Documente Academic
Documente Profesional
Documente Cultură
00
Printed in Great Britain. C 1991 Pergamon Press pk.
Abstract-This paper presents a method for considering fatigue life requirements in the optimal
design of structures. The basic concept is to use the load history data combined with the finite
element stresses of the structure and the material fatigue properties to calculate the fatigue life
during the optimization process. The life requirement is considered as side constraints and the
structure weight as the objective function. To demonstrate the concept, the optimization task with
fatigue life constraints and the fatigue life calculation, based on the contemporary approach, are
discussed. Some optimum design test cases, using the SAE keyhole specimen experimental fatigue
data, are presented.
INTRODUCTION
UNTIL the last two decades the only fatigue property of metals determined by testing and reported
in handbooks and literatures was the so-called endurance limit or fatigue limit. This was the
stress amplitude in a completely reversed test. Below this stress amplitude the fatigue life was
considered to be infinite. The pioneering work of Coffin and Manson in 1954, where the cyclic
plastic strain and fatigue life were shown to be related by a simple power function, led to a more
detailed formulation of fatigue properties with metals over the entire life range from only a half
cycle to millions of cycles. This formulation of fatigue properties led to the contemporary approach
for calculating fatigue life in terms of plastic strain for variable amplitude load histories.
In the design of many structures, the fatigue life requirement is one of the major design criteria
for the safety of the structure. Early research efforts to address the safety aspect in the minimum
weight design of structures were based on the structural reliability or failure probability[l-31. The
complicating factors in these methods are the statistical dependence and the need for prior test data
for the given types of structures. In this paper we present a method for including the fatigue life
requirement, in the optimization of structures, without statistical dependence or prior test data
requirement. The method utilizes the finite element stresses, the entire load history of the required
life span and the material properties directly in the optimization process. We first discuss the
optimization task with fatigue life constraints, then we discuss the basic requirements for a fatigue
life routine based on the contemporary approach for fatigue life calculation. To test the developed
computer program some test cases, using the keyhole specimen experimental data given by Socie
and Morrow[4], are presented.
where n is the number of variables contained in X and m is the number of inequality constraints.
The objective function to be minimized is taken as the total weight of the structure. The side
constraints of eq. (3) are simple limits imposed on the design variables to provide practical limits
on member sizes.
The inequality constraints of eq. (2) are the constraints derived from performance require-
ments. These inequality constraints include limits on element stresses and vibration frequencies of
the structure. Thus the following constraints are specified.
Stress
Lower and upper bounds may be imposed on any stress components that can be computed
using a finite element analysis program. The typical stress constraint has the form
aij/6 - 1 < 0 (4)
i = element number
j = loading condition
Z = allowed stress.
Displacement
Limits may be imposed on displacement components at any grid point or on the maximum
resultant displacement at any grid point, and this constraint is typically
s,js- 1 GO (5)
i = grid number
j = loading condition
k = direction
C?= displacement limit.
Frequency
Any fundamental vibration frequencies of the structure may be constrained by
1 -nJn; <o @a)
n,/nr - 1 < 0 (6b)
where ni and ny are lower and upper bounds, respectively, on the ith eigenvalue.
The stress constraints must be satisfied for all loading conditions. While the magnitudes of
the fatigue loads are small compared to the static load, the effect on the structure is due to their
entire history. If the fatigue loads are used as static loading they will hardly have any effect in sizing
the structure. It is the change in magnitude and direction of these loads that causes the fatigue
damage of the structure. Therefore, the entire load history of the required life span should be used
in the optimum design of the structures. This can be accomplished by imposing a lower bound on
the fatigue life, at any elment in the finite element model of the structure, during the optimization
process. This constraint takes the form
1 -L&E<0 (7)
i = element number
j = loading condition
L = allowed fatigue life.
The evaluation of the fatigue life constraints requires the calculation of the fatigue life L,, at
every step of the optimization. This requires a fatigue life calculation routine to be connected to
the finite element and optimization routines. From the finite element stresses, the load histories and
the material fatigue properties data the fatigue life routine calculates L, during the optimization
Structural optimization with fatigue life constraints 1151
process. In the following section we discuss the basic requirements for such fatigue life calculation
routine.
The A, in the above equation, indicates completely reversed ranges of stresses and strains and
subscripts e and p stand for elastic and plastic strain. The two material properties, n and K, are
the cyclic strain hardening exponent and cyclic strength coefficient, respectively.
For the calculation of fatigue life, it is convenient to incorporate mean stress effects as an
equivalent change in static strength. Equation (10) may then be modified as follows
Aa
2 = (a; - a0)(2N,)b. (11)
To obtain an expression relating total strain, mean stress and life, eq. (11) is divided by
the Youngs modulus, E, to obtain the elastic strain and added to eq. (9) for the plastic strain to
yield
The two exponents and the coefficients are regarded as fatigue properties of the metal, and
they are designated as follows
By manipulating eqs (9) and (10) to eliminate life and changing the result to the form of
eq. (8), it can be shown that the cyclic strain hardening exponent, n, is determined by the fatigue
strength and ductility exponents as follows
(13)
Similarly, it can also be shown that the cyclic strength coefficient can be determined from the
fatigue properties as follows
(14)
Cycle counting
Some of the cycle counting methods in use today for fatigue analysis are peak, level crossing,
range, range-mean, range-pair, and rainflow. Of these various methods, rainflow or its equivalent
range-pair has been shown to yield superior fatigue life estimates[8]. The basic idea behind rainflow
counting is to treat small events in the load history as interruptions over larger overall events and,
in the simplest terms, to match the highest peak and deepest valley, then the next largest and
smallest together, etc., until the peaks and valleys of the load history have been paired.
Notch analysis
In dealing with real components, it is often necessary to relate the nominal loads or strains
to the maximum stresses and strains at the critical location. Neuber derived a rule which applies
when the material at the notch root deforms nonlinearly. The theoretical stress concentration, K,,
is equal to the geometric mean of the actual stress and strain concentration factors, K,, and K,
K, = (K, Kc). (15)
Topper[lO] modified Neubers rule for use in cyclic loading applications by substituting the
fatigue notch factor, K,, for the stress concentration factor and rewriting eq. (15) in the following
form
K t=& AC (16)
Structural optimization with fatigue life constraints 1153
Combining this form with eq. (81, an expression for the notch stress becomes:
(K, A,!Q2
Z------_ (191
E
This equation is easily solved using the Ne~on-Raphson iteration technique. If finite element
analysis program is used to calculate the stresses, Kf will depend on the mesh size at the high stress
area. For a fine finite element mesh Kf = 1. Once the notch stress is obtained, it can be used in eq.
(8) to solve for the elastic and plastic strains at the notch root. After each reversal, a new axis is
defined and the right-hand side of eq, (19) recalculated in order to solve for the new stress and
strain range.
~urnulat~~~edamage analysis
Cumulative damage fatigue analysis is usually based on the Palmgren-Miner linear damage
rule. Fatigue damage is computed by linearly summing cycle ratios for the applied loading history,
as indicated in the following equation
Damage = c pfi
fi
ni = observed cycles at amplitude, i
Nfi = fatigue life at constant amplitude, i.
After the fatigue damage for a representative segment or block of load history has been
determined, the fatigue life in blocks, L, of eq. (7), is calculated by taking the reciprocal.
The fatigue life for any cycle or reversal can be determined from eq. (12). The mean stress,
do and cyclic strain range, AC, has been determined from the material response model. Equation
(12) cannot be explicitly solved for life because of the negative fractional exponents involved, but
can easily be solved using Newton-Raphson or interval halving iteration techniques.
ALGORITHM
The optimization procedure discussed was developed into a computer program. As shown in
the overall flow diagram of Fig. 1, the finite element routine calculates the displacements, stresses
and frequencies of the structure. For the fatigue life calculation, the maximum von Mises stress
in the structure is then sent to the fatigue analysis routine and the fatigue life is calculated following
the approach discussed earlier. The displacement, stresses, frequencies and fatigue life are then sent
1154 M. E. M. EL-SAYED and E. H. LUND
and increment
derlgn variables
Fig. 1.
to the optimization routine to evaluate the objective function and constraints. The new search
direction is determined using the finite difference approximation. The optimization search
procedure is continued until the optimum is found.
TEST CASES
To study the effect of the fatigue life on the optimization process and test the developed
program, some test cases were performed using the keyhole specimen experimental test data given
by[4]. The minimum weight design of the SAE Keyhole specimen with fatigue constraints for a
specified maximum load was performed to determine the optimum thickness for required life equal
to the actual tested fatigue life. The tests were made using data for Man Ten steel with the bracket,
suspension and transmission load histories. The optimization problem solved for each test case was:
Pivot point of
lwding clrvis
\
------y-T
5.00
I.50
2.70 4 1.00 -
I=- I t
I t
0.75 -
318
* 1.50
where
p = density
A = area
X = thickness
Li = life calculated with current thickness
L, = allowable life.
An initial guess for the design variable X (thickness) was supplied to the optimization routine
as well as the bounds for the inequality constraints. For some load levels the upper and lower
experimental fatigue life bounds were used to obtain upper and lower bounds for the optimum
thickness. To perform the finite element analysis, the keyhole specimen is shown in Fig. 2, was
modeled using a fine mesh of plane stress elements.
The predicted optimum thickness results, for the different maximum loads for each load
history along with the number of blocks to failure, are presented in Tables l-3. As shown in the
tables, the predicted optimum thicknesses are reasonable when compared to the actual thickness,
considering the variation in the test data and the approximation in the fatigue life prediction.
CONCLUSION
A method for including the fatigue life requirement, in the optimization of structures is
developed. The method does not have statistical dependence or prior test data requirements, for
the types of structures being designed. The finite element stresses, the entire load history of the
required life span and the material properties are used to calculate the fatigue life during the
optimization steps. The fatigue life is obtained using the contemporary approach for calculating
fatigue life in terms of the plastic strains for variable amplitude load histories.
Reasonable predicted optimum thicknesses are obtained using the SAE Keyhole specimen
experimental data, for different maximum load levels and different load histories. Due to the
number of fatigue life evaluations and the time required for each evaluation, the optimization
process is slow. For complicated structures with many design variables and large load history data
further research is required to reduce the solution time.
REFERENCES
VI H. H. Hilton and M. Fe&en, Minimum weight analysis based on structural reliability. J. Aerospace Sciences 27,
641-652 (1960).
PI H. Switzky, Minimum weight design with structural reliability. AIAA 5th Annual Structures and Materials Cor$
pp. 316322 (1964).
[31 F. Moses and D. E. Kinser, Optimum structural design with failure probability constraints. AIAA 1. 5, 1152-I 158
(1967).
[41 D. F. Socie and J. Morrow, Review of contemporary approaches to fatigue damage analysis. FCP Report No. 24,
University of Illinois (1976).
[51 G. N. Vanderplaats, H. Miura and M. Chargin, Large scale structural synthesis. Finite Element Anal. Design 1, 117- 130
(1985).
I61 R. T. Haftka and M. P. Kamat, Elements of Srructurul Optimization. Martinus Nijhoff, The Hague (1985).
H. 0. Fuchs and R. I. Stephens, Meful Fatigue In Engineering. John Wiley, New York (1980).
i; N. E. Dowling, Fatigue failure predictions for complicated stress-strain histories. J. Mater. 7, 71-78 (1972).
L91R. M. Wetzel, A Method of Fatigue Damage Analysis. Ph. D. Thesis, University of Waterloo, Canada (1971).
WI T. H. Topper, R. M. Wetzel and J. Morrow, Nubers Rule Applied to Fatigue of Notched Specimens. J. Mufer. 4,
200-209 (1969).