heat Transfer 17, 205–214 (1974)
[25] Yovanovich, M.M., De Vaal, J., Hegazy, A.: A statistical model to predict thermal
gap conductance between conforming rough surfaces. AIAA Paper No: 82-0888
(1998)
[26] Colas, R., Torres, M.: Modeling heat conduction through an oxide layer during hot
rolling of steel. ASME Manufacturing Science Engineering 68(2), 577–582 (1994)
[27] Browne, K.M., Dryden, J., Assefpour, M.: Modeling scaling and de-scaling in hot
strip mills. Recent Advances in Heat Transfer and Micro Structure Modeling for Met-
al Processing 67, 187–201 (1995)
[28] Ranta, H., Larkoila, J., Korhonen, A.S.: A study of scale effects during accelerated
cooling. In: Modeling of Metal Rolling Process Conference Proceedings, pp. 638–647
(1993)
[29] Tseng, A.A.: A numerical heat transfer analysis of strip rolling. ASME Journal of
Heat Transfer 106, 512–517 (1984)
[30] Sellars, C.M., McLaern, A.J.: Modeling distribution of microstructure during hot roll-
ing of stainless steel. Materials Science and Technology 8, 1090–1095 (1975)
[31] Militzer, M., Nakata, N.: Modeling of microstructure evolution during hot rolling of a
780 MPa high strength steel. ISIJ 45, 82–90 (2005)
[32] Tseng, A.A., Lin, F.H., Gunderia, A.S., Ni, D.S.: Roll cooling and its relationship to
roll life. Metallurgical Transactions 20A, 2305–2320 (1988)
[33] Sun, C.G., Hwang, S.M.: Prediction of roll thermal profile in hot strip rolling by the
FE Method. ISIJ 40, 794–801 (2000)
[34] Azene, Y.T., Roy, R., Farrugia, D., Onisa, C., Trumann, M.H.: Work roll cooling sys-
tem design optimization in presence of uncertainty. In: CIRP Design Conference Pro-
ceedings, pp. 57–65 (2009)
[35] Partender, E., Mitter, S.: Ein Modell zur Prognose des Arbeitswalzenverschleiß der
Breibandstraße in Linz. In: XXII Verformungskundlisches Kolloquium (2003)
[36] Simulia, http://www.simulia.com/products/abaqus_fea.html
[37] Lasso GmBh, http://www.lasso.de/index.php?id=28
[38] GOM, Optische 3D-Koordinatenmessmaschine
http://www.gom.com/de/messsysteme/
systemuebersicht/tritop.html
[39] Franzke, M.: Zielgrößenadaptierte Netzdiagnose und -generierung zur Anwendung
der Finite Elemente Methode in der Umformtechnik. Dissertation, RWTH Aachen
University (1999)
Chapter 21
Increase of the Dimensional Accuracy of Sheet
Metal Parts Utilizing a Model-Based Path
Planning for Robot-Based Incremental Forming
21.1 Introduction
In the field of sheet metal forming, the production of prototypes or parts in low
batch sizes is being realized by high-priced conventional forming processes. The
lack of flexibility as well as high investment costs clearly point out the limits of
conventional sheet metal forming for these parts. Recently, different kinematics-
based sheet metal forming processes have been developed. In most cases, these
methods only require simple dies or supports ([2], [3]).
In this regard, the Chair of Production Systems (LPS) at the Ruhr-University of
Bochum has developed a robot-based incremental sheet metal forming process
(Roboforming) [4]. The Roboforming principle is based on flexible shaping by
means of freely programmable path-synchronous movements of two industrial 6-
axis robots driving universal workpiece-independent forming tools. Figure 21.1a
shows the machine setup, which consists of two KUKA KR360 robots and a frame
with a blank holder. The final shape is produced by the incremental inward motion
of the forming tool in depth direction and its movement along the contour in later-
al direction by driving either on parallel layers or on a helical path. The supporting
tool, also being realized with a simple geometry, holds the sheet on the backside
by moving synchronously along the outer contour, constantly on the same level or
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 459–473.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
460 H. Meier et al.
directly opposed to the forming tool. Thereby, a pre-defined gap between the two
hemispherical tools is being generated (Figure 21.1b).
Due to the high flexibility regarding the reachable geometries due to the robots’
kinematics and the tools’ simplicity, Roboforming provides a rather interesting
method for industrial prototyping and low batch size production. Apart from free-
form surfaces (Figure 21.1c), it is possible to form undercuts in a multi-step strat-
egy in a single clamping with the two 6-axis kinematics (Figure 21.1d).
Fig. 21.1 a) Roboforming setup, b) basic forming strategies, c) free-form surface and d) cy-
linder with undercut (97° wall angle) [5]
serial kinematic posture and the elasticities of its six joints. Each joint elasticity is
composed of elasticities and clearances in the gears and bearings, which are not
detected by the rotary encoders of the motors and therefore not compensated by
the cascaded control structure. To handle these deviations an entire model-based
method is developed, where no additional sensors or other modifications to the ro-
bot hardware are necessary. The implementation and experimental validation of
this model-based compensation method are described in this chapter.
Fig. 21.2 Measured process forces in a) radial, b) tangential, c) infeed direction and d) re-
sulting absolute tool path deviation for a frustum geometry (Ø90x40mm x 60°) produced
by roboforming with an incremental infeed of 1 mm. (material: DC06, thickness: 0.8 mm)
system modeling the robot. Here, the actual tool path and the robot compliance is
obtained due to the previously computed forces. Once the actual path is known,
the path correction compensating the force-dependent deviations is executed and
the correction data-set is derived. It contains the path, which has to be driven to
reach the reference path. When compiled and transferred to the robot control unit,
the corrected tool path allows the production of a workpiece without robot-
influenced errors.
Fig. 21.3 Path generation concept including robot compliance compensation [5]
Stress tensors
∂ψ e −T ∂ψ kin −T
S = 2 Fp−1 Fp , X = 2 Fpi−1 Fpi , Y = C S − CP X , Ykin = C p X (21.1)
∂Ce ∂C pe
Evolution equations
Y DC p b 2
C p = 2 λ , C pi = 2 λ YkinD C pi , κ = λ (21.2)
Y ⋅ (Y )
D D T c 3
Yield function
Φ = Y D ⋅ (Y D )T −
2
3
(σ y − R ) , R = −Q (1 − e− βκ ) (21.3)
Kuhn-Tucker conditions
λ ≥ 0, Φ ≤ 0, λ Φ = 0 (21.4)
Recent research has been focusing on the large deformation version of a new
eight-node solid-shell finite element based on reduced integration with hourglass
stabilization [14]. The major problem of low-order finite-elements, used to simu-
late thin structures as the sheet metal, is locking, a non-physical stiffening effect.
Therefore, in our recent solid-shell formulation the enhanced assumed strain
(EAS) as well as the assumed natural strain (ANS) concepts are implemented to
circumvent locking. To cure transverse shear-locking, the corresponding trans-
verse shear terms evaluated at locking-free sampling points are interpolated within
464 H. Meier et al.
the element domain. The same procedure applied to the transverse normal strain
cures curvature thickness locking. For the EAS concept, the derivation is based on
the well-established two-field variational functional
g1 (u , Ee ) = S ( E ) : δ Ec dV + g ext = 0 (21.5)
B0
g 2 (u , Ee ) = S ( E ) : δ Ee dV = 0 (21.6)
B0
∑ ’ ’
Fig. 21.4 Simplified structure of the robot model including compliance compensation
21 Increase of the Dimensional Accuracy of Sheet Metal Parts 465
The calculation of the joint angles necessary to reach a specified end-effector posi-
tion requires an inverse kinematic solution. In the implemented model, a numeri-
cal solution based on the Denavit-Hartenberg convention [15] for the description
of the kinematical structure is used. It makes use of the pseudo-inverse of the ma-
nipulator Jakobian to iteratively calculate the six joint angles of the robot for a
given position and orientation of the TCP. [16]
three DOF. The detail in Figure 21.5 exemplarily shows the design of the steady
damping and the torsion angle-dependent stiffness function for one tilting direc-
tion. Element Joint_i_stiffness_R3 considers the required stiffness characteristic as
shown exemplarily in Figure 21.6. All 17 stiffness characteristics were identified
through appropriate experiments described in chapter 21.2.2.4.
The rigid forward model of the robot is the same as the compliant forward model
but without the compliant part of the joints and the required dummies.
In order to measure the stiffness characteristics of all six joints in the 3 rotational
DOF an optical 3-D motion and position measurement system was used. The Me-
tris K600 system measures a position and orientation of a marker, which is
attached to at least 3 infrared ray emitting LEDs by use of 3 CCD cameras at an
volumetric accuracy (+/-2σ) of 90 μm. The system also allows the measurement
of relative movements between two objects by preparing both objects with LEDs.
This feature was used to measure the torsion angle of each robot joint under a de-
fined load by measuring the relative movement between the previous and the fol-
lowing link. The load was applied by a pneumatic cylinder, which was mounted
on the effector of the measured robot. Supporting the pressurized cylinder by the
second robot, a defined static force with varying amplitudes was applied to the
measured robot. To apply high torques in a defined direction to each joint opti-
mized postures of the robot were used. This procedure has been significantly
21 Increase of the Dimensional Accuracy of Sheet Metal Parts 467
assuaged due to the use of the second robot as the support can easily be located in
the workspace. The cylinder force was measured by means of a force torque sen-
sor (accuracy: ~5 % of full-scale load) mounted between the cylinder and the ro-
bots’ effector. The relevant torque was computed considering the distance
between the cylinder and the analyzed joint.
b)
a)
Fig. 21.6 Identification of the stiffness characteristic of axis 1 in rotational direction: Expe-
rimental setup and resulting stiffness characteristic Torque vs. Torsion Angle
Figure 21.6 exemplarily shows the experimental setup to identify the stiffness
characteristic of the first axis in rotational direction. The robot was set to its initial
posture to maximize the arm lever and therefore the applied torque and to realize
an exclusive deflection in the DOF of interest. Three LEDs were attached to the
carrousel to measure its deflection against the ground. Therefore, there was no
need for additional LEDs. At steps of 100 N, the force was increased up to a max-
imum of 1,400 N, decreased to 0 N, furthermore decreased up to -1,600 N and
increased up to 0 N again. Special care was taken to ensure that the brakes of the
robot remain unclamped during the experiment. The simplified characteristic
shown in Figure 21.6 was derived considering the arm lever of about 3,300 mm.
one hand, this geometry can be simply described analytical and allows an easy in-
terpretation of the experimental results. On the other hand, it represents the typical
forming strategy of geometries with Roboforming, where the effector mostly
moves with ring-like paths. As shown in Figure 21.7b, in the experiment, the ef-
fector moves along a helical path from outside to inside and at the same time
gradually increases the processing depth. The tool velocity v is set to 25 mm/s.
During the forming process, the actual TCP path was measured by an optical mea-
suring device, which detects the position of LEDs applied to the robot’s effector
(volumetric accuracy (+/-2σ): 90 μm). The process forces were measured using a
force-torque sensor located between robot effector and tool (accuracy: ~5 % of
full-scale load) (Figure 21.7a). Furthermore, the interpolated reference path and
the motion of each robot joint, measured by the rotary encoders and therefore not
including the elasticities of the axes, were logged by the robots’ control unit.
The description of the tool path for the programming of the robot and for the
simulation model as well as the tool position measurement are based on the local
Cartesian coordinate system CS located in the centre of the geometry, as shown in
Figure 21.7b. For a better presentation of the results, the measured and simulated
paths were transformed to the basis ϕ, which describes the angle between each
point of the path and the starting point of the forming process. ϕ is set to 0° at the
starting point of the forming process and continuously increases along the helical
tool path. The total rotation of the path during the whole forming process was
about 19,000°, which equals to about 53 revolutions. Based on the independent
variable ϕ, the vectors z , x2 + y2 , and x 2 + y 2 + z 2 can be used to describe the
tool position.
the robots. The x and y-ax xes describe the plane of the sheet. The z-axis is used foor
the thickness direction.
The time-dependent paath is given by x(t), y(t) and z(t). These coordinates oveer
time are translated into a user sub-routine to specify prescribed boundary conddi-
tion (DISP). At this, the influence
i of acceleration and velocity are neglected. Thhe
simulation of a complicatted forming process like Roboforming requires a realisttic
description of the materiaal behavior. As presented above, a large deformation elaas-
to-plastic material model with combined non-linear kinematic and isotropic haar-
dening is used to handle the springback effect of the sheet. It should describe thhe
Bauschinger and ratchetin ng effect, which belongs to the most characteristic phee-
nomena of the hardening behavior of metals. The Bauschinger effect is defined bby
the fact that straining in one
o direction reduces the yield stress in the opposite ddi-
rection, while ratcheting denotes the progressive increase of the mean strain as a
result of unsymmetrical stress cycles. Due to the fact that the contact behaviour is
very complex the solution n time is dominated by the number of contact iterationns,
which is relatively high in
n this simulation. Therefore, the contact formulation is aan
important aspect. Here, thhe forming and supporting tools are modeled as rigid boo-
dies. The interaction betwween them and the deformable sheet metal is defined bby
a penalty surface-to-surfaace contact method with finite sliding. The friction coefffi-
cient is 0.05. The resultss shown in this chapter are computed using an 8-nodde
continuum-shell element with reduced integration (SC8R). The equivalent plasttic
strain of the deformed sheet metal for an infeed of 20.04 mm can be seen iin
Figures 21.8a and b.
The process forces fromm the simulation of the sheet metal forming applying thhe
measured tool path of thee experiment are compared to the process forces provideed
by the experiment. Due to o the lack of available experiments to fit the material pa-
rameters differences are to
o be expected. Moreover, the simulation of one compleete
forming process of up to 20 mm in the infeed direction is very time-consuming.
A more efficient way is required
r to apply the correct material parameters for thhe
simulation.
470 H. Meier et al.
Fig. 21.9 Comparison of the reference path and the unloaded measured path to evaluate the
robots’ path accuracy. Comparison of the simulation results and experimental results for the
loaded robot to evaluate the robot model accuracy.
21 Increase of the Dimensional Accuracy of Sheet Metal Parts 471
To validate the compliant robot model the experiment was repeated with the
same tool path but with a sheet metal placed in the clamping device. Based on the
measured process forces (Figure 21.2) during this experiment, a simulation was
executed to calculate the deviated path. Figure 21.9 shows the results of both the
loaded robot experiment (dotted blue) and the simulation of the loaded robot (dash
dotted red). Considering the result of the real robot, in the first half of the experi-
ment, the average deviation increases up to 1.1 mm and the oscillation amplitude
enlarges up to 1.4 mm for the loaded robot. During the second half of the experi-
ment, these deviations stay constant. These characteristics are caused by the
process forces in all directions, which increase up to the middle of the experiment
and the rotation of the forces in radial and tangential direction about Z at a fre-
quency of ϕ = 360°. Further investigations of the results (not shown in the
diagram) have shown that the portion of the deviation in radial and tangential
direction is noticeably higher than in infeed direction, which is caused by the
higher stiffness characteristics of the robot in infeed direction. In conclusion, the
comparison of the simulated path and the experimental path shows a very good
qualitative and quantitative agreement. The maximal difference is about 0.3 mm.
Hence, the compliant robot model shows a behavior close to reality.
For the validation of the compensation approach, the compensation path al-
ready calculated in the previous simulation was used in another forming experi-
ment. Figure 21.10 shows the calculated compensation path (dashed green) and
the resulting measured path of the experiment (dash dotted red). With reference to
Fig. 21.10 Comparison of the resulting tool path before and after compensation to evaluate
the performance of the compensation approach
472 H. Meier et al.
the reference path (solid black), the compensation path is nearly symmetrical to
the uncompensated path of the loaded robot (dotted blue). Considering the meas-
ured tool path of the final experiment it can be noticed that it matches the refer-
ence path very well. Not only the average deviations but also the oscillations have
been reduced in all directions. A decrease from 1.1 mm to 0.1 mm for the average
deviation and from 1.4 mm to 0.8 mm for the oscillation amplitude can be ob-
served. These remaining oscillations are mostly the results of the robot´s inaccura-
cy itself, as already shown in Figure 21.9.
21.5 Conclusion
The chapter shows that the presented model-based approach has a high potential to
significantly increase the geometrical accuracy in Roboforming. The demonstrated
experimental results document a high efficiency of the compensation method. In
order to complete the model-based approach further optimization and the integra-
tion of the FEM model are necessary. Due to the fact that the computation time is
still very high and path deviations still exist, different investigations still have to
be carried out. With the presented material model we can further investigate the
influence of the kinematic hardening on the compliance of the sheet metal and the
springback. This is expected to provide significant impact on the simulation re-
sults. Additionally, the contact algorithms have to be improved. Finally, the ap-
proach has to be tested with complex geometries.
References
[1] Vladimirov, I.N., Pietryga, M.P., Reese, S.: On the modelling of nonlinear-kinematic
hardening at finite strains with application to springback – comparison of time inte-
gration algorithms. International Journal for Numerical Methods in Engineering 75,
1–28 (2008)
[2] Douflou, J., Szekeres, A., Vanherck, P.: Force Measurements for Single Point Incre-
mental Forming: A Experimental Study. In: Proceedings of the 11th Int. Conference
on Sheet Metal, pp. 441–448 (2005)
[3] Jeswiet, J., Micari, F., Hirt, G., Bramley, A., Douflou, J., Allwood, J.: Asymmetric
Single Point Forming of Sheet Metal. In: Proceedings of the 55th CIRP General As-
sembly in Antalya, pp. 88–114 (2005) (in Antalya)
[4] Meier, H., Smukala, V., Dewald, O., Zhang, J.: Two Point Incremental Forming with
Two Moving Forming Tools. In: Proceedings of the 12th International Conference on
Sheet Metal, pp. 599–605 (2007)
[5] Meier, H., Laurischkat, R., Zhu, J.: A Model Based Approach to Increase the Part
Accuracy in Robot based Incremental Sheet Metal Forming. In: AIP Conference Pro-
ceedings, vol. 1315, pp. 1407–1412 (2010)
[6] Vladimirov, I.N.: Anisotropic material modelling with application to sheet metal
forming RWTH Aachen, Institut für Angewandte Mechanik, Technische Universität
Braunschweig, Dissertation (2009)
21 Increase of the Dimensional Accuracy of Sheet Metal Parts 473
[7] Choi, Y., Han, C.S., Lee, J.K., Wagoner, R.: Modeling multi-axial deformation of
planar anisotropic elasto-plastic materials. part I: Theory, International Journal of
Plasticity 22, 1745–1764 (2006)
[8] Dettmer, W., Reese, S.: On the theoretical and numerical modelling ofArmstrong-
Frederick kinematic hardening in the finite strain regime. Computer Methods in Ap-
plied Mechanics and Engineering 193, 87–116 (2004)
[9] Hakansson, P., Wallin, M., Ristinmaa, M.: Comparison of isotropic hardening and ki-
nematic hardening in thermoplasticity. International Journal of Plasticity 21, 1435–
1460 (2005)
[10] Menzel, A., Ekh, M., Runesson, K., Steinmann, P.: A framework for multiplicative
elastoplasticity with kinematic hardering coupled to anisotropic damage. International
Journal of Plasticity 21, 397–434 (2005)
[11] Svendsen, B., Levkovitch, V., Wang, J., Reusch, F., Reese, S.: Application of the
concept of evolving structure tensors to the modeling of initial and induced anisotro-
py at large deformation. Computers & Structures 84, 1077–1085 (2006)
[12] Wallin, M., Ristinmaa, M.: Deformation gradient based kinematic hardening model.
International Journal of Plasticity 21, 2025–2050 (2005)
[13] Armstrong, F., PJ, CO: A mathematical representation of the multiaxial Bauschinger
effect, C.E.G.B. Report RD/B/N731, Berkeley Nuclear Laboratories, Berkeley, U.K
(1966)
[14] Schwarze, M., Reese, S.: A reduced integration solid-shell finite element based on the
EAS and the ANS concept – geometrically linear problems. International Journal for
Numerical Methods in Engineering 80, 1322–1355 (2009)
[15] Hartenberg, R.S., Denavit, J.: A kinematic notation for lower pair mechanisms based
on matrices. Journal of Applied Mechanics 77, 215–221 (1955)
[16] Corke, P.I.: A Robotics Toolbox for MATLAB. IEEE Robotics and Automation
Magazine 3(1), 24–32 (1996)
[17] Gerstmann, U.: Robotergenauigkeit: Der Getriebeeinfluss auf die Arbeits- und Posi-
tionsgenauigkeit, Dissertation, Universität Hannover (1997)
[18] Pham, M.T., Gautier, M., Poignet, P.: Identification of Joint Stiffness with Bandpass
Filtering. In: Proceedings of the IEEE Int. Conf. on Robotics & Automation, pp.
2867–2872 (2001)
[19] Beyer, L.: Genauigkeitssteigerung von Industrierobotern, Dissertation, Universität der
Bundeswehr Hamburg (2004)
Chapter 22
Gear Rolling Process
22.1 Introduction
The gear rolling process with round tools is one example of cold bulk metal
forming technologies. Cold rolling of high gearings is a new technique with new
requirements on the rolling equipment and process. Initially, pitch-related ques-
tions are of primary importance. The profile depth and the pitch size of high gear-
ings are larger than those of stub tooth and normal gearings. This results in the
displacement extent of the rolling tool by about half of the initial pitch value and a
changing pitch value during the process (initial and calibration pitch). Due to the
occurring strain hardening during cold forming the material is characterized by an
enormous flow stress that generates high stress rates during the forming process.
Resulting effects are elastic deformations of the rolling tools and a significant dis-
placement of machine components. The process forces affect the structure of the
machine, cause deviations in the rolling kinematics and influence the forming
forces that need to be generated by the machine. This complex interaction thus re-
quires a combined consideration of the forming process and structure of the ma-
chine in terms of rolling process cause-and-effect relationships as well as reactions
of involved machine components. The forming result largely depends on these
factors (Figure 22.1).
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 475–490.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
476 R. Neugebauer et al.
Existing machine and process design procedures applied to high gearings insuf-
ficiently comply with the rolling standards DIN 3960, 3961 and 3962. The main
reason is the inadequate understanding of interactions and relations between ma-
chine structure / components and forming process [1]. The aim of the presented
investigations was to determine the influence of machine components and tool
displacements in the rolling process on the gear quality of rolled high gearings.
The results lead to a better understanding of the system machine-process-tool and
can be used to enhance the quality of rolled high gears by implanting them in the
tool, process and machine design operations.
The process of gear rolling with round tools is divided into three phases:
• Initial rolling phase: rolling of tool tip pitches into the pre-form-diameter of the
workpiece (WP)
• Penetration phase: penetration of the round tools into the workpiece until de-
manded parameters of gearing are reached (tip diameter and root diameter)
• Calibrating phase: calibration of the finished rolled gearing to achieve complete
and uniform tooth form filling and circularity (Fig. 22.3)
The gear qualities are measured according to DIN 3962. This standard defines a
range of quality measures based on dimensions and accuracies for each relevant
gear variable. The technological limits in gear forming have been considerably ex-
tended. Ranges of gear types, formerly impossible to form, have been produced by
the round rolling technique. State of the art qualities for the rolling of stub tooth
gearings are qualities 7 to 8. First experimental investigations resulted in rolled
high gearings for automotive components with tooth height coefficient y > 2.7
22 Gear Rolling Process 479
with quality 9 to 10. The tooth height coefficient is defined by the following rela-
tion, where da and df are the tip and root diameters, mn is the normal module and hz
is the tooth height:
da − d f hz
y= = (22.1)
2mn mn
In the past, the accuracy of the rolled gear wheels depended to a large extent on
the quality of the tools. Nowadays, conventional manufacturing processes can
produce gear qualities 5 or 6. Even quality 4 can be reached by profile grinding.
These qualities are sufficient for reaching gear qualities 7 or 8 in the rolling
process.
480 R. Neugebauer et al.
Uneven numbers of teeth on the tool and gearing cause an unsymmetrical dis-
tribution of force impact points on the left and right side of the workpiece. In con-
sequence, left and right tool are horizontally displaced by about half pitch at the
tool tip diameter. The parameters of the gear used in the experiments are listed in
Table 22.1.
Based on the value of the pitch at the tool tip diameter ptool,A, the initial pitch pa
is calculated from the billet pre-form diameter dV and tooth number zpart according
to the equation (22.2). The pre-form diameter results from the cross-section area
of the gearing with the parameters given in Table 22.1.
dV ⋅ π 73.2 mm ⋅ π
pa = ptool,A = = = 11.5 mm (22.2)
z part 20
Subsequently, the initial tool displacement can be calculated:
pa 11.5 mm
= = 5.75 mm (22.3)
2 2
In addition to the problem of tool displacement the problem of an inconstant pitch
is a significant factor. The pitch value changes during the rolling process in a non-
linear progression (Fig. 22.8).
The initial pitch pa is calculated by using the pre-form diameter dv, while the
calibration pitch pk is calculated by using the target root diameter df:
d f ⋅π 69 mm ⋅ π
pk = = = 10.84 mm (22.4)
z part 20
Due to the pitch difference between initial pitch (pa = 11.5 mm) and calibration
pitch (pk = 10.84 mm) the tool teeth are subjected to bending stress. This stress
furthermore affects machine components and the rolling kinematic. The aim of
the investigations was the identification of the character and magnitude of the
described problems. The displacements and elastic deformations of machine com-
ponents were measured. The ‘Faro Laser Tracker’ was used to detect and record
displacement and deformation. This 3D portable laser measuring system is used
22 Gear Rolling Process 483
for the measurement of devices and machinery. Hereby, a laser beam from the
tracker is focused onto a mirror reflector and projected back by this mirror. The
reflected light waves allow the determination of distances as well as horizontal
and vertical angles. Using the obtained data the exact 3D positions can be calcu-
lated. Another advantage of the laser tracker system is the ability to measure mo-
tion. The mirrors rotate and the tracker, driven by a highly sensitive servomotor,
follows the mirror movement [5]. Seven rotating mirrors were mounted on specif-
ic points of the machine for the tests (Fig. 22.9).
Measuring points 1 and 2 were installed on the bearing blocks in front and be-
hind the tool to record the movement of the left tool axis under load. Furthermore,
the interaction effect with the left carriage could be analyzed by the rotating mir-
rors 6 and 7. To determine the tool displacement during the rolling process mirrors
were mounted in front of and behind the round rolling tools (1, 2 and 4). The inte-
raction between tool holder and rolling carriage could be measured by the rotating
mirrors installed parallel to the tool measuring points (5, 6 and 7). At the tips of
each clamping device a measuring point (3) was fixed in axial workpiece position
484 R. Neugebauer et al.
to determine the interaction between tools, workpiece, and workpiece holder. The
measuring points on the tool axes (measuring points 1 and 4) were used to detect
the relative positions between tools and workpiece during the rolling process. The
experiments were carried out with the tool rotation speed set of 12 revolutions per
minute and 4 reversion points. The graph in Figure 22.10 displays the progress of
the rolling force and tool path during the rolling process. The maximum recorded
force was 99.2 kN, which corresponds to a machine load of about 25 % (maxi-
mum load capacity 400 kN).
22.2.4 Results
The following diagrams show the displacement of the measuring points (MP) dur-
ing the rolling process recorded by the laser tracker system. The beginning of the
process represents the zero-situation, i. e. the unloaded initial position of the ma-
chine system. During the rolling process the direction of the tool revolution re-
versed four times. At the points of reversion during the penetration phase the
entire machine system is briefly unloaded. Such reversion points correspond to the
points of sharp amplitude changes of the measurement curves.
Fig. 22.11 Displacement in the x-y plane (left: tool axle; right: rolling slide)
Fig. 22.12 Displacement in the x-y plane (left: initiation; right: displacement)
Fig. 22.13 Displacement in the x-z plane (left: tools; right: rolling slides)
486 R. Neugebauer et al.
Fig. 22.14 Displacement in the x-z plane (left: initiation; right: displacement)
Fig. 22.15 Displacement in the y-z plane (both tools und clamping tips)
Fig. 22.16 Displacement in the y-z plane (left: initiation; right: displacement)
22 Gear Rolling Process 487
Because of the increase of available computing capacities and the latest findings in
the FEM simulation of incremental forming technologies an illustration of the de-
manding process kinematics became possible. For the selected reference gearing
of module mn = 3.45 mm and zpart = 20 a regular process containment with tools as
deformable geometries, the workpiece, the rolling kinematic and the boundary pa-
rameters was defined. The aim of simulating the rolling process was to calculate
the three-dimensional component material flow, the maximum achievable defor-
mation, the extent of workpiece strain hardening, the resulting orientation of ma-
terial fiber and the occurring tool and workpiece loads. The simulation was carried
out using the Transvalor Software Forge. The rolling machine itself was defined
as non-deformable. The measured displacements of tools, workpiece and machine
components shall be included into the simulation model in further projects. The
process was computed in an acceptable calculating time and could be verified by
practical investigations. The following tables provide an overview of setting pa-
rameters of the simulation.
discretization
discretization
Figure 22.17 shows the stress distribution on the tool and deformed workpiece
in the normal section at the end of the penetration phase (left figure) and a picture
of the rolled component (right figure). A detailed validation and evaluation of the
quantitative simulation resulting in verifiability is intended to provide a compari-
son with the practice tests or with the experimental-numerical method for the de-
termination of the tool loads. The equivalent stress by v. Mises was used to
represent the three-dimensional stress state, showing high values of equivalent
stress. The results of the simulation lead to the selection of high performance
powder metallurgical steel with high strength and toughness as rolling tool materi-
al. Compared to the real material behavior the simulation shows a rotation kine-
matics-related volume increase in the forming simulation process in spite of
Runge-Kutta interpolation (RK2), which results in a faster "form filling" of the
tool gap with workpiece material in the simulation than in the real process.
22 Gear Rolling Process 489
22.5 Conclusion
The demand for manufacturing methods, which ensure a high utilization of ma-
terial with a high quality of the final products and short production times, conti-
nuously pushes the improvement of existing and the development of new process
technologies. The manufacturing of transmission-typical gears by cold forming is
one new approach. Current industrial applications of gear rolling are in the range
of stub tooth gearing, tappet gearings and pinions in defined limits to maximum
module 1.6. First investigations demonstrated that a forming capacity of high
gearings can be obtained by adapting and optimizing the rolling process to the
specific gear parameters. The results of the trial series proved that tooth heights
can be manufactured, which so far have not been producible by cold cross rolling
490 R. Neugebauer et al.
processes. The main tasks of future research are to improve the tool and process
design regarding the accuracy of the rolling kinematic and the precision of the
pitch. The machine concepts of future rolling machines have to be adapted to and
optimized for high gear rolling as currently available machines are derived from
guidelines for the rolling of stub tooth gearings. Especially the stiffness and/or de-
sign of the workpiece holder and rolling tool axles have to be enhanced, regarding
the high loads in the process of high gear rolling. The first steps into the field of
cold forming of high gears have been taken but major research work on the analy-
sis of the interaction between process, machine and part is needed in order to de-
velop the technology to a reliable level for industrial application. From a scientific
point of view the detection of tool stresses within the process boundary conditions
provided the basis for further steps in a technology research and process optimiza-
tion. The knowledge of load parameters as well as their local and temporal occur-
rence allows the performance of direct discharges for parameter-adjusted tool and
process dimensioning. Adequate tool materials and heat treatments become select-
able. Thus, the FEM simulation is an implementation for shortening development
times and product innovations while maintaining the product quality requirements.
References
[1] Neugebauer, R., Hellfritzsch, U., Lahl, M.: Advanced process limits by rolling of heli-
cal gears. In: 11th ESAFORM Conference on Material Forming, Lyon (2008)
[2] DIN 3960, Definitions, parameters and equations for involute cylindrical gears and
gear pairs. Beuth Verlag GmbH, Berlin (1987)
[3] DIN 3961, Tolerance for cylindrical gear teeth, bases. Beuth Verlag GmbH, Berlin
(1978)
[4] DIN 3962, Tolerance for cylindrical gear teeth. Beuth Verlag GmbH, Berlin (1978)
[5] Neugebauer, R., Klug, D., Hellfritzsch, U.: Description of the interactions during gear
rolling as a basis for a method for the prognosis of the attainable quality parameters.
Prod. Eng. 1(3), 253–257 (2007), doi:10.1007/s11740-007-0041-9
[6] Neugebauer, R., Hellfritzsch, U., Lahl, M.: Interaction of Machine, Tool and Process
by Rolling of High Gears. In: Proceedings of the 1st International Conference on
Process Machine Interactions, Hannover (2008)
Chapter 23
Investigation of the Complex Interactions
during Impulse Forming of Tubular Parts
23.1 Introduction
Implementing lightweight construction concepts has been becoming more and
more important over the last years. Thereby, material as well as design aspects
have to be considered. The application of space frame structures from hollow pro-
files made of the typical lightweight metals as aluminum and magnesium alloys is
a common approach. However, to gain maximum weight reduction, it is necessary
to apply the most suitable material for each separate component [1]. Consequent-
ly, a material mix including high strength steels, (fiber-reinforced) polymers and
the already mentioned lightweight alloys results. Further weight reduction can be
achieved by combining the functions of different parts and thus eliminating sever-
al components completely. This constructive approach usually leads to a higher
complexity of the remaining components, which have to fulfill different and some-
times contradictory functions [2]. This concerns the geometry and material as-
pects. It could be necessary to combine the corrosion resistance and strength of
aluminum, steel or titanium with the high electrical and thermal conductivity of
aluminum or copper, for example, guaranteeing a thin-walled but stiff geometry at
the same time. Consequently, highly complex forming and joining tasks have to be
performed in modern production processes so that conventional technologies fre-
quently reach their limits. Considering conventional forming operations, such as
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 491–511.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
492 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
quasi-static hydroforming alone, the formability might be too low to realize the
complex part geometry without wrinkling, buckling and/or cracking. When think-
ing of thermal joining technologies, problems might arise due to the high affinity
of aluminium, steel and titanium to the atmospheric gases oxygen, nitrogen
and hydrogen. Especially at elevated temperatures in the range of 300°C and
more, brittle intermetallic phases can be formed, which decrease the joint quality
significantly.
Here, high speed or impulse-forming technologies, such as explosive forming
by a gas detonation and electromagnetic forming (EMF), offer a high potential.
They can be applied for forming with and without a form defining tool as well as
for joining by forming, where the connections can be based on force-fit, form-fit
and impact-welded material joints. The processes use the energy density of a de-
tonating gas (forming by gas detonation) or pulsed magnetic fields (EMF) to acce-
lerate and form tubular workpieces without any mechanical contact to the tools
and machines, for example. Although the principle and the advantages of both
technologies have already been known for several decades no wide implementa-
tion into industrial series production has been realizable up to now. One important
reason for this deficit is the lack of a deep understanding of the process and, based
upon this, the lack of general and easily applicable design strategies for the
process dimensioning [3]. To fill these gap fundamental investigations considering
the complex interdependencies occurring during high-speed forming and the join-
ing of tubes by gas detonation and EMF, respectively, were carried out bearing in
mind aspects of forming technology, material science and process technology. The
results of this work are presented in this article, structured according to two differ-
ent sub-systems:
• Sub-system 1 deals with the acceleration and the free forming of a workpiece.
Here, the investigations focus on the interdependencies between the pressure
pulse and the workpiece deformation, which, in turn, retroact on the temporary
and local development of the pressure. These investigations were carried out
disregarding the special application (forming or joining) but considering the
two different acceleration technologies separately. In the case of forming by
gas detonation, the interdependencies between the reactive gas, the energy
source, and the working medium at the same time and the workpiece is rele-
vant, while in the case of EMF, the interdependencies between the tool coil, the
magnetic field and pressure, and the workpiece are considered.
• Sub-system 2 deals with interdependencies during the impact of the workpiece
and an according obstacle, which can be a form-defining tool or a joining part-
ner. Here, the impacting parameters (e. g. impacting velocity and angle), the
properties of the two components (workpiece and obstacle) and the formation
of the contact are analyzed. The evaluation of the result is carried out consider-
ing the special application. While in the case of forming against a form defining
tool, the form filling and the accuracy of the deformed part are decisive,
the formation of the joint and the transferable load are relevant in the case of
joining.
23 Investigation of the Complex Interactions during Impulse Forming 493
In both variants, the setup can be represented by an oscillating circuit, where the
forming machine serves as the energy supply while the tool coil and the workpiece
represent the consumer load.
Fig. 23.1 Principle sketch of the electro-magnetic tube compression and expansion
In Figure 23.1, the forming machine is represented by its equivalent circuit dia-
gram consisting of the capacitor C, the inner resistance Ri and the inner inductance
Li. Closing the high current switch causes a sudden discharge of the capacitor and
thus a damped sinusoidal current I(t) flows through the tool coil so that a corres-
ponding magnetic field H(r,z,t) results. The time-dependent field induces a current
in the workpiece, which flows opposed to the coil current and shields the magnetic
field. Consequently, radial Lorentz’s forces act on the workpiece and deform it in
the direction away from the tool coil. These volume forces can be mathematically
transferred to virtual surface forces, the so-called magnetic pressure p(r,z,t).
Due to the short process times in the range of 100 µs or less in impulse forming
and joining, high demands have to be made on the measurement technique. During
EMF, the coil current is measured via a Rogowski probe; on this basis, the mag-
netic pressure can be calculated analytically or via a coupled structural mechanic
494 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
In EMF, significant interdependencies between the acting loads and the workpiece
deformation were identified. These become evident when comparing current
courses measured during a forming process and during experiments with the same
process parameters but with prevented workpiece deformation. Figure 23.2
presents such curves together with the pressure curves and the radial displacement
of the smallest cross-section measured during a tube compression.
The diagrams clearly indicate that, at the beginning of the process, the coil
current and the magnetic pressure increase accordingly while the workpiece re-
mains undeformed. Plastic deformation occurs as soon as the inertia is overcome
and the stresses in the workpiece reach the flowstress of the material. It is corre-
lated to an increasing gap width between tool coil and workpiece; consequently,
the inductivity of the consumer load (tool coil and workpiece) steadily increases.
Therefore, the amplitude and the frequency of the coil current are lower compared
to the reference experiment with prevented workpiece deformation. Thus, consi-
dering the direct comparison of the current measurements the beginning of the
workpiece deformation is indicated by the point of time, where the curves are no
longer collinear. After the saddle point in the current curve measured during the
forming experiment, the inductivity of the oscillating current remains constant.
This can indicate that the forming is completed. In the displacement curve, an
elastic radial oscillation of the tube can be recognized because the high kinetic
energy resulting from the high forming velocity cannot be dissipated spontaneous-
ly. [4] Under certain circumstances, the increasing gap volume between coil and
tube can cause the magnetic pressure to decrease faster than the current. Moreo-
ver, with time, the magnetic field penetrates the workpiece causing a further pres-
sure diminution. This penetration is faster, the lower the discharging frequency,
the thinner the workpiece wall and the lower the electrical conductivity of the
workpiece [5].
23 Investigation of the Complex Interactions during Impulse Forming 495
Fig. 23. 2 Interdependency of current course, magnetic pressure and radial displacement in
electromagnetic tube compression
The experimental setup used for investigation forming by gas detonation is illu-
strated in Figure 23.3 and explained in the following: A tube, which is fixed in the
testing apparatus designed by the SWL1 and located in the explosion-protected
area of the SWL, is filled with a stoichiometric mixture of hydrogen and oxygen
using a filling connector via mass flow controllers. The gas is then detonated ex-
ternally using a wire, which is ignited by a high voltage discharge. The detonation
wave, which is immediately formed, travels through the pipework, which bifur-
cates and then enters the detonation tube from two sides. The gas mixture in the
apparatus is now ignited, which a homogeneous conflagration produces over the
entire cross-section. A tubular-shaped probe, located in the middle of the appara-
tus, is installed co-axially to the workpiece and fixed at both ends of the testing
1
Stoßwellenlabor RWTH Aachen.
496 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
apparatus. Two pressure sensors, located at different positions in this probe, are
connected to the gas-filled space by radial holes and shielded from direct contact
with both the active components as well as the combustion gases by high tempera-
ture grease.
To keep the apparatus' design simple, all the tested workpieces have an initial
diameter of 40 mm. In the course of the test setup, the tubes are slid to the right
end onto a sealing ring and fixed by means of a split sleeve. The tube's axial
movement is prevented by an O-ring seal at the left end. Four planar tool compo-
nents, which constrain the forming but still permit the formed region to be directly
observed, can be arranged around the workpiece. Moreover, the workpiece can be
replaced by a thick-walled tube and thereby enable pressure profiles to be meas-
ured without influencing the volume changes due to the forming process [6].
The optically accessible forming section can be photographed during the
process by a high-speed (HS) camera. The camera enables the workpiece's contour
to be investigated at different times during the forming. The DRELLO type HS
camera has a framing frequency of 10 MHz for 24 images. It is triggered by the
first pressure sensor due to the travelling pressure wave [6].
The two typical, measured pressure profiles depicted in Figure 23.4 were recorded
during free and constrained expansion tests using Al 99.5 and thick-walled tubes
respectively.
23 Investigation of the Complex Interactions during Impulse Forming 497
Fig. 23.4 Pressure courses during free tube expansion and reference experiments with pre-
vented workpiece deformation [6]
An initial pressure p0 of 8.5 MPa was used for the tests. Owing to the strong
oscillations the measured pressure signals required smoothing. The zero point on
the time axis is defined by the initial signal of the first pressure sensor. The initial-
ly measured signal deflection of the pressure curve cannot be resolved in detail
due to the smoothing or the sensor signal's strong vibrational behavior. The theo-
retical maximum value for this initial pressure is 17 MPa. In the diagrams of the
first and second pressure sensors, the second, smaller rise in pressure is explained
by the partial reflection of the detonation wave at the sealing ring depicted in Fig-
ure 23.3. The third noticeable pressure rise after approx. 280 µs and 180 µs, re-
spectively, is generated by the reflection of the detonation wave at the end wall.
Consequently, the expansion of the tube leads to a faster and earlier drop in the
pressure profile based on the position of the detonation wave [6].
To characterize the forming sequence high-speed photographs were taken dur-
ing selected free expansion tests. The initial gas mixture's pressure, which deter-
mines the maximum pressure operating in the process, is limited to p0 ≈ 0.9 MPa
for the 40 x 2 mm Al 99.5 tubes. Exceeding this value leads to the specimen's fail-
ure by crack formation during the expansion process. The forming sequence is de-
picted in Figure 23.5 with the simultaneous reference to the pressure profiles. In
Figure 23.5a), an original HS-camera image showing the process status 121 µs af-
ter triggering by means of the first pressure sensor using the arrival of the detona-
tion wave is depicted. This geometry can be seen as the form's final contour. The
diagram in Figure 23.5b) shows various stages of the tube's expansion. Additional-
ly, the position of the wave front at 21 µs and 41 µs is plotted. The tube wall's
highest radial speed of 45 m/s is attained at the right hand end of the free-region
shortly before the second pressure sensor indicates the second large rise in pres-
sure (compare Fig. 23.4), which is explained by the waves’ reflections [6].
Further elevating the initial pressure p0 produces a still larger expansion, which
continues beyond the observation time (Fig. 23.5, experiment no. 33)). The final
diameter obtained is plotted on the graph's right ordinate.
498 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
Fig. 23.5 Results of online measurements of the expansion during forming by gas detonation
[6]
To be able to also track the effect of the higher pressures, the four planar tool
components are employed. In Figure 23.5c, the commencing phase of the expan-
sion is shown for the initial pressure of p0 = 1.13 MPa. The maximum radial speed
amounted to 65 m/s at this pressure. This value approximately corresponds to a
strain rate of 3,000 s-1[6].
By supplementing the described experimental investigations a numerical analy-
sis of the forming by gas detonation was realized. As a first step, an uncoupled
simulation strategy was applied in order to avoid problems due to changing boun-
dary conditions resulting from the displacement of the tube wall. Such problems
typically occur in fluid mechanical calculations, which are usually based on Euler
formulations. The shock tube simulation program Kasimir, developed by SWL,
was employed to determine the pressure-time profile [7]. With the aid of a Rie-
mann solver, this software solves the thermo-dynamic equations for shock waves
taking into account the chemical reaction. The results obtained using this program
were employed as time-dependent input parameters for the structural mechanical
simulation in Msc.Marc in combination with various user sub-routines developed
at the IUL2. In order to take into account the influence of the inertial masses dur-
ing the forming, an implicit Newmark time-integration scheme was employed to
solve the equations of motion, ignoring the damping effects [8]. An ALE strategy
using a second FE mesh was developed and implemented for interpolating the im-
ported data. In this way, the forces' time and spatially-dependent boundary condi-
tions were introduced into the structurally dynamic model. This mesh, which is
only necessary for the time and spatial interpolations in the first step of the
modeling, is simultaneously distorted by the tube's forming. For this purpose, an
2
Institut für Umformtechnik und Leichtbau, Technische Universität Dortmund.
23 Investigation of the Complex Interactions during Impulse Forming 499
Figure 23.6c shows the results achieved by applying this simulation strategy to
a free tube expansion. After initiating the detonation wave, the tube's wall remains
in its initial position for about 20 µs until the forming begins. The reason for this
is the effect of the inertial masses and their associated forces, which counteract the
acceleration. The elastic-plastic deformation of the initially accelerated tube re-
gion ceases when the detonation wave reaches the end of the tube. It was also
possible to experimentally verify this time delay between the detonation wave and
the tube wall's movement.
The reflected wave re-accelerates the tube's wall, which is already moving. Fol-
lowing the wave's second reflection, the deformation process has terminated and
the final form has been obtained. The entire process lasts for approx. 200 µs, whe-
reby the quantitative data for the process sequence depends to a large extent on the
starting parameters. The numerically determined contour fits the measurement
well qualitatively (compare Figure 23.6 and Figure 23.8). The suggested modeling
including the simplification employed here, such as the decoupling strategy, can
therefore be considered as permissible.
A completely coupled simulation, possessing simultaneous and step-wise solu-
tions of both field problems, is necessary for a mathematically correct model of
the process and, in particular, of the interactions. However, the required effort for
500 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
this is disproportionate to the gain in knowledge with respect to the state of the
current investigations. For this reason, the fall in gas pressure due to the increase
in volume was also simply modeled by means of an exponential pressure drop.
However, the results obtained using this strategy showed no substantial influence
of the deformation process.
Fig. 23.7 Roundness and wall thickness profiles of free expansion tests a) gas detonation, b)
quasi-static water pressure (according to 6)
Figure 23.8 shows the contour and the according distribution of the hardness
measured for a specimen expanded by gas detonation.
It can be clearly recognized that the hardness increases with increasing defor-
mation. This effect corresponds to the strain hardening known from conventional
forming processes. Comparing the hardness values to according measurements for
a quasi-statically expanded tube shows good qualitative and quantitative agree-
ment. This indicates that the high strain rates occurring during forming by gas
detonation do not significantly influence the strain hardening behavior of this ma-
terial. This tendency again corresponds to similar experiments comparing quasi-
statically formed and electromagnetically compressed tubes [10].
However, the temperature gradient, the impulsive forces and the accordingly
high forming velocities lead to a specific deformation of the workpiece during
forming by gas detonation; consequently, the corresponding forming mechanisms
are expected to differ from conventional ones. Therefore, more fundamental ma-
terial investigations were carried out to provide more comprehensive knowledge
about the process specific influences on the material behavior.
502 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
High-speed tensile tests were carried out in a drop tower testing-rig at the IW3
to determine the influence of the strain rate on the material properties. By using
this drop tower it was possible to investigate high strain rates in round-bar speci-
mens. In this test, the specimen is accelerated using various falling test weights in
a 4 m high drop tower testing-rig with a maximum falling speed of 6.3 m/s. A test
weight impacts the anvil; and the resulting impulse is transmitted to the lower spe-
cimen's grips via four rods. In this way, the specimen is subjected to straining at a
high speed. The installed load cell (Hottinger 50kN, type U3) detects the resulting
load. In addition to this, specimens were successfully prepared in an appropriately
equipped metallurgical laboratory for subsequent investigations in scanning elec-
tron and light microscopes (see below).
Moreover, in collaboration with the IUL and SWL, it was possible to success-
fully carry out thermographic investigations of both the gas denotation procedure
as well as the comparative static procedures. This enabled the investigation of the
effects of the heat, generated during the method and forming. The thermal camera
employed for this purpose, ThermaCAM SC 3000 made by the company FLIR,
incorporated a high-speed facility.
To investigate these influences metallographic investigations were carried out
by the IW. The micro-structural changes as a result of deformation can be de-
scribed by a multiplicity of structure-relevant parameters. Dislocations and grain
boundaries have a substantial influence on the dynamics of the forming process
[11, 12]. For example, dislocations within a grain show a different behavior than
at or near grain boundaries. These differences are caused by the structure and type
of the dislocations [13]. Previous analyses of the dislocation movement and the
dislocation density changes as well as the analyses of grain boundary migration
took place under thermal activation or at low strain and strain rates. The present
results of the conducted analyses can serve as the basis for the description of the
plastic deformation mechanisms at high strain rates in the finite element simula-
tion. The structural changes were examined using aluminum samples. The
processes during plastic deformation are related to different dislocation-drag me-
chanisms [14]. These are influenced by the following factors: - thermally activated
dislocation movement energy of dislocation drag - impurity viscosity - phonon
viscosity - electron viscosity. The first three mechanisms are particularly relevant
in static tensile tests with the conventional universal testing machines (tension
speed of 2 µm/s to 1,000 mm/min). It is well-known from literature that the
mechanism of phonon viscosity is very important at tension speeds of 1 m/s to
1,500 m/s [15].
The processes at the grain boundaries are extraordinarily important as well as
the dislocation drag in grains for the measured characteristics of the process of gas
detonation. The mobility of grain boundary and its driving strength are the funda-
mental parameters of the grain boundary slip [16]. The origin of this driving
strength has different causes. For most cases, the strength results from the differ-
ences in the free energy of the adjacent grains. This difference can be caused by
the different surface energies, permeabilities and disorientations of the adjacent
grains or by different dislocation densities on both sides of the grain boundary
3
Institut für Werkstoffkunde, Leibniz Universität Hannover.
23 Investigation of the Complex Interactions during Impulse Forming 503
[17]. A grain boundary slip sets the deformation energy stored in the dislocations
free; and the energy of the complete system is minimized. One of the most impor-
tant procedures for the analysis of the dislocation and the grain boundary structure
is the light-microscope etching pit method. Numerous methods have already been
suggested for the determination of the defect structure in aluminum. At the IW, a
new etching agent and a special procedure were developed on the basis of an etch-
ing agent, recommended by Barret and Levenson [18]. When using this procedure
the dissolution speed does not depend on the crystallographic orientation so that
the etching speed is evenly distributed on specific crystal faces. This procedure
can be used for the determination of changes in the defect structure after the de-
formation of poly-crystalline aluminum. A further application is to simultaneously
corrode grain boundaries, inner-grain dislocation pits and grain boundary disloca-
tion pits, using the described etching agent in combination with a certain parame-
ter set for electrolytic polishing. The reproducibility of the metallographically
determined etching pit density can be determined by comparison with trans-
electron-microscopy (TEM) measurements of the dislocation density. The sample
material was poly-crystalline aluminum (99.5 %), which was tested in the form of
rolled strips of 1 to 4 mm thickness and 70 mm width and of cast slabs of 7 mm
thickness and 70 mm width. The samples for the tension tests were prepared from
strips and slabs, homogenized for 72 hours at a temperature of 580°C and cooled
down along with the furnace.
After the tension tests with different test speeds, the samples were examined by
light-microscopy parallel to the tension direction (longitudinal axis). The samples
for the etching pit method were electro-polished in electrolytes (20 % perchloric
acid, 10 % butylglycol and 70 % ethyl alcohol) by using a voltage of 25 V and a
flow rate of 5 for a duration of 10 s. The etching agent for this corrosion con-
tained: 10 % HF, 30 % H20, 45 % HCI and 15 % HNO3.
Fig. 23.9 Light optical and SEM investigations of the dislocation structure. The specimens
are pulled with different velocities
504 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
In Figure 23.9, examples of the etched micro-structure are shown. Different test
speeds were used for that. The etching pit arrangement corresponds to a disloca-
tion structure after deformation whereas the original grain boundaries are marked
by pit-free seams. The light microscopic and the REM investigations of the spe-
cimens of the tension tests show that two processes influence the microstructure of
the material. On the one hand, the dynamics of the grain boundaries cause an in-
crease of the grains and a frequency movement of the grain boundaries. On the
other hand, dislocation structures were formed, which, on their part, are able to
cause the development of a new structure. Both processes influence each other. A
selection of the results of the investigations according to this effect is shown in
Figure 23.9. Increasing pulling velocities cause a higher density of dislocations at
the grain boundaries and a growing deformation of the grains. Smaller pulling ve-
locities (2 mm/s) cause so-called dislocation fringes at the grain boundaries. An
accumulation of dislocations can be detected near this phenomenon. This causes a
significantly higher density of dislocations in this area than in the middle of the
grains. A further increase of the pulling velocity causes higher lengths at first and
then continuous devolutions of the walls of dislocations. The density of the dislo-
cations in those walls increases in relation to the degree of forming.
The joining quality was evaluated using metallographic methods. The investi-
gations have shown that for velocities up to 25 m/s no metallic bonding resulted
whereas higher impact velocities led to local impact welding. Figure 23.11 shows
the contact area of specimens, joint with an impact velocity of approximately
30 m/s, approximately 100 m/s and approximately 130 m/s in representative mi-
crographs. It can be observed that increasing the impact velocity results in larger
welded areas. Statistic interpretations of the micro-structural data suggest a linear
correlation between this ratio of welded area and contact surface and the impact
velocity in the regarded span of values. [19]
Fig. 23.11 Influence of different impact velocities on the contact area of joined specimens
[19]
506 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
However, also the intactness of the microstructure in the contact area has to be
considered when evaluating the welding quality. As shown in Figure 23.14, micro-
fractures occur in the aluminum parallel to the contact surface, indicating a deteri-
oration of the joining quality, if a critical value of the impact velocity is exceeded.
Here, this is shown for an exemplary impacting velocity of 130 m/s. In conclusion,
it can be said that an optimum value for the impact velocity exists. It has to be
high enough to maximize the ratio of the welded area and the contact surface but
low enough to avoid micro-fractures. For the regarded material combination and
setup, this optimum lies between 100 m/s and 130 m/s. [19]
To identify the resulting phases in the joining area, polished and etched speci-
mens were analyzed by an energy dispersive x-ray (EDX) and micro-analysis; and
the element distribution in the welding area was quantified. It was shown in [19]
that in-between the titanium and the aluminum base material an extremely thin
(thickness < 1 µm) titanium-rich transition area is located, however, according to
[20], intersection layers of such small thickness can be regarded as uncritical.
welding quality is located adjacent to this area. For the charging energy of 1,500 J,
this zone continues up to the axial end of the joining area. Contrarily, in the case
of the specimen welded with a charging energy of 1,000 J, another region charac-
terized by a strongly inhomogeneous layer between the aluminum and the titanium
was found. This area extends to the end of the compression zone and, again, indi-
cates an inadequate joining quality.
Fig. 23.13 Distribution of impact velocity and angle along the workpiece axis [19]
where the high welding quality can be observed, favorable combinations of the
impacting velocity and the impacting angle were reached. This approach suggests
that not only high workpiece velocities are required for the generation of good
welding properties but matching conditions of local impact velocity and angle also
have to be granted. In order to determine such parameter constellations the model
experimental setup can be used as it allows a separate modification of the impact
velocity and impact angle.
The tooling system employed permits reference tests to be carried out using
identical tool geometries in the technically comparable procedure of hydro-
forming (IHU). For the die-forming tests, the aluminum alloy (AlMgSi1) was se-
lected as the material. During the testing, the tool's internal contour was modified
by means of exchangeable tool inserts. The inserts' internal contour exhibited a cy-
lindrical form, whose diameter was increased step-by-step. It is shown that using
AlMgSi1, a maximum diameter of 48 mm can be obtained without cracking.
23 Investigation of the Complex Interactions during Impulse Forming 509
Fig. 23.15 Effect of initial pressure on the workpiece surface (left) and quality of the form-
ing for increasing initial pressure (right)
During the tests, failures also occurred due to crack formation when using the
large die-inserts. Owing to the cracking, the detonation gases escape into the eva-
cuated tool space and then leak out through the tool's mating surfaces. This pro-
duces combustion marks of up to 3 mm deep in the steel inserts or, as may be the
case, in the tool's mating surfaces.
During the static reference tests using the hydroforming test tool, it was not
possible to successfully carry out experiments using even the smallest tool-insert
with a 44 mm diameter. By using an Al 99.5 tube, it was possible to form a
workpiece to this maximum diameter although, here, failure had also been ob-
served already for slightly larger die-inserts. To guarantee comparability, no axial
forces were applied during the hydro-forming tests. The die-force was only used
for sealing.
23.4 Conclusions
To investigate the complex interdependencies during the impulse forming of tubu-
lar parts, complementary experimental and numerical investigations were carried
out considering two different sub-systems. Sub-system 1 considers the workpiece
510 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
References
[1] Lesemann, M., Sahr, C., Hart, S., Taylor, R.: SuperLIGHT-CAR – the Multi-Material
Car Body. In: Proceedings of the 7th LS-DYNA Anwenderforum, Bamberg (2008)
[2] Kleiner, M., Geiger, M., Klaus, A.: Manufacturing of Lightweight Components by
Metal Forming. In: Annals of the CIRP Manufacturing Technology, 53rd General As-
sembly of CIRP, vol. 52(2), pp. 521–542 (2003)
[3] Psyk, V., Risch, D., Kinsey, B.L., Tekkaya, A.E., Kleiner, M.: Electromagnetic form-
ing – A review. J. Mater. Process. Tech (2011),
doi: 10.1016/j.jmatproctec.2010.12.012
[4] Beerwald, C.: Grundlagen der Prozessauslegung und –gestaltung bei der elektromag-
netischen Umformung. Dr. Ing. Dissertation, TU Dortmund (2005)
[5] Kaden, H.: Wirbelströme und Schirmung in der Nachrichtentechnik. Springer, Berlin
(1959)
[6] Weber, M., Hermes, M., Brosius, A., Beerwald, C., Gersthteyn, G., Olivier, H.,
Kleiner, M., Bach, F.-W.: Process Investigation of Tube Expansion by Gas Detona-
tion. In: Proceedings of the International Conference on High Speed Forming
(ICHSF) 2006, pp. 161–174 (2006) ISBN 3-00-018432-5
23 Investigation of the Complex Interactions during Impulse Forming 511
[7] Esser, B.: Die Zustandsgrößen im Stoßwellenkanal als Ergebnisse eines exakten Rie-
mannlösers. Dissertation, RWTH Aachen (1991)
[8] Bathe, K.J.: Finite Elemente Methoden, 2nd edn. Springer, Berlin (2002)
[9] Psyk, V., Beerwald, C., Homberg, W., Kleiner, M.: Extension of Forming Limits by
Using a Process Combination of Electromagnetic Forming and Hydroforming. In:
Proceedings of the 8th International Conference on Technology of Plasticity, ICTP
(2005) ISBN 88-87331-74-X
[10] Psyk, V.: Prozesskette Krümmen – Elektromagnetisch Komprimieren – Innenhoch-
druckumformen für Rohre und profilförmige Bauteile. Dr. Ing. Dissertation, TU
Dortmund (2010) ISBN 978-3-8322-9026-9
[11] Gottstein, G., Shvindlerman, L.S.: Grain boundary migration in metals: thermody-
namics, kinetics, applications, vol. 385. CRC Press, Boca Raton (1999)
[12] Johnston, W.G., Gilman, J.J.: Dislocation Velocities, Dislocation Densities and Plas-
tic Flow in Lithium Fluoride Crystals. Journal of Applied Physics 30(2), 129 (1959)
[13] Gottstein, G.: Physik. Grundlagen der Materialkunde. Springer, Berlin (2001)
[14] Roos, A.: Fast-moving dislocations in high strain rate deformation. Groningen Uni-
versity Press (1999)
[15] Alshitz, V.I., Indenbom, V.L.: Dynamic dragging of dislocations. Soviet Physics Us-
pekhi 18(1) (1975)
[16] Mattissen, D.: Insitu Untersuchung des Einflusses der Tripelpunkte auf die Korngren-
zenbewegung in Aluminium, Dissertation, RWTH Aachen (2004)
[17] Czubayko, M.: Korngrenzenbewegung in Aluminium und Zink, Dissertation, RWTH
Aachen (1998)
[18] Vovk, V.T.: Gasexplosion als Werkzeug in der Fertigungstechnik. Habilitation, Mag-
deburg, Univ. (1999)
[19] Psyk, V., Gershteyn, G., Demir, O.K., Brosius, A., Tekkaya, A.E., Schaper, M., Bach,
F.-W.: Process Analysis and Physical Simulation of Electromagnetic Joining of Thin-
Walled Parts. In: Proceedings of the International Conference on High Speed Form-
ing, pp. 181–190 (2008) ISBN: 3-9809535-3-X
[20] Kreimeyer, M., Wagner, F., Zerner, I., Sepold, G.: Laser beam joining of aluminium
with titanium with the use of an adapted working head. In: DVS-Berichte, vol. 212,
pp. 317–321. DVS-Verlag, Düsseldorf (2001)
Author Index
a posteriori error estimate 54, 61, 64, cold rolling 475, 479
132 compensation 398
residual 58 geometrical approach 259
abrasive machining 39 model based 258
acoustic emission 19, 104 compliance 246
adaptive finite element method 132 compliance simulation 185
additional stability lobes 238 comprehensive spatial multiple degrees
AE sensor 19 of freedom (MDOF) model 265
AFPM 385 contact
analogous models 35 bilateral 364
analytical-empirical process model 110 contact length 114
arbitrary lagrangian eulerian approach contact pressure 452
31 contact problem 54
axes dynamic 64
elasticities 248 elasticity 54
friction 54
frictional 60
back-step test 11
control system 111
backward can extrusion 426
convergence criterion 414
balance quality grad 330, 331
coolant forces 106
balancing 72, 74, 330, 352
cooling system 450
dual-plane balancing 332, 334
coulomb's Law 125, 131
single-plane balancing 331, 332,
coupled model 110
358
coupled simulations 48
balancing plane 331, 332, 334
coupling 345
bending vibration 184
coupling of models 44
blank holder 388
coupling of the models 302
bulk forming 40
critical chip thickness 272
critical depth of cut, 232
Campbell diagram 279 cross rolling 476, 478, 489
chatter 47, 179 cutting edge radius 273
chatter criterion 206, 278 cutting force 253
chatter detection 206 action reaction law 216
chatter frequencies 278 assumptions 217
chip longitudinal section 113 comparison 369
chip thickness modulation 189 modular model 366
chisel process structure interaction 215
orientation 371 spectrum 221
circularity test 10 temperature 210
cold forging 419, 420 uncut chip thickness 215
cold forming 475, 487, 489 cutting force coefficients 36
516 Subject Index
high gearing 475, 476, 478, 479, 480, measurement errors 208
489 mechanical (tactile) techniques 24
high performance cutting 179 mechanistic cutting force models 36
high-speed press 404 meshing 487, 488
Hopf instability 277 metal cutting simulation
hot rolling experiments 448 FEM 367
hydrodynamic pressure 107 friction conditions 368
hysteresis 6 material laws 367
metal forming 40
inconel 718, 109 metal forming processes 403
incremental forming 487 metrological acquisition 188
infrared radiatio 21 micro cutting 265
instability islands 238 micro force model 342
integrated FE simulation model 425, micro milling machine tool 266
426, 428, 429, 432 micro sensors 21
interacting system 45 micro turning 341
interaction effects 441 mixed formulation 60
interferometry 23 microscopic approaches for grinding
inverse problem 54, 67, 69 models 40
milled surfaces 280
milling 227, 245
joint
milling system 45
actuated 248
modal analysis 182, 183, 266
virtual 248
robot 254
modal analysis 9, 124
kinematics 10 modal models 237
modal parameters 36
lagrange multiplier 60 mode shapes 269
laser interferometer measurement 12 model coupling 384
laser tracker system 483, 484 model integration 48
light fringe projection 24 multi body simulation 110
linear cutting force models 37 multi degree of freedom system 229
multi frequency solution 226
multi body models 30
machine behavior 480
multi body simulation 30
machine concept 479, 480, 490
multi body system 213
machine tool behavior 123
multi body system
machining tests 204
contact 363
macroscopic approaches for grinding
MBSim 362
models 40
multi body system
main spindle systems 180
dynamics 245
mass spring damper systems 35
multi dexel model 251
material behavior 488
material flow 476, 487
material laws NC-shape grinding 121
metal cutting simulation 367 Neo-Hookean model 33
material removal 114 net-shape production 419
calculation 245 newmark-scheme 34
material removal model 217 nonholonomic constrain 271
material removal process 348 nozzle 114
measurement numerical simulation 218
temperature 369 nyquist criterion 227
518 Subject Index