Sunteți pe pagina 1din 26

International Journal of Machine Tools & Manufacture 45 (2005) 301326 www.elsevier.

com/locate/ijmactool

Tracking and contour error control in CNC servo systems


R. Ramesh, M.A. Mannan*, A.N. Poo
Department of Mechanical Engineering, National University of Singapore, 10 Kent Ridge Crescent, Singapore 119260, Singapore Received 21 May 2004; accepted 3 August 2004 Available online 8 October 2004

Abstract The dynamic response of a machine tool is a complex interaction of several factors that includes the basic design of the machine as well as the capability and performance of the corresponding axis servo system. In terms of the performance of the servo system, tracking and contour errors are two important aspects that signicantly affect the machine tool. Extensive research has been conducted over the last several decades into the study of these error components. A number of different approaches have been proposed by various researchers that attempt to effectively address this problem. This paper attempts to study the work that has been carried out in minimising tracking and contour errors in CNC machine tools. q 2004 Elsevier Ltd. All rights reserved.
Keywords: Tracking error; Contour error; Controller; Error control; Feedback

1. Introduction While basic machine tool errors form one of the major sources of inaccuracy in machine tools, achieving high precision in actual machine tool performance depends a lot on the dynamic performance of the individual axis controllers as well. During the early 1970s, one of the important advances in the philosophy of machine tool control was the use of computers in numerical control of the machine. CNC systems have since then became very popular on account of their exibility, low investment and also the declining cost and availability of microprocessors. CNC systems primarily use both velocity and position control loops in monitoring and controlling the machine tool axes with the velocity feedback normally provided by a tachogenerator and the position feedback by digital encoders or other feedback devices [1]. The design of servo control systems that generate accurate coordinated multi-axis motion is therefore of great importance in high precision machining applications. As several factors such as the dynamic response of the machine axes, stability of the control systems, time and frequency response of the servo system, etc. are involved in determining the nal
* Corresponding author. Tel.: C65 874 6889; fax: C65 779 1459. E-mail address: mpemma@nus.edu.sg (M.A. Mannan). 0890-6955/$ - see front matter q 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijmachtools.2004.08.008

performance of the machine tool, the performance criteria for such servo control systems needs to be clearly established as well as closely monitored both to serve as a benchmark as well as to continuously improve upon the capabilities of the machine tool. Tracking error and contour error are two very important aspects that need to be effectively handled by any servo control system. Researchers over the past several decades have developed very innovative and ingenious ways of addressing these two vital issues. In this paper, the various techniques that have been developed in minimising and eliminating tracking and contour errors are reviewed.

2. Control systems for machine axes Many machining applications involve the generation of complex proles or, at the very least, need to synchronize the motions of the different axes to generate the required machined surface. In a CNC machine tool, this is usually done with the help of a multi-axis servo control system that consists of an interpolator and several machine axis controllers. The interpolator generates the desired tool motion relative to the workpiece and then decomposes the desired motion into the reference position commands for the separate driving axes. Actual positioning of the axes is done

302

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

by the individual axis controllers. The function of an axis controller is to provide the appropriate drive signal to the actuator, usually in the form of an electric motor, so that the actual position of the axis accurately tracks the desired axis reference command to eliminate the position error along each driving axis [2]. The purpose of axis control systems is to control the position and velocity of the particular machine tool axis. In general, each machine axis in a CNC machine is separately driven and follows the command signal produced by the respective interpolator. Controllers are of two types: the open-loop or the closed-loop. Open-loop controls use stepping motors to drive the machine axis while in the closed-loop control systems, a position feedback element, such as an encoder or a resolver, is used to provide information of the actual position of the axis during motion. There are two types of closed-loop position control systems, namely the point-to-point and the contouring systems. In point-to-point systems, the path of the machine tool axis from one point to the other and its velocity during motion are not signicant. This is because, in point-to-point motions, the tool is not in contact with, or machining, the workpiece during the motion. What is of interest is only the accuracy of positioning the axis to the desired end point. In contouring systems on the other hand, the path that the cutting tool takes, relative to the workpiece, called the tool path, is of paramount importance. This is because the contour, or prole, of the machined part depends primarily on this path. A typical closed-loop, axis control diagram for a contouring system is as shown in Fig. 1 [3]. One of the signicant requirements of a CNC machine is to generate coordinated movement of the different axes of the machine so as to trace a predetermined path. A typical 2axis contouring system is shown in Fig. 2. The interpolator handles the task of generating the reference position commands for the coordinated movement of the different axes. The most signicant factor in the performance of contouring systems is the accuracy of the overall system or the contour error of the system. This is dened as the actual difference in distance between the programmed path and the actual path. The problem of contour error arises in linear

motion, circular motion, in the cutting of a corner radius and in the cutting of any desired contour. In order to reduce contour error in linear motions, it is necessary to have high average gains and also to match the individual axis gains. The difference in axis open-loop gain values represents the mismatching of the system, which creates the contour error. In the case of circular motion, for perfectly matched axes, radial error is very small when the damping factor is 0.707. However, due to mismatch in the axis dynamics, the contour error leads to the generation of an elliptical prole. In the case of cutting a corner, if the gain is too low, the axis dynamics will be overdamped, in which case, the following error of the axis is large and the system is sluggish, which results in an undercut. If, however, the gain is too high, the system will be underdamped and the following error and thus the axis overshoots the desired prole causing an overcut. Lower axis feedrates will reduce the three types of contour error. To obtain high contouring accuracy at high speeds, the system should be critically damped and the system gains well matched [3]. High speed machining techniques require faster feed motion between the tool and workpiece, in proportion to the increased spindle speed. However, due to the limited bandwidth achievable by using only proportional (P), proportional-derivative (PD), or proportional-integralderivative (PID) types of servo controllers, which are the most common ones used in industrial CNCs, there will be tracking errors in each axis as the closed-loop control system is not able to follow the rapidly varying position commands. On the other hand, the friction in the guideways as well as cutting forces, constitute disturbances to the servo loop, which are detrimental to the tracking accuracy. Furthermore, even if the servo controller is tuned to maintain accurate tracking at high speeds, the performance may degrade if the drive parameters change due to varying workpiece mass and lubrication conditions. The disturbance and parameter variation sensitivity is much more severe in direct drive systems compared to ball screw drives and a well designed servo controller must be capable of handling both types without requiring any major changes [4].

Fig. 1. Control loop of a contouring system [3].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

303

Fig. 2. Two-axis contouring system [3].

3. Research in machine tool axis control Errors affecting a multi-axis machine are dened as the difference between the real and measured position and orientation of the end effector. These errors cause a deterioration of the machine performance, which in turn directly inuences the nal product specications. To maintain high quality of performance for a machine, these errors have to be detected and eliminated. Machine tool control systems maintain two control objectives: to follow a pre-specied trajectory as closely as possible, and to maintain a pre-specied advancing speed. These systems break the desired trajectory into small segments, and step along these segments to form the path. When the trajectories are multi-dimensional, the systems divide the motion into simultaneous movements in different axes. The drivers of different axes have to move synchronous with one another to obtain the desired trajectory. If the controllers operate independent of each other, then any load disturbance or any difference in the performance of each axis may cause contour error [5]. Hence, two major approaches have emerged for improving the control performance in drive systems. The rst approach, known as tracking control concentrates on only reducing the tracking error in each axis, which indirectly results in the reduction of the contour error. The second approach, known as contouring control aims at estimating the contour error in real-time and using this estimate in the feedback control law. Poo et al. [6] studied the effect of dynamic errors in twoaxis, contouring systems. Two specic cases of constant acceleration and constant velocity (shown in Figs. 3 and 4) along a straight line contour were studied. It was found that when the axis gains of the two axes (Kx and Ky) were perfectly matched, the contour error was zero even though there were individual axis errors with respect to time. However, when there is mismatch between the axis

characteristics, there results a contour error. In the case of circular contours, it was found that for perfectly matched axes, the amount of radial error was very small and a perfect circle was generated with a radius larger or smaller than the desired radius depending upon the amount of damping and the angular velocity. Mismatch in system gains however has considerable effect on the maximum radial contour error with the steady-state contour being elliptical in shape. Many researchers have developed control algorithms that primarily seek to improve the tracking accuracy for an individual axis. Traditional algorithms are based on the feedback principle. In addition, feedforward control algorithms have been implemented to improve the tracking performance. A signicant contribution has been made by Tomizuka, who proposed a zero phase error tracking controller (ZPETC). On the basis of the ZPETC method, some variational or accessory algorithms (e.g. adaptive ZPETC, etc.) have been proposed [2].

Fig. 3. Contour Error-Constant acceleration along a line [6].

304

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 4. Contour Error-Constant velocity along a straight line [6].

Although the tracking performance for each individual axis can be signicantly improved by the above methods, good overall control performance for the multi-axis machine tool is not always guaranteed. A typical performance index for evaluation of multi-axis servo control is the contour error, which is the deviation from the desired tool path. To conduct an effective reduction of the contour error, Koren proposed a cross-coupling controller (CCC) that is constructed between and parallel to the axial controllers. A typical cross-coupling controller consists of a real-time calculation of the contour error and a control law to eliminate the contour error. Based on the concept of the CCC method, numerous cross-coupling controllers (with different contour-error models and/or controller laws) have also been proposed [7]. These are the two broad areas in which research has been going on to improve the performance of control systems. Thus, control of system performance involves careful attention to two critical areas namely control of tracking or following error and control of contour error. In order to improve the performance of machine tool axis drives, many variations and improvements in these areas have been introduced to the design of axis control systems over the last several years. A study of research in these two major areas has been presented in subsequent sections. 3.1. Tracking/following error control systems Tomizuka [8] developed the zero phase error tracking control (ZPETC) system for tracking with zero error. In order to achieve superior tracking, a feedforward controller is required in addition to the feedback controller. Assuming that the feedback controller already exists, a feedforward controller was set up that utilised the future desired output to compensate for the delay in the closed-loop transfer function. The controller thus provided perfect tracking as long as all the initial conditions were zero requiring that all the closed-loop zeros to be inside the unit circle in the z-plane. The system became highly unstable if any of

the zeros were outside or on the unit circle or even on the negative real axis and close to K1. To circumvent this, the ZPETC system was developed that assures zero phase error. The ZPETC utilises the (dCs)-step ahead desired output, yd(kCdCs), where d is the number of delay steps in the closed loop transfer function and s is the number of closed loop zeros which are unacceptable for pole/zero cancellation. ZPETC was examined in detail for a motion control problem, where the controlled plant was a pure double integrator, the input is the force (torque) and the output is the position (angular position). The tracking performance was seen to be excellent both in terms of tracking error as well as smoothness of velocity. Tomizuka et al. [9] developed a tool positioning system for non-circular cutting in a lathe. The tool carriage was driven by a dc motor in the set up as shown in Fig. 5. A low spindle speed of 93.7 rpm was used for cutting. Both feedforward and feedback controllers for carriage motion were used in the radial direction normal to the workpiece surface for successful non-circular cutting. The feedback controller was for regulation against disturbances and the feedforward controller for positioning the tool along the desired time varying output signal. Two feedforward controllers, one based on magnitude and phase compensation for a particular frequency and the other based on ZPETC were attempted. Because of the low cutting speed, machining tests were performed on polyethylene workpieces. It was found that ZPETC was an effective method for non-circular machining with the actual position being very close to the desired position. Although the experimental work supported the effectiveness of the ZPETC, the maximum experimental error was 0.635 mm. The largest diameter machined was 44.45 mm.

Fig. 5. Experimental set up for non-circular cutting [9].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

305

Tsao and Tomizuka [10] developed an adaptive zero phase error tracking algorithm in order to allow the output of a plant with stable and unstable zeros to track a time varying desired output without any phase error. In the ZPETC, the closed loop system, which consists of a plant and a feedback controller, is asymptotically stable and is described by yk qKd BqK1 qKd BKqK1 BCqK1 Z Z rk AqK1 AqK1 (1)

where qK1 is a one step delay operator, r(k) and y(k) are the reference input and plant output, respectively, AqK1 Z 1C a1 qK1 C .C an qKn ; BqK1 Z b0 C b1 qK1 C .C bm qKm ; B(1)s0 and B(qK1) is factorised into BK(qK1) and BC(qK1). The zero phase error feedforward controller (ZPETC) is rk Z AqK1 BKq ym k C d BCqK1 BK12 (2)
Fig. 6. Experimental set up for verifying EBZPETC algorithm [11].

where ym(kCd) is a d-step advanced bounded desired output signal and BK(q) is obtained by replacing (q)K1 in BK(qK1) by q. From these two equations, the overall transfer function becomes yk BKqK1 BKq Z rk BK12 (3)

When the plant parameters are unknown, the feedforward controller as given above cannot be directly implemented and therefore was made adaptive. The closed loop system was separated into a known part and an unknown part as given below yk qKd BqK1 BKqK1 B qK1 0 0 rk AqK1 A0 qK1 (4)

The feature of this adaptive control system is that the feedback loop is unperturbed by the adaptation algorithm and hence the adaptive feedforward controller can be switched on or off as a module. System stability is easily established because of the moving average characteristics of the feedforward controller and because the feedback loop is untouched. The preview action of the feedforward controller upgrades the tracking performance. It was shown by simulation that superior tracking performance was maintained when unmodelled dynamics were stable and moderately fast. Torfs et al. [11] introduced an improved algorithm called the extended bandwidth zero phase error tracking control (EBZPETC). This algorithm computes the feedforward signal in two steps namely by rst generating a feedforward signal based on the ZPETC algorithm and then compensating the remaining error by adding additional feedforward signals which repeatedly reduce the tracking error proportional to 32, 34, 36,., where 3 is the ZPETC tracking error. This algorithm was veried on the experimental test set up shown in Fig. 6. The end point of the exible beam

was to be controlled. The end point position was not directly measured but was calculated as a linear combination of beam deection and motor position. A tracking test was conducted with a 9th order polynomial trajectory having zero start and end velocity, acceleration and jerk. A large tracking error of 1402 mm was observed using the optimised feedback controller without feedforward. Feedforward was then calculated in three different ways namely (a) velocity and acceleration, (b) based on ZPETC algorithm and (c) based on EBZPETC algorithm using one additional feedforward term. The maximum tracking error using velocity and acceleration feedforward was 100 mm, that using ZPETC was 60 mm and that using the EBZPETC algorithm was 8 mm. In some approaches like the Generalised Minimum Variance (GMV) and the pole-placement method that have been applied in industries, it was found that excellent control behaviour was achieved provided that the various design parameters were chosen properly. To overcome the drawbacks of these systems it was found that an algorithm was required that retained the design exibility and performance of the earlier methods yet was robust against incorrect choice of parameters. The Generalised Predictive Control (GPC) was one such approach attempted. In such a system, the controller predicts the changes in the controlled variable that are likely to occur in the future using present process knowledge and candidate control actions [12]. The GPC algorithm was found to be superior to many accepted adaptive controllers when used on a plant that has large dynamic variations [13,14]. The GPC algorithm was applied to industrial applications in order to evaluate its performance over conventional PID controllers. In a robot arm application, wherein accurate, high speed trajectory following of the arm tip was required, it was found that the PID controller was difcult to tune because of the complex

306

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

nature of the arm dynamics. The GPC on the other hand provided tighter control and nearly ideal tracking for sine wave variations. Simulation tests have shown that GPC can cope with the control of complex processes under realistic conditions. As it is relatively insensitive to basic assumptions (model order, etc.) about the process, GPC can be easily applied in practice without a prolonged design phase. These features ensure that the method provides an effective approach to the adaptive control of an industrial plant. Boucher et al. [15] developed a generalised predictive cascade control (GPCC) for machine tool drives using the basic idea of the GPC system enabling control of two different variables together. The plant is divided into two parts. The output of the rst part called the inner system is the rst variable to be controlled and represents the input to the second part called the external system. The output of this second part is the second variable to be controlled. In order to do this, two GPC algorithms were computed. The results on tests conducted on a brushless motor used to drive machine tool axes indicate that the predictive control enables good performance in terms of rapidity and cancellation of steady state errors and tracking errors. In addition, this law was found to be well adapted to servomechanisms using brushless motors, which need very small sampling period that is allowed by the simplicity of this control algorithm. Weck and Ye [16] designed a compensation method for tracking errors using the inverse compensation lter (IKF) control strategy. The IKF works according to the feedforward principle. The idea of the feedforward controller was to inuence the transfer error of the proportional controlled position control loop with a suitable command variable distortion. One of the essential characteristics of the IKF solution is the integrated ltering away of high frequency signals to the inverted system model. This was found to be very effective for the smoothening of discontinuities. Tomizuka et al. [17] developed the discrete-time repetitive control to handle periodic signals and applied the same to track-following in a disk-le actuator system. In the disk le servo system, the read/write head needs to be kept on a circular track on the spinning track. The dynamics of the disk le actuator were assumed to be second order and a PD controller was selected. The digital repetitive controller was added to the existing servo system as a plug-in module. The periodic disturbance comprised a constant signal and four out-of-phase sinusoidal signals from the rst to the fourth harmonic of the spindle frequency. The repetitive controller was found to reduce the tracking error due to the periodic; the sampled error converged to zero very rapidly. Chew and Tomizuka [18] developed a modied discretetime prototype repetitive controller for a disk-le actuator system application. The dynamics of the disk le actuator were assumed to be second order and the analog controller was selected to be of PD (proportional plus derivative) type. The modied prototype repetitive controller was added to

the servo system. The amplitude of the tracking error with the modied repetitive controller was only about 1/3 that without the repetitive controller. It was also seen that although a perturbed internal model led to imperfect tracking, the residual errors could be reduced by increasing the repetitive gain. Haack and Tomizuka [19] discussed the effect of adding zeroes to two types of feedforward controllers namely the stable pole zero cancelling (SPZC) controllers and the zero phase error tracking controller (ZPETC). When there are uncancellable zeroes in the feedback system, it was seen that additional zeroes in the feedforward controller could reduce the tracking/contour error to zero for low and medium frequency trajectories and also reduce the error signicantly for trajectories with high frequency components such as sharp corners. However, since the SPZC controller and the ZPETC are based on pole/zero cancellation and phase cancellation, their performance depends upon the accuracy of the plant model used in the design. Erkorkmaz and Altintas [4] proposed an axis tracking control scheme (shown in Fig. 7) for high speed tests on a three-axis vertical machining centre. The control scheme was tested in tracking circular and diamond shaped trajectories. It was seen from the test results in air cutting that both the RMS and maximum values of tracking and contour error exhibited a clear descend with the activation of each component in the control scheme. The largest improvement was seen with the addition of the ZPETC. However, it should be noted that feedforward controller on its own would not have been so effective if the parameter variations and disturbances were not handled by the disturbance cancellation feature. On the other hand, the feedforward friction compensation did not appear to be providing a dramatic improvement on the tracking performance, but this was accounted to the effect of friction being relatively small compared to the process uncertainty and measurement noise. For drive systems with higher quality feedback devices, the advantage of pre-compensating for the friction would be much more evident. Lim et al. [20] developed a simple and effective algorithm to reduce the tracking error of the table position with motor encoder information only. A torsional displacement feedback control was proposed. To use this algorithm, the torsional displacement had to be measured or estimated. The laser interferometer was used for validation of the test result. The XY table of a vertical CNC machining center was used for the experiments. Experiments were performed for a circular trajectory with an 8 mm radius. With only the motor position feedback, friction caused large quadrant glitches when the axes changed directions, and its effects on the motors position and the tables position were quite apparent. The torsion of ballscrew caused a large error in the tables position. The quadrant glitches in the motor position were compensated for quite well when the feedforward friction compensator was added. However, the error of the tables position due to the torsion of

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

307

Fig. 7. Axis tracking control scheme [4].

the ballscrew was quite large. The error of the motors position became larger, but the maximum error of the tables position reduced considerably when the additional torsional displacement feedback was introduced. The errors thus obtained were comparable with the position feedback as obtained using a laser interferometer. Van Den Braembussche et al. [21] proposed a controller consisting of feedback of the difference between the actual state of the system and a reference state calculated using a reference state observer and feedforward consisting of inverse model pre-ltering. Ripple compensation was based on an experimentally identied rst order approximation of the magnetic ripple (cogging) and electro-magnetic ripple (force ripple). This was implemented in the position controller as an additional feedforward term (as shown in Fig. 8). Control experiments on two linear synchronous

motor machine tool axes showed that the proposed advanced controller yielded acceptably small tracking errors if motor ripple compensation was applied. It was found by Renton and Elbestawi [22] that the largest source of path error, given a well-tuned system, was a commanded path that exceeded the velocity or acceleration capabilities of one or more of the axes. Minimum-time path optimisation determines the maximum feedrate along an arbitrary multi-axis path that is possible without exceeding the capabilities of any axis. These capabilities are limited by the voltage and current limits of the amplier/ motor combination. Error was minimised by determining the capabilities of the axes, and modifying the velocity (feedrate) of the tool along the path such that these measured limits were not exceeded. It was also found that cutting time could be signicantly reduced without

Fig. 8. Control scheme using ripple model feedforward [21].

308

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

compromising path accuracy by using the entire capability envelope of the system for path planning, instead of xed acceleration and velocity limits. A Minimum-Time Tracking Controller was developed which focussed on the primary sources of path error in high-speed servo applications-specically, limits on ampliers, system inertia, and damping. This controller was computationally efcient and robust. The advantages of the controller were seen in its consistently superior disturbance-rejection properties. Torfs et al. [23] presented a new digital algorithm for quasi-perfect tracking control (QPTC) of non-minimal phase systems. The QPTC algorithm consisted of two steps namely a feedforward calculation based on the ZPETC algorithm followed by a compensation of the remaining tracking error with an additional feedforward. First, a closed form expression was derived for the relation between the z-transform of the desired output signal and the z-transform of the tracking error resulting from the ZPETC algorithm. From this expression the time and frequency domain properties were derived. These properties give insight into the parameters that inuence the tracking error: the prole for the desired time varying output signal, the effect of the sampling rate and the location of the unstable zeroes in the z-plane. The QPTC algorithm was then veried on simulation results. The results showed a drastic improvement of tracking error over the ZPETC algorithm especially in the case of unstable zeroes close to zZ1. Tomizuka et al. [24] identied the need for contour error control in automatic welding applications, wherein, in order to guarantee uniform quality of the weldments, the relative position error between the seam and the welding torch needs to be kept small while maintaining a uniform tracking speed. To accomplish this, preview control concepts were used in addition to conventional PID control. A separate driving mechanism with a controller was provided to each axis in the 2-D space. Coordination of the X and Y axis controllers was accomplished by properly generating reference positions for both the X and Y axes. The basic schematic of the tracking system used is as shown in Fig. 9. The coordinator insures that the two adjacent reference

points on the weld path are uniformly spaced in order to guarantee a constant (desired) tracking speed. Tomizuka et al. [25] carried out an experimental evaluation of the preview control scheme as developed above. The complete hardware system consisted of a twoaxis welding manipulator, electronic interface hardware and a microcomputer as shown in Fig. 10. Two added features were introduced for improving the performance. A velocity estimation program was introduced using a Kalman lter and a corner preview program was added to decrease and increase the relative velocity of the welding torch and workpiece for reduced path error at sharp corners. The Kalman lter produced smoothed position and velocity data for use in the control algorithm. The corner preview scheme substantially reduced both the positioning and velocity errors at sharp corners. The concept of the feedforward control as extended to preview control was illustrated by Tomizuka and Rosenthal [26]. A digital state vector feedback controller, called the PIDP controller, with integral action (PID controller) and preview actions (P) was derived based on linear quadratic optimal control theory. Herein, future information of the disturbance as measured in advance is utilized. It was found that preview of disturbances drastically improves the performance of the system. Tomizuka and Fung [27] presented the design of a feedforward/preview (F/P) controller based on parameter optimization. This controller assumes that the feedback controller gains are xed a priori so that the feedback system is asymptotically stable. The role of the feedback controller is to assure good system performance under known and unknown disturbances as well as modelling errors. The F/P controller on the other hand is for a particular disturbance that is measured/ previewed. The feedback control gains affect the closed loop stability as well as sensitivity to variations in the plant parameters. The F/P controller appears outside the closed loop and does not affect system stability. Given the system with the predetermined feedback controller, the F/P control gains are optimised so that a quadratic cost criterion, which penalises error due to disturbance as well as incremental changes in the F/P controller output, is minimised.

Fig. 9. Schematic diagram of the 2-D welding head tracking system [24].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

309

Fig. 10. Hardware system for two-axis welding manipulator [25].

Pak [28] developed an adaptive matching and preview controller for feed drive systems with the objective of eliminating dynamic parameter mismatch between axes feed drives by using a single discrete time reference model of the feed drive speed regulation loop. To achieve this, the speed regulation loop for each axis was made adaptive using a direct approach and for contour tracking purposes, a preview feedforward controller was combined with a conventional state feedback control to form the input to the reference model. This control scheme was simulated for an axis feed drive of a CNC lathe. It was observed that considerable following errors result when no preview action was employed. On the other hand, the tracking response with preview gains showed substantial improvement. The response of the preview controller with the adaptive control switched off showed considerable following errors since the actual feed drive dynamics no longer follows the reference model. However, when the adaptation was switched on, almost perfect model following was observed. 3.2. Contouring error control systems Several specic attempts have been similarly made in the elimination of contour error. To provide a direct and effective elimination of deviation error, orientation error and tracking-lag error that are main concerns in ve-axis tool-path tracking control, a new tool-path control scheme was presented by Lo [2]. The control system, as shown in Fig. 11, consisted of a real-time transformation between the drive-coordinate and the workpiece-coordinate bases, an error model for calculation of the deviation error, etc.

and control laws to eliminate them. The proposed control system constructs a global and coupled loop to achieve an effective control of the overall performance in terms of the deviation error, the orientation error, and the tracking lag. The deviation error, etc. are error components dened in the workpiece coordinate basis. In contrast, the fed back position signals (Pd) and the control signals (Ud) sent to the axial drives are both dened in the drive-coordinate basis. Coordinate transformations were introduced to the proposed control system. The proposed tool-path control method was evaluated and compared with the conventional method on the basis of on their capability in elimination of the deviation error, the orientation error and the tracking-lag error. The deviation error was found to be signicantly reduced by the proposed method for all the three surface proles tested. The orientation error was found to be signicantly reduced for the cylindrical surface and only slightly for the canonical and ruled surfaces. Tracking error was, however, not signicantly reduced [2]. In a signicant breakthrough to the design of axis control systems to minimise contour error, Koren [29] proposed a cross-coupled controller (CCC) for biaxial control of machine tool axes. The rst such system was developed by Sarachik and Ragazzini in 1957 with a storage device f(x) that provides the desired value of y as its output for an input x. The Y-axis has a closed loop control while the X-axis has an open-loop with the Y-axis error as input to the X-axis. Although this non-symmetrical system was found to be suitable for some contouring operations, it was not found to be practical under most circumstances.

310

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 11. Proposed control system for ve-axis machine tools [2].

In a sampled-data type CNC system, each axis is controlled independently in a loop closed by software within the computer. The axis is driven by a servomotor with an encoder feedback as shown in Fig. 12. The control program compares the two types of input namely the reference number R proportional to the required speed of the axis and the feedback signal F proportional to the actual speed of the axis. The difference represents the speed error. In the CCC, the control program is fed by two reference numbers R1 and R2 proportional to the required speeds of axes 1 and 2 as shown in Fig. 13. The operation of the CCC is based on providing corrections proportional not only to the individual axial errors but to the contour error as well [29]. Doraiswami and Gulliver [30] proposed a control strategy using robust servomechanism control theory for CNC machines in order to improve precision and rapidity in machining a given contour. The conventional controller that was commonly used in machine tools was the PID type controller that ensured zero tracking error only for constant

command signals. In such controllers, contouring precision was achieved by lowering the velocity of tool movement, matching the parameters of the control system in each axis and by increasing the closed loop control gains. The mathematical model of the NC machine axis consisting of a ballscrew and nut feed drive powered by a DC motor was developed. Based on this, a state variable representation of the composite system consisting of the motor, drive amplier, feed drive and load was arrived at. In a contouring system, the cutting tool experiences a varying reaction force depending upon the feedrate, depth of cut, etc. This force was also accounted for in the mathematical model of the machine tool axis. In order to ensure that the closed loop system is stable and the tracking error is less than the specied tolerance, a linear, discrete, time-invariant servomechanism as shown in Fig. 14 was developed. The control strategy includes a tracking error driven lter termed the servocompensator in a feedback loop and stabilises the system. This strategy was found to be superior to conventional strategies. The contouring task was performed

Fig. 12. Conventional CNC loop [29].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

311

Fig. 13. Basic structure of cross-coupled control system [29].

accurately for varying speeds and in the presence of disturbances. Improved response was shown for both analytic and piecewise analytic signals. Srinivasan and Kulkarni [31] carried out stability analysis and experimental evaluation of a cross-coupled controller for a precision two-axis table. The table was driven by precision ground ballscrews supported on ball bearings. DC servomotors were directly coupled to the axes and were powered by single-phase, full-wave SCR ampliers. Three different representative trajectories namely a straight line at 458, a right-angled cornering contour and a circular contour were used to evaluate the contouring performance of the controllers. Both an uncoupled as well as a cross-coupled controller was studied. In the uncoupled controller, the corrective control action for each axis was dependent only on the commanded input and response for that axis. In the cross-coupled controller, the position loops were closed with proportional controllers. An additional

input term proportional to the component of the contour error along the corresponding axis was included and resulted in the cross-coupling of the two axes as shown in Fig. 15. In executing straight line contours, the crosscoupled controller was found to reduce the contour error more rapidly than the uncoupled controller but was found to overshoot while the uncoupled controller exhibited slower and overdamped responses. In executing cornering contours, the cross-coupled controller was effective in reducing contour error without overshoot at lower speed but was oscillatory at higher speed. The uncoupled controller resulted in slower but well damped reduction of the error. Yamazaki et al. [32] developed a exible servo controller for machine tool and robot control so as to facilitate the application of advanced control schemes. A fully softwarebased control scheme which can be applied for DC, AC synchronous and AC induction servo motors was proposed as shown in Fig. 16. An equivalent current and voltage

Fig. 14. Machine axis control system using servocompensator and stabiliser [30].

312

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 15. Cross-coupled controller for two-axis table [31].

conversion method was introduced so that either type of motor could be controlled. A software-based, two-modes switching speed detection method was implemented for precise speed measurement without a speed sensor. The control hardware was designed and fabricated using a highly integrated 16-bit microcontroller. Successful control for step velocity input, constant speed and quick reverse were demonstrated for all the three types of motors. The exible servo controller (Fig. 16) was realised using a general microprocessor CPU and additional peripheral circuits consisting of primitive electronics. This implementation method, however, suffered from the fact that control performance was limited due to insufcient execution

speed. Compactness and cost-effectiveness of hardware were also limited. To circumvent this, a less-chip-count solution was envisioned through the application of ASIC (Application Specic Integrated Circuit) in order to integrate the exible servo controller into a single silicon chip. Based on an evaluation of the functional elements that had to be integrated into the chip, three kinds of ASIC chips in terms of cost effectiveness, performance and functional exibility were considered required. These were termed as FSP-1, FSP-2 and FSP-3, respectively, (FSP stands for Flexible Servo Peripheral) with each chip having a modular functionality of the exible controller as shown in Fig. 17. The PWM generation and a pre-settable up-down

Fig. 16. Software control scheme [32].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

313

Fig. 17. ASIC integration of the exible servo controller [33].

counting were integrated into the FSP-1 using the gate array technology. The FSP-2 featured the fact processing capability for the current loop control and was developed using the standard cell technology. The development of the FSP-3 was intended to take care of the position and speed control loop and was designed using a silicon compiler. The FSP-1 was fabricated using NECs gate array process. Through tests on the chip it was found that noise on the analog circuit was remarkably less. The FSP-2 was also successfully implemented and evaluated in terms of performance and size of the hardware system. The execution time of the current loop for AC motors was faster than achievable by the standard DSP coprocessor [33,34]. In order to eliminate the disadvantages posed by conventional electromechanical drive systems like nonlinearities such as backlash and hysteresis, Pritschow and Philipp [35] developed a linear drive system on the basis of a single-comb linear asynchronous motor. This drive system has a number of characteristics like high acceleration, load stiffness and positioning accuracy which are not generally reached by conventional drives. Although the cross-coupled controller was found to be effective in reducing contour error for multi-axis motion, increasing feedrates degraded contouring performance. In order to improve contouring accuracy for various types of contours due to varying feedrates, Chuang and Liu [36] developed a self-tuning adaptive control strategy. To obtain feasible solutions and to apply adaptive control techniques, a linear perturbed model between the contour error and the feedrate was established. This was done by introducing a feedback loop from the calculated contour error to

the commanded feedrate. Experiments were carried out to evaluate contour error during circular contouring as well as corner contouring. It was found that coupled controllers perform better than the uncoupled ones. In addition, the adaptive controllers perform better that the non-adaptive ones. The experiments indicated that the combination of cross-coupled and adaptive feedrate control improved substantially the contouring performance for multi-axial motion. The major drawback of the cross-coupling controller (CCC) is its low effectiveness in dealing with non-linear contours such as circles and parabolas. Another drawback is that even for linear contours, the contour error oscillates when the steady-state error goes to zero. Koren and Lo [37] introduced a new non-linear CCC that was based on control gains that vary during the contour cut. This controller was termed as Variable-Gain Cross-Coupling Controller. The gains are adjusted in real-time according to the shape of the contour. The structure of the variable-gain CCC is as shown in Fig. 18. A comparison of the contour error between the variable-gain CCC and a conventional controller are as shown in Figs. 1921. As seen from the gures, the error reduction on account of the variable-gain CCC was signicant. Due to their low rst natural frequency, conventional electro-mechanical drives with ballscrew mechanisms generally limit the velocity gain factor kv to values below 50. When using direct drives for the same applications, kv values can be achieved that are higher by a factor of 5 to 10. It was shown by Pritschow [38] that the kv value improves the path accuracy for a circular path according to the square law. Moreover, the use of a PI path controller reduced

314

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 18. Variable gain cross-coupling controller [37].

the settling time to zero inversely proportional to the kv value. It was observed that a completely new quality of dynamic accuracy, difcult to achieve with conventional axes, could be reached through the use of direct drives on account of the extraordinary inuence of the kv value on path accuracy. Makino and Ohde [39] developed a motion control system for the direct drive actuator. The direct drive (DD) actuator is a high-torque and low-speed electric motor which can drive a mechanism without a speed reducer and can reduce unfavourable effects of non-linear backlash

and low stiffness of gears thereby accomplishing high precision, high speed and reliable positioning. Two kinds of controlling methods were adopted to control the DD actuator. One was the I-PD control for larger load uctuations and the other was the P-ID control with torque feedforward control used for high frequency repetitious motion of stable load. The P-ID control was tested using several cam curves on two kinds of apparatuses namely

Fig. 19. Variable gain cross-coupling controller [37].

Fig. 20. Variable gain cross-coupling controller [37].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

315

Fig. 21. Variable gain cross-coupling controller [37].

a single arm robot and an indexing table. An exact response was obtained by using the P-ID position control. It was found that by using these methods, the DD actuator could achieve high repetitious motion excellently. Knapp and Weikert [40] developed a measuring instrument to check contouring performance of a machine tool in 68 of freedom by simultaneous measurement of the three translational and three rotational relative movements between the tool side and the workpiece side. The enhanced (C) coordinate (K) grid (G) measuring (M) system, called the system KGMC, is based on a Heidenhain grid encoder which is a two-dimensional linear scale. The set up of the KGMCsystem is as shown in Fig. 22. The grid plate is mounted on the work table. The two optical heads and the three probes are mounted in a xture which is held in

Fig. 22. The system KGMC[40].

the blocked tool spindle. Measurements were carried out at low and high feedrates on a horizontal and vertical machining centre. Deviation from the straight line in the XZ plane was found to be 1.5 mm and deviations from the straightness in the XY plane went up to 8 mm. Kulkarni and Srinivasan [41] developed the mathematical formulation of a class of cross-coupled compensators using an optimal control formulation. The block diagram of the controlled feed drive including optimal coupled control action is as shown in Fig. 23. The test contours used to evaluate the controller were (a) straight line contours at 458. At 2.25 and 5.63 m/min, (b) right-angled cornering contour at the same speeds as before and (c) circular contour of radius 30 mm at 7.2 m/min. The results obtained indicated that cross-coupled controllers using an optimal control formulation were practical and effective on experimental hardware representative of machine tool feed drives. In comparison with an uncoupled controller, the optimal controller was found to very effectively reduce contour error. The advantage of the cross-coupled controller was that it could reduce contour error without affecting the tracking error signicantly. The uncoupled controller on the other hand could reduce contour error as well as tracking error but in the process requiring higher gains than the crosscoupled controller for the same contour error level. Lo and Hsiao [42] proposed a compensation method for contour error in repeated machining process. In this method, the prole of the rst machined part was measured by a coordinate measuring machine (CMM). Based on the measured data, the tool path was modied by a compensation algorithm and then represented by a series of linear segments. The compensated tool path was then fed to the CNC machine tool for the machining of subsequent parts. Experiments were carried out on a three-axis CNC milling machine using a square aluminium block to machine a square prole. Although rst machining contour errors were very large, the compensation algorithm was found to effectively compensate the errors subsequently. Hong et al. [43] proposed an efcient method for the identication of motion error sources from circular test results in NC machines. Error signals from the circular tests were classied into two fundamental patterns namely nondirectional and directional error patterns. These patterns were readily available from two circular tests carried out under identical conditions except for the direction of motion. A simple algebraic average of the two test signals produced the non-directional error and then subtraction of the non-directional error pattern from each signal yielded the directional error. Circular tests were performed at an interpolation speed of 550 mm/min for a command circle radius of 95 mm. This method proved to be very efcient as it ultimately generated simple, linear equations for identication. Law and Geddam [44] presented a simple tool deection model for predicting contour errors in straight segments and corners for rough milling with slot cutting as well as nish

316

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 23. Block diagram of controlled feed drive including optimal coupled control action [41].

milling with immersion cutting. To compensate the tool path for such end milling of pockets, instantaneous tool deection was predicted by estimating the instantaneous cutting force by using the instantaneous radial depth of cut. The predicted results of the contour accuracy of pocket milling were veried using the CMM. The values of the predicted results closely matched those that were actually measured. Yang and Yellowley [45] took an approach based upon the use of a very simple high-speed control architecture which allowed real-time modication of path velocity at each loop closing. The work was intended to allow more exible path control in the presence of signicant process constraints and to examine methods that would allow further gains to be made in contouring speed. A fairly comprehensive set of tests were carried out to verify the approach. The experimental system comprised an open architecture controller with the slave axis code being run on two NEC V53 (80286 equivalent) based cards each congured as a temporary master. The system ran the master software (rst stage interpolation, etc.) on an Intel 80486 congured as permanent master. The experiments were conducted using the control in conventional fashion and having the slave processor write the actual position at loop closing to memory so that it could be retrieved and plotted versus the reference later. The motors used were small brushed DC servos with PWM type ampliers. The corner tracking performance of the system was tested with both single and multiple state line congurations. The experiments showed that the controlled servo with multiple state lines generated much less path error. Circular interpolation was also

conducted with and without multiple state lines. The circle tracking experiments showed that the controlled servo with multiple state lines again generated a much more accurate contour. Barakat et al. [46] proposed and tested a methodology for modelling and compensation of errors in multi-axis machines. An intrinsic CMM was chosen to test and validate the proposed methodology because the CMM would encounter errors in both measurements for quality and contouring for scan probing. The kinematics physical model of the CMM was enhanced by upgrading it to a hybrid of both physical and analytical modelling. Systematic modelling of the machine was carried out in two stages. The rst stage was based on physical modelling and consisted of using the HTMs to model the machine kinematics in a reference thermal state and arrive at an error model. This model was used in conjunction with the proposed compensation strategy (shown in Fig. 24) to correct the machine measurement errors. The proposed strategy was tested using the ASME standard CMM performance evaluation test and resulted in 93% measurement error reduction. In the second stage, the ambient thermal changes were incorporated in the kinematics models through the use of basic theoretical models as a start. These error function models were then veried and calibrated according to the actual laser measurements at varying thermal states. The nal models were tested through a modied version of the standard performance evaluation of the CMM, in conjunction with the proposed compensation strategy. The standard ASME performance evaluation of the CMM was repeated while having random

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

317

Fig. 24. Compensation strategy for CMM measurement error correction [46].

variations in the ambient thermal conditions. The performance of the CMM deteriorated by the introduction of the thermal disturbances and instability. However, the application of the proposed compensation strategy improved the performance from its worst case by 97%. These results depicted the correctness of the proposed models to estimate the errors and the effectiveness of the proposed strategy to make the model adaptive to thermal changes. Based on the error model developed, the inverse kinematics modelling was carried out for both stages using non-linear regression analysis. This compensation strategy proposed the incorporation of modications in the feed forward path of the commanded positions to the machine motion commands. This strategy provided a viable alternative to attempting to modify the control loop of the machine which introduced different problems ranging from stability issues to controller accessibility. The proposed methodology including the models and the adaptive compensation strategy was tested by simulating the machine in contouring different trajectories. These trajectories were selected to be the diagonal straight lines in the CMM work volume because they provided coverage of a wide range in the work volume and extensive interaction between the machine elements. The adaptive compensation strategy for contouring was applied while contouring the selected trajectories in a range of temperatures close to shop oor environment. Results obtained showed a signicant error reduction to values below one micrometer. Ho et al. [5] proposed a new path following controller with decoupled tangential and the normal control. The controller decouples the tracking system into normal

contour error dynamics and tangential speed regulation dynamics. The normal control is responsible for maintaining the tool on the target trajectory, and the tangential control moves the tool along the trajectory at the desired speed. Since the errors must be calculated on-line, an on-line interpolator was introduced in the controller. This approach made the new controller more robust to system delays because the on-line command generator was not affected by the delay caused by different material hardness or path obstacles. The new decoupled controller can be regarded as a modication to the cross-coupled controller. This approach maintains the advantages of the cross-coupled controller, while clearly identifying the control effort to drive the tool along the desired trajectory. The new controller thus allows for a distinct trajectory regulation and feedrate specications. The experimental results demonstrated that the decoupled controller was capable of achieving a separate trajectory following feedrate requirements. In addition, the new controller also achieved nonovershoot control to avoid overcutting. Luo et al. [47] proposed a neural network approach (shown in Fig. 25) to simultaneously control the average resultant cutting force and the contour error in multidimensional milling. The control system consisted of a neural force controller, a neural dynamic lag error controller and a feedforward input to compensate for static friction in the feed drives. To simplify the procedure for on-line learning, the neural force controller used an appropriate identication model to specify the feedrate. The neural dynamic lag error controller, on the other hand, was based on non-parametric process identication. Experimental results verifying the proposed methodology were presented for machining two-dimensional circular slots and a threedimensional spherical surface. The proposed methodology was applicable to any contour. Lo and Hsiao [48] proposed a CNC interpolator for repeated contour machining as shown in Fig. 26. In addition to a conventional contour interpolation algorithm, the CNC interpolator also includes contour error calculation, data extraction and contour error interpolation algorithms so that the previous contour machining result can be introduced to improve the accuracy of the subsequent repeated machining. During the rst machining task, the contour error at every sampling instant is calculated and the data selection is conducted so that a representative set of data is extracted. When the same contour machining task is repeated, the contour error is interpolated based on the previous extracted data and is then added to the reference position commands as illustrated in Fig. 26. The contour error calculation and the data extraction are conducted on-line for the rst machining while the contour error and the contour interpolations are on-line for subsequent machining. Experiments were conducted on a biaxial positioning table consisting of two linear axes driven by a servo motor through a ballscrew and sliding guideway. The proposed compensation method was applied to linear, circular

318

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 25. Neural Network control scheme [47].

and spiral contours. It was seen that the compensation method could achieve high performance while requiring only a small amount of sampled data. Oldknow and Yellowley [49] proposed a novel machine tool control architecture and mechanism for real-time manipulation of the nominal reference trajectory in three orthogonal directions. 3D dynamic interpolation was one of the approaches for the optimisation of machining economics through the maximisation of feedrate while avoiding

process constraints. The approach allowed the manipulation of the nominal reference trajectory so that small orthogonal offsets from the desired machined surface due to effects such as tool deection and wear can be compensated. This is in addition to contour errors resulting from system dynamics. In contrast to off-line (pre-processing) compensation techniques, variations in real-time cutting conditions due feedrate overrides, adaptive feed modulation, material property variations, etc. did not adversely affect

Fig. 26. Interpolator for repeated contour machining [48].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

319

Fig. 27. High speed, High precision XY table [50].

the performance of the approach. The implementation of this system provided robust performance under non-linear conditions. Mei et al. [50] investigated the effect of stick-slip motion for a high speed, high precision XY table (shown in Fig. 27) using the ballscrew and roller type guides through experimental and numerical simulation. A two-component mixed friction model was involved in the mathematical model of the servo feed systems. The study showed that the error caused by the friction of the ballscrew and roller guides could be precisely predicted through carefully tuning such parameters as surface roughness, etc. in the friction model. The dynamics and some other characteristics generated by the stick-slip motion could be described to some extent. With the mathematical model of the servo feed system developed by Mei et al., the effects of dynamic parameters, including the stiffness of the worktable, the dampness of the system, etc. on the performance of the system could be conveniently determined. The results obtained demonstrated that the friction error could be predicted off-line with the mathematical model. Shih et al. [51] proposed a variation of the cross-coupled controller (CCC). Combined with multiple-loop cascaded control design method, the new CCC structure with compensation at its reference position command is as shown in Fig. 28. The new structure facilitated the CCC to directly compensate the reference position commands of both axes thereby serving to integrate into any kind of axial tracking controller. A simple proportional position-loop controller was used to demonstrate the feasibility of this kind of integration. The concept of contour error transfer function was adopted and the pole-placement method was used to design a PI-type CCC compensator. Simulation and experimental results of the study both showed that this dualaxis motion control system could achieve satisfactory contouring accuracy under different motion conditions such as large feedrate or small radius of curvature cases.

This new controller could be easily implemented on a majority of motion systems currently in use via reprogramming the reference position command subroutine only without any change in hardware of the motion systems. Tarng et al. [52] presented an intelligent cross-coupled contour controller design using an efcient robust genetic learning algorithm (shown in Fig. 29). An appropriate weighting factor W had to be selected in coupled loop to speed up the control response. In addition, the cross-coupled biaxial control system had more than one input and one output. The parameters of the control namely Vx and Vy, which are also called the average unit velocity components, vary with time. Hence, the contour control system is a multivariable, non-linear, and time-varying control system. The stability analysis of such a control system or the selection of controller parameters (gains) is difcult to perform. It was shown that genetic algorithms could automatically search the controller parameters based on

Fig. 28. Novel cross-coupled controller [51].

320

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 29. Intelligent biaxial cross-coupled control system [52].

a tness function. As a result, contour performance of CNC machine tools could substantially be improved based on the optimal controller parameters. To further reduce the contour error, a cross-coupled fuzzy feedrate contour controller (shown in Fig. 30) was also developed. It was shown that the cross-coupled fuzzy feedrate contour controller could achieve an on-line adjustment of feedrate to optimize contour performance. Modern CNC machines have two principal systems namely the cutting system and the feeding system. Workpiece quality in terms of surface nish or contour accuracy is the overall outcome of cutting and feeding systems. Studies in general have tended to approach CNC machining from the perspective of either cutting or feeding. Cheng and Chin [53], however, investigated the contour errors on the basis of a complete CNC system. A system model that included all groups of functions in CNC machining, namely trajectory planning, trajectory tracking, the cutting process, and the machine structure (spindle-tool structure and table structure) was established (as shown in Fig. 31). Empirical data obtained for a table-top CNC milling machine was used to build the empirical cutting force model, the spindle-tool structure model and the table structure model. Contour errors were examined using this complete system. The limitations of traditional studies were examined and discussed. According to the results, the approach that

considers purely cutting suffers from ignoring the curvaturerelated features of the workpiece prole and the existence of tracking errors. Contour errors are local phenomena, and different errors occur at different locations on trajectory. Overlooking prole (trajectory) features therefore makes the efforts of cutting controls inefcient. Besides, overlooking tracking errors is equivalent to assuming perfect feeding, hence leaving the variation of cutting forces and its consequences due to this cause unchecked. Cutting was also shown to worsen contour errors in a manner that might eliminate some of the advantages gained by tracking techniques. This nding may lead to a re-thinking and redeveloping of tracking strategies. Some feeding and cutting parameters, including curvature, feedrate, cutting depth and tracking control, were examined with reference to the proposed complete CNC machining system. It was found that contour errors increased with feedrate at a rate that escalated with curvature. The approach proposed in this work represented a new way of understanding CNC machining. Tarng and Cheng [54] carried out an investigation of the contouring error of CNC machine tools due to stick-slip friction. Simulation and experimental studies indicated that quadrant protrusion produced by stick-slip friction is a primary contouring error in a circular test. It was also demonstrated that an appropriate tuning of the velocity loop

Fig. 30. One input, one output fuzzy contour error control system [52].

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

321

Fig. 31. Structure of CNC machining [53].

integral gain could efciently suppress quadrant protrusions and torque disturbance on the circular prole. The Circular Test is a quick method of testing the accuracy of three-coordinate measuring machines and NC machine tools. The test is inuenced by the 21 geometric error components of a three-axis machine. The 21 error components of a three-axis machine consist of the six error components per axis (i.e. positioning, straightness (two), roll, pitch and yaw) and the three angles of nonperpendicularity between the axes. Computer simulations have shown that each of the 21 error components yields either an excessive (or short) axial movement, a nonperpendicular movement, an out-of-straightness or a combination of these three basic effects (as shown in Fig. 32). It can be seen that such a three-dimensional test is sensitive to all geometric error components simultaneously. The Circular Test shows directly the two- or threedimensional uncertainty of the machine. In contrast, calculation of the two- or three- dimensional uncertainty from measured geometric error components is generally not possible because of the inuence of temperature (time delay between measurements), insufcient accuracy of the test equipment and random error effects [55]. Chuang and Liu [56] proposed a cross-coupled, modelreference adaptive feedrate control strategy for improving the contouring performance of multi-axis CNC machine tools. The proposed model (as shown in Fig. 33) enclosed a feedback loop between the input commanded feedrate and the output contour error. The feedrate is adjusted adaptively so that the resultant contour errors are maintained within the pre-specied tolerance. Experiments showed that the adaptive control strategy could increase the nominal feedrates for well-matched axial dynamics thereby yielding better productivity than the conventional non-adaptive control method. In addition, for mismatched axial dynamics,

the adaptive control strategy could maintain the prespecied contour error tolerance by adjusting the feedrates. Huang and Yan [57] devised a one-step, ahead adaptive control with input non-linearity strategy to control the converted system of a traditional milling machine table with lead screw transmission. Experiments were carried out and it was observed that the controller could overcome the nonlinear effect due to coulomb friction and construct a stable closed loop system. The performance of the adaptive control system was found to depend on the moving speed. The accuracy increased with reference input speed. Chattopadhyay [58] provided a basis of assessing magnitudes of dynamic errors associated with commanded motion by combining nite element modelling of the sliding interface, compliance and structural elements (links) with the well known 4!4 matrix dynamic modelling techniques. The output motions and the impact forces at the joints were computed as shown in Fig. 34. The results were used to assess dynamic accuracy. Canudas et al. [59] discussed the possibilities of improving the performance of a servo by non-linear compensation of friction. Several different models of friction have been proposed in the past. Canudas used a friction model that included Coulomb friction and viscous friction. Different ways to compensate for friction as well as to estimate the coefcients of the friction model have been investigated in the past. The adaptive techniques have been found superior because the friction depends on the operating conditions. Adaptive friction compensation was applied to an experimental system. Tracking experiments were carried out with a constant gain regulator without friction compensation and a controller with adaptive friction compensation. It was seen that with a constant gain regulator, the tracking performance was found to deteriorate as the friction was increased. However, with a regulator

322

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 32. Inuence of the 21 geometric error components [55].

based on adaptive friction compensation, the improvements were quite noticeable. Based on the concept that by pre-ltering the command trajectory in an appropriate way the contour error could be effectively reduced, Chang et al. [60] applied a hybrid approach to pre-lter design. A MLP neural network was used to enhance the performance of an adaptive calibrating controller. Due to the presence of non-linearities such as friction and signal quantization, this method further reduced contouring error by taking advantage of the non-linear and learning nature of the neural network. Ang et al. [61] studied the performance of variations in the zero-order hold (ZOH) algorithm designed to

compensate for trajectory tracking errors introduced due to sampling and ZOH actuations in computer based implementations of model-based motion control of robotic-type mechanisms. These algorithms provided a modied zero-order hold output to actuate the robot instead of providing a zero-order hold actuation output based on the current state. The 1-ZOH computes a modied actuation force F based on a linear extrapolation of the new actuation input at one half sample step later from the previous actuation. This new actuation force is used to actuate the robot. Of the algorithms that were tried out, the performance of the modied 1-ZOH algorithm was found to be superior to the rest, providing

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

323

Fig. 33. Model reference adaptive control scheme [56].

over 500% improvement in trajectory tracking errors compared to the standard ZOH algorithm. Astrom and Wittenmark [62] considered the problem of controlling a system with constant but unknown parameters. The analysis was restricted to discrete time single-input single-output systems. An algorithm obtained by combining a least squares estimator with a minimum variance regulator computed from the estimated model was analysed. The main results were two theorems which characterize the closed loop system obtained under the assumption that the parameter estimates converge. The rst theorem stated that certain covariances of the output and certain crosscovariances of the control variable and the output would vanish under weak assumptions on the system to be controlled. In the second theorem it was assumed that the system to be controlled was a general linear stochastic nth order system. It was shown that if the parameter estimates converge, the control law obtained is in fact the minimum variance control law that could be computed if the parameters of the system were known. It was shown that the algorithm could be feasibly implemented on a small process computer. Richalet et al. [63] described a new method of digital process control, called the Model Predictive Heuristic Control (MHPC), based on three principles namely: (a) The multivariable plant was represented by its impulse responses which was used on line by the control computer for long range prediction; (b) The behaviour of the closedloop system was prescribed by means of reference trajectories initiated on the actual outputs; (c) The control variables were computed in a heuristic way with the same procedure used in identication, which appeared as a dual of

the control under this formulation. The control algorithm is as shown in Fig. 35. This method was continuously and successfully applied to a dozen large scale industrial processes for more than a years time. Its effectiveness was found to be due to the ease of its implementation (e.g. constraints on the control variables) and to its amazing robustness as concerns structural perturbations.

Fig. 34. Computational block diagram [58].

324

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326

Fig. 35. Control algorithm for the MHPC [63].

4. Conclusion Tracking and contour errors are two very important aspects that need to be effectively addressed in order to maintain and enhance the dynamic performance of CNC machine tools. Extensive research, as analysed above, has been conducted in these two elds with the aim of arriving at new and improved methods of minimising and eliminating these errors. Traditional tracking control algorithms are based on the feedback principle. Using a high feedback gain has severe limitations manifesting in the form of increased noise sensitivity, generation of undesirable oscillations, etc. Since the reference trajectory is general known in advance in most CNC systems, the future reference points are available to use in improving tracking accuracy. This led to active research in the area of feedforward controller design involving the introduction of a feedforward control. This was followed by several improvements and modications to the control algorithm and has resulted in the development of many powerful tracking control systems like the ZPETC. Although generating high tracking accuracy, feedforward control suffered from the disadvantage of producing high frequency components in the control signal as well as being very sensitive to parameter changes and disturbances.

The introduction of low pass lters to limit the high frequency content of the control signal and the development of other adaptive measures to address parameter variation and improve disturbance robustness were some of the proposed solutions. The development of such systems has served to minimise tracking error and improve performance of the machine. Improving the overall control performance of a multiaxis machine tool is not necessarily ensured only by improving the tracking performance of each individual axis. Contour error, which is a typical performance index for the evaluation of multi-axis servo control systems, also needs to be effectively minimised. Several different systems like the CCC were introduced in order to address the issue of contour errors in servo systems. Cross-coupling, however, requires accurate knowledge of the kinematic conguration of the machine tool since the different axis controllers have to be coupled amongst themselves. Variations to the cross-coupling control concept with different contour-error models and/or controller laws based on the concept of the CCC method have also been proposed. In addition, other control algorithms such as repetitive control, predictive control, adaptive control and optimal control have also been developed in order to address and improve contouring performance in machine tools. The development of such systems plays an important role in addressing the needs of modern manufacturing especially with the advent of high-speed machining techniques and fast, PC-based machine tool control systems that provide sufcient processing speed. Such efforts to minimise and eliminate tracking and contouring errors reect directly on machine performance and enable the production of high precision machined components.

References
[1] Y. Koren, Design of computer control for manufacturing systems, Journal of Engineering for Industry, Transactions of the ASME 101 (1979) 326332. [2] C.-C. Lo, A tool-path control scheme for ve-axis machine tools, International Journal of Machine Tools and Manufacture 42 (2002) 7988. [3] Y. Koren, Computer Control of Manufacturing Systems, McGrawHill Inc., New York, 1983. [4] K. Erkorkmaz, Y. Altintas, High speed CNC system design Part III: high speed tracking and contour control of feed drives, International Journal of Machine Tools and Manufacture 41 (2001) 16371658. [5] H.-C. Ho, J.-Y. Yen, S.-S. Lu, A decoupled path-following control algorithm based upon the decomposed trajectory error, International Journal of Machine Tools and Manufacture 39 (1999) 16191630. [6] A.N. Poo, J.G. Bollinger, G.W. Younkin, Dynamic errors in type 1 contouring systems, IEEE Transactions on Industry Applications IA-8 (4) (1972) 477484. [7] Y. Koren, C.C. Lo, Advanced controllers for feed drives, Annals of the CIRP 41 (2) (1992) 689698.

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326 [8] M. Tomizuka, Zero phase error tracking algorithm for digital control, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 109 (1987) 6568. [9] M. Tomizuka, M.S. Chen, S. Renn, T.C. Tsao, Tool positioning for noncircular cutting with lathe, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 109 (1987) 176179. [10] T.-C. Tsao, M. Tomizuka, Adaptive zero phase error tracking algorithm for digital control, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 109 (1987) 349354. [11] D. Torfs, J. De Schutter, J. Swevers, Extended bandwidth zero phase error tracking control of nonminimal phase systems, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 114 (1992) 347351. [12] D.W. Clarke, Application of generalised predictive control to industrial processes, IEEE Control Systems Magazine 04/88 (1988) 4955. [13] D.W. Clarke, C. Mohtadi, P.S. Tuffs, Generalised predictive control Part I. The basic algorithm, Automatica 23 (2) (1987) 137148. [14] D.W. Clarke, C. Mohtadi, P.S. Tuffs, Generalised predictive controlPart II. Extensions and interpretations, Automatica 23 (2) (1987) 149160. [15] P. Boucher, D. Dumur, K.F. Rahmani, Generalised predictive cascade control (GPCC) for machine tool drives, Annals of the CIRP 39 (1) (1990) 357360. [16] M. Weck, G. Ye, Sharp corner tracking using the IKF control strategy, Annals of the CIRP 39 (1) (1990) 437441. [17] M. Tomizuka, T.-C. Tsao, K.-K. Chew, Analysis and synthesis of discrete-time repetitive controllers, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 111 (1989) 353358. [18] K.K. Chew, M. Tomizuka, Steady-state and stochastic performance of a modied discrete-time prototype repetitive controller, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 112 (1990) 3541. [19] B. Haack, M. Tomizuka, The effect of adding zeroes to feedforward controllers, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 113 (1991) 610. [20] H. Lim, J.-W. Seo, C.-H. Choi, Torsional displacement compensation in position control for machining centers, Control Engineering Practice 9 (2001) 7987. [21] B. Van Den, J.H. Swevers, H. Van Brussel, P. Vanherck, Accurate tracking control of linear synchronous motor machine tool axes, Mechatronics 6 (5) (1996) 507521. [22] D. Renton, M.A. Elbestawi, High speed servo control of multi-axis machine tools, International Journal of Machine Tools and Manufacture 40 (2000) 539559. [23] D. Torfs, J. Swevers, J. De Schutter, Quasi-perfect tracking control of non-minimal phase systems, Proceedings of the 30th IEEE Conference on Decision and Control 1 (1991) 241244. [24] M. Tomizuka, D. Dornfeld, M. Purcell, Application of microcomputers to automatic weld quality control, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 102 (1980) 6268. [25] M. Tomizuka, D. Dornfeld, X.-Q. Bian, H.-G. Cai, Experimental evaluation of the preview servo scheme for a two-axis positioning system, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 106 (1984) 15. [26] M. Tomizuka, D.E. Rosenthal, On the optimal digital state vector feedback controller with integral and preview actions, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 101 (1979) 172178. [27] M. Tomizuka, D.H. Fung, Design of digital feedforward/preview controllers for processes with predetermined feedback controllers, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 102 (1980) 218225.

325

[28] H.A. Pak, Adaptive matching and preview controllers for feed drive systems, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 113 (1991) 316320. [29] Y. Koren, Cross-coupled biaxial computer control for manufacturing systems, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 102 (1980) 265272. [30] R. Doraiswami, A. Gulliver, A control strategy for computer numerical control machine exhibiting precision and rapidity, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 106 (1984) 5662. [31] K. Srinivasan, P.K. Kulkarni, Cross-coupled control of biaxial feed drive servomechanisms, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 112 (1990) 225232. [32] K. Yamazaki, F. De Schepper, M. Kamiyama, Development of exible actuator controller for advanced machine tool and robot control, Annals of the CIRP 36 (1) (1987) 285288. [33] F. De Schepper, K. Yamazaki, M. Imai, J. Satou, Application of ASIC-technology to mechatronics control: development of the exible servo peripheral chip, Annals of the CIRP 37 (1) (1988) 389392. [34] F. De Schepper, K. Yamazaki, T. Shigemura, M. Imai, Development of an ASIC performing high speed loop processing of servo motion control for mechatronics applications, Annals of the CIRP 38 (1) (1989) 355358. [35] G. Pritschow, W. Philipp, Direct drives for high-dynamic machine tool axes, Annals of the CIRP 39 (1) (1990) 413416. [36] H.-Y. Chuang, C.-H. Liu, Cross-coupled adaptive feedrate control for multiaxis machine tools, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 113 (1991) 451457. [37] Y. Koren, C.C. Lo, Variable-gain cross-coupling controller for contouring, Annals of the CIRP 40 (1) (1991) 371374. [38] H.C.G. Pritschow, On the inuence of the velocity gain factor on the path deviation, Annals of the CIRP 45 (1) (1996) 367371. [39] H. Makino, T. Ohde, Motion control of the direct drive actuator, Annals of the CIRP 40 (1) (1991) 375378. [40] W. Knapp, S. Weikert, Testing the contouring performance in 6 degrees of freedom, Annals of the CIRP 48 (1) (1999) 433436. [41] P.K. Kulkarni, K. Srinivasan, Optimal contouring control of multiaxis feed drive servomechanisms, Journal of Dynamic Systems, Measurement and Control, Transactions of the ASME 111 (1989) 140148. [42] C.-C. Lo, C.-Y. Hsiao, A Method of tool path compensation for repeated machining process, International Journal of Machine Tools and Manufacture 38 (3) (1998) 205213. [43] S.-W. Hong, Y.-J. Shin, H.-S. Lee, An efcient method for identication of motion error sources from circular test results in NC machines, International Journal of Machine Tools and Manufacture 37 (3) (1997) 327340. [44] K.M.Y. Law, A. Geddam, Prediction of contour accuracy in the end milling of pockets, Journal of Materials Processing Technology 113 (2001) 399405. [45] L. Yang, I. Yellowley, High-speed contouring using a novel dynamic interpolation mechanism, International Journal of Machine Tools and Manufacture 41 (2001) 773794. [46] N.A. Barakat, A.D. Spence, M.A. Elbestawi, Adaptive compensation of quasi-static errors for an intrinsic machine, International Journal of Machine Tools and Manufacture 40 (2000) 22672291. [47] T. Luo, W. Lu, K. Krishnamurthy, B. McMillin, A neural network approach for force and contour error control in multi-dimensional end milling operations, International Journal of Machine Tools and Manufacture 38 (1998) 13431359. [48] C.-C. Lo, C.-Y. Hsiao, CNC machine tool interpolator with path compensation for repeated contour machining, Computer-Aided Design 30 (1) (1998) 5562. [49] K.D. Oldknow, I. Yellowley, Three-dimensional dynamic interpolation using stateline based control architectures, International Journal of Machine Tools and Manufacture 42 (2002) 16271641.

326

R. Ramesh et al. / International Journal of Machine Tools & Manufacture 45 (2005) 301326 [57] S.-J. Huang, M.-T. Yan, The adaptive control for a retrot traditional milling machine, IEEE Transactions on Industry Applications 32 (4) (1996) 802809. [58] S. Chattopadhyay, Dynamic errors in computer controlled machine tools, Proceedings of the Third International Conference on Computer Integrated Manufacturing, 1992 pp. 480489. [59] C. Canudas, K.J. Astrom, K. Braun, Adaptive friction compensation in DC motor drives, Proceedings of the IEEE International Conference on Robotics and Automation 3 (1986) 15561561. [60] B.C. Chang, R.C. Ko, S.K. Halgamuge, Pre-lter design for high speed contouring machines, International Conference on Neural Information Processing (ICONIP99), Perth, Australia, 1999 pp. 11001105. [61] M.H. Ang Jr, A.N. Poo, C.L. Teo, Q. Li, Compensation of trajectory tracking errors introduced by sampling in computer-control implementations of model-based robot control, Proceedings of the International Conference on Industrial Electronics, Control and Implementation IECON93, 3, 1993 pp. 16511653. [62] K.J. Astrom, B. Wittenmark, On self tuning regulators, Automatica 9 (2) (1973) 185199. [63] J. Richalet, A. Rault, J.L. Testud, J. Papon, Model predictive heuristic control: application to industrial processes, Automatica 14 (5) (1978) 413428.

[50] X. Mei, M. Tsutsumi, T. Yamazaki, N. Sun, Study of the friction error for a high-speed high precision table, International Journal of Machine Tools and Manufacture 41 (2001) 14051415. [51] Y.-T. Shih, C.-S. Chen, A.-C. Lee, A novel cross-coupling control design for bi-axis motion, International Journal of Machine Tools and Manufacture 42 (2002) 15391548. [52] Y.S. Tarng, H.Y. Chuang, W.T. Chu, Intelligent cross-coupled fuzzy feedrate controller design for CNC machine tools based on genetic algorithms, International Journal of Machine Tools and Manufacture 39 (1999) 16731692. [53] Y.-M. Cheng, J.-H. Chin, Machining contour errors as ensembles of cutting, feeding and machine structure effects, International Journal of Machine Tools and Manufacture 43 (2003) 10011014. [54] Y.S. Tarng, H.E. Cheng, An investigation of stick-slip friction on the contouring accuracy of CNC machine tools, International Journal of Machine Tools and Manufacture 35 (4) (1995) 565576. [55] W. Knapp, Circular test for three-coordinate measuring machines and machine tools, Precision Engineering 5 (3) (1983) 115124. [56] H.-Y. Chuang, C.-H. Liu, A model-referenced adaptive control strategy for improving contour accuracy of multiaxis machine tools, IEEE Transactions on Industry Applications 28 (1) (1992) 221227.

S-ar putea să vă placă și