Vous êtes sur la page 1sur 9

Finite Elements in Analysis and Design 5 (1989) 281-289

281

Elsevier SciencePublishers B.V., Amsterdam -- Printed in the Netherlands

D E S I G N O P T I M I Z A T I O N O F S T R U C T U R E S S U B J E C T T O STATIC AND DYNAMIC C O N S T R A I N T S Gopichand SOMAYAJULA


CAE Methods Section, Powertrain Planning and Engineerin~ Ford Motor Company, Dearborn, M1 48121, U.S.A.

and Jim B E R N A R D
Department of Mechanical Engineering~ Iowa State Unioersity, Ames, 1,4 50011, U.S.A.

Revised March 1989 Abstract. This paper presents a finite element based optimization technique in which the designer interactively chooses penalty functions which lead to the desired optimum. Sensitivitiescomputed using MSC/~ASTRANare the basis for approximate relationships between the response variables and the design changes. New software was developedto perform the optimizations.The techniqueis illustrated with two examples.

Introduction A typical design process involves iterative analysis which is computationally intensive and time consuming. As the complexity of the design increases, a purely intuitive design approach becomes inefficient in guiding the analysis. In these situations, a more formal approach called design optimization provides guidance for the systematic modifications to the model of the structure. The final result of this optimization process may be called an optimal design. In order to achieve an optimal design, it is necessary to have some function for which a minimum value is sought. This function is known as the objective function. The objective function includes quantities known as design variables which can be varied to achieve the desired optimum. The number and type of design variables will depend upon the problem. For example, these variables may be cross-sectional properties of a reinforcement, skin thicknesses of a fiber glass hood panel, etc. The goal of structural optimization is to design a structure that satisfies all the constraints and minimizes the specified objective function. Commonly used constraints include limits on the size of design variables, displacements and stresses under various static load conditions, and the frequency spectrum of the structure. This paper presents a finite element based design optimization technique. In the formulation considered here, the cost function becomes smaller when: (1) Natural frequencies of the system do not fall within undesirable ranges. (2) Maximum stress and deformations drop below the set limits. (3) Weight of the system decreases. (4) Changes from the original design remain within given bounds. (5) Design changes satisfy other combination constraints. Structural optimization has been widely discussed in the literature (see, for example, [1,2,4,6]). This work departs from most previous efforts in its reliance on simple intuitively based penalty functions which remain under the interactive control of the design engineer. This follows a path laid out in [3,5]. The contribution here is to apply these techniques to large 0168-874X/89/$3.50 1989, ElsevierSciencePublishers B.V.

282

G. Somayajula, J. Bernard / Design optimization c (structures

problems using MSC/ NASTRAN. The procedure followed here involves unconstrained optimization. As in [5], the cost functions are usually quadratic. For example, consider the cost P ( x ) , where

P(x)=a(x-x, P(x)=b(x-xo)
P ( x ) = 0,

)2,
2,

x>x,, x<xo,
x 0 < x <.v I.

(lal ilb)
(lc)

This type of function is handy for typical constraints on size, stresses and displacements--large values of a and b will ensure that the optimal design remains close to the range x 0 < x < .vj Often the desired attributes of the frequency spectrum of a structure do not concern where the designer wants the frequencies to lie. Rather, the designer would like to specify undesirable bands where frequencies do not lie to limit occurrence of resonance. This line of thinking leads to penalties which penalize frequencies in the range x 0 < ~0 < x 1. A useful form of the penalty is P(~0l=a[1-cos(2,~(~0-x P(~0) = 0, otherwise. 0)/(x,-x 0))], x 0<~0<x,, (2a) (2bl

It is obvious that this is not a hard constraint, i.e., there is a small penalty for a small violation of the x 0 to x I range. On the other hand, undesirable frequency bands are in fact usually fuzzy around the e d g e s - - t h e designer probably does not have in mind hard numbers for the band. Furthermore, the optimization process remains under interactive control of the designer, so if the undesirable range is invaded too far, the designer can increase either the weight a or the frequency range x 0 to .v~. There are similar penalties that are also useful in dealing with spectra of undesirable frequencies. For example, it is common to need to have all frequencies be above some lower limit. It is easy to see how we can modify (2) to meet this need. For the purposes of this paper, sensitivities of the response variables (displacement, stress, and frequency) with respect to design variables were computed within MSC/NASTRAN. These sensitivities were used in the optimization procedure to establish approximate relationships between design variables and response variables in order to determine the optimal solution. The sensitivities simplify the objective function, turning the optimization into an interactive exercise. Thus the designer can change the weight a n d / o r the shape of the various penalties and examine the effects on the optimal solution. On the other hand, the sensitivities are linear approximations to nonlinear functions, thus the range of validity of the approximation may be limited. Therefore, when the appropriate penalties are established and the optimization is complete, the finite element analysis needs to be redone to establish the validity of the approximation. In the event validity is not established, the optimization must be redone based on new local sensitivities [3]. The next section of this paper presents two illustrations of the technique. The first is a fairly straightforward example in which the design variables are the thicknesses of various areas of a rectangular plate and the objective is to remove natural frequencies from two separate ranges. The second is a far more complicated problem concerning the design of a vertical exhaust pipe mounting system for a heavy duty truck. This example includes constraints on size of changes. displacements under loading, frequencies, and stresses in key areas.

Examples
Example 1. A cantilever plate structure

The design of a cantilever plate structure is presented here to illustrate the proposed design technique. The finite element model was developed using MSC/NASTRAN. The model is shown

G. Somayajula, J. Bernard / Design optimization of structures

283

Fig. 1. Finiteelementmodelof the cantileverplate.


in Fig 1 and consists of 32 CQUAD4 plate elements and 200 displacement degrees of freedom The material was steel with an elastic modulus of 3.0 x 107 psi and weight density of 0.283 l b / i n 3. The plate was divided into four regions as shown in Fig. 2. The thicknesses of the four regions were the design variables which were allowed to vary between 0.05" and 0.50" Two undesirable frequency bands, 100-150 Hz and 800-850 Hz were imposed on the design. The thickness of the entire plate was initially set to 0.25". The design was analyzed for natural frequencies and the results are shown in Table 1. There were two modes within the undesirable frequency ranges: the first mode at 13081 Hz and the third mode at 827.89 Hz. The objective of our design was to decrease the weight of the plate and shift the natural frequencies out of the undesirable frequency bands. To do this, we used a penalty function of the form: P = a E (frequency penalties) + b E (size penalties) + c(welght) , and determined a, b and c interactively.
2

(3)

_~'~,,,~_~.j-- Region 1
~ Region2

3 Region

gion4

Fig. 2 Designvariablesfor the cantileverplate.

284

G. Somayajula, Z Bernard / Design optimization of structures

Table 1 Structural characteristics of the cantilever plate Initial design Design variables (in) Thickness of region Thickness of region Thickness of region Thickness of region Total weight (lbs) Natural frequencies (Hz) Mode 1 Mode 2 Mode 3 Mode 4 Mode 5 Mode 6 1 2 3 4 0.2500 0.2500 0.2500 0.2500 2.264 130.81 554.79 827.89 1769.88 1831.31 2401.76 Design iteration 1 0.2807 0.0504 0.0581 0.0500 0.994 47.87 170.72 319.47 573.16 806.64 926.71 Design iteration 2 0.2864 0.0500 0.0561 0.0500 1.001 47.25 166.88 313.34 567.39 794.63 919.69

~
upprTbe /

~ - - ExhaustPipe

Low.r,.~,--~/// ~

Framerail J

.......:'::-\-~'~:'::.~?-~;i ..~,, ,~\v

~reCtio n

Fig. 3. Exhaust pipe mounting arrange. ment.

G. Somayajula,J. Bernard / Design optimization of structures

285

The process took two iterations, i.e., after the first interactive optimization, the change in the design variables was too large to preserve the accuracy of the linear approximation. Thus the sensitivities were recomputed at the plate thicknesses indicated by design iteration 1, and the process was repeated. This second iteration procedure resulted in changes within the range of the linear sensitivities. Some of the details are given in Table 1.
Example 2. Design of an exhaust pipe mounting

The design of a vertical exhaust pipe mounting system for a heavy duty truck is considered in this example. Figures 3 and 4 show the isometric and top views of the mounting arrangements. The material is steel of specific weight 0.283 l b / i n 3 and modulus of elasticity E = 3.0 5< 107 psi. The exhaust pipe is mounted to the web of the framerall between the cab and the van by two rectangular tubes. The outboard ends of the tubes are welded to brackets which are fastened to the exhaust pipe. The inboard ends of the tubes are welded to a single bracket which is bolted to the framerail. A flat plate gusset is welded to the lower tube's bottom surface to reinforce the tube in the longitudinal direction. (Refer to Fig. 3 for the nomenclature.) The finite element model of the mounting structure was developed using MSC/NASTRAN.The supporting tubes were modelled with 8 CBAR elements each. The framerail bracket was modelled using 72 CQUAD4elements. The gusset was modelled using two CONM2 elements. The exhaust pipe was modelled using 20 CBEAMelements. The finite element model, which is shown in Fig. 5, has 816 displacement degrees of freedom. There are seven design variables in this example: (1) upper tube width 14,'1; (2) upper tube height D1;

i/-

Framerail T

Van

Cab

)
Exhaust Pipe

Fig. 4. Top view of exhaust pipe mounting bracket.

286

G. Somayajula, J. Bernard / Design optimization of structurex

Fig, 5. Finite element model of the exhaust pipe mounting.

(3) (4) (5) (6) (7)

u p p e r t u b e t h i c k n e s s tl; l o w e r t u b e w i d t h W2; l o w e r t u b e h e i g h t D2; l o w e r t u b e t h i c k n e s s t2; f r a m e r a i l b r a c k e t t h i c k n e s s T. T h e initial v a l u e s of t h e d e s i g n v a r i a b l e s a n d s o m e o f t h e c o m p u t e d r e s u l t s are s u m m a r i z e d in T a b l e 2. M o d e s h a p e s are s h o w n in Figs. 6 a n d 7.

Table 2 Structural characteristics of the initial design Design variables (in) Upper tube Width W1 Height D~ Thickness t~ Lower tube Width W2 Height D 2 Thickness t 2 Framerail bracket Plate thickness T Total weight (lbs) Natural frequencies (Hz) Mode 1 (longitudinal) Mode 2 (lateral) Static displacement due to 6g longitudinal gravity load (in) at Node 18 at Node 120 Maximum stress due to 10g laterial gravity load (psi) at Element 1

2.0 2.0 0.083 2.0 2.0 0.083 0.25 52.8 9.73 10.57 0.472 1.lll 20, 530

G. Somayajula, J. Bernard / Design optimization of structures

287

/
ZL- x

Y
Z"L X

Z ~X

Y zLx

zJ-x

zdx

Fig. 6. First mode of exhaust pipe mounting (longitudinal mode, frequency = 9.73 Hz).

Fig. 7. Second mode of exhaust pipe mounting (lateral mode, frequency = 10.57 Hz).

A prototype of the initial design was made for testing. In over the road testing, the maximum measured accelerations at the upper exhaust pipe mounting bracket were 6g and 10g in longitudinal and lateral directions respectively. The lower tube of the prototype cracked during testing at its upper edge near the weld to the framerail bracket (dement 1 in the model). This failure was the result of lateral vibration. The failure of the prototype indicated that we needed to give attention to the area of the model designated by element 1. We decided to attack this in two ways, namely, to increase the fundamental longitudinal and lateral frequencies from their current values of about 10 Hz to at least 15 Hz. We felt this increase would help isolate the system from typical excitation frequencies. Secondly, we decided to try to limit the stress in element 1 under a 10g lateral gravity load to 70% of its original calculated value of 20,530 psi: STR a < 14,370 psi, (4)

where STR a is the maximum stress in element 1. There were also several displacement constraints to consider based on anticipated use. For example, the displacement at the top of the exhaust pipe (Node 120 in FE model) is limited by the 3.0" clearance between the exhaust pipe and the van body. The displacement of the upper exhaust mounting bracket (Node 18 in FE model) is limited by the 1.0" clearance between the upper tube and the cab. These constraint equations are DLN120 < 3.00", DLN]8 < 1.00", (5)

where DLNa20 and DLN]8 are the displacements at the top of the exhaust pipe and at the upper exhaust pipe mounting bracket respectively. Our design goal was to meet the constraint under a 6g longitudinal load. The supporting tubes lie in the narrow gap between the cab and the van body. Therefore, the width of the upper tube is limited by the size of the gap and the longitudinal displacement of the upper tube to avoid hitting the van.
W1 -~- DLN18 < 3.0tt.

(6)

The tubes are welded to the framerail bracket at an angle; therefore, the total projected height of the upper and lower tubes on the framerail bracket cannot exceed the height of the bracket: D1/cos 60 o + D2/cos 35 < 9.75". (7)

The angles 60 and 35 are the inclination angles of the upper and lower tubes with respect to the frame rail bracket and 9.75" is the available height of the frame rail bracket. Finally, common practice and our experience led to size constraints shown in Table 3.

288 Table 3 Size constraints Design variable Upper tube width W1 Upper tube height D1 Upper tube thickness t 1 Lower tube width W 2 Lower tube height D2 Lower tube thickness t 2

G. Somayajula, J. Bernard / Design optimization of structures

Lower limit (in) 1.000 1.000 0.050 1.000 1.000 0.050 0.200

Upper limit (in) 3.000 3.500 0.250 3.000 3.500 0.250 0.500

Framerail bracket thickness T

Table 4 Structural characteristics of the optimized design Design variables (in) Upper tube Width W1 Height D 1 Thickness t 1 Lower tube Width W2 Height D2 Thickness t 2 Framerail bracket Plate thickness T Natural frequencies (Hz) Mode 1 (longitudinal) Mode 2 (lateral) Static displacement due to 6g longitudinal gravity (in) at Node 18 at Node 120 Maximum stress due to 10g laterial gravity load (psi) at Element 1

2.77 2.09 0.088 2.72 2.91 0.128 0.499 15.29 21.87 0.160 0.438 8, 720

A p e n a l t y al o n g the lines of e q u a t i o n s (1) a n d (2) was d e d u c e d i n t e r a c t i v e l y w h i ch i n c l u d e d all the desired features discussed here, a n d the m o d e l was o p t i m i z e d by i n t er act i v el y c h o o s i n g the weight of the p e n a l t y functions. T h a t p r o c e d u r e led to the results s u m m a r i z e d in T a b l e 4. T h e finite e l e m e n t c a l c u l a t i o n with the design variables set to the T a b l e 4 values verified that the linear sensitivities were a valid a p p r o x i m a t i o n an d T a b l e 4 i n d i c a t e d the desired m i n i m u m .

Conclusions

This p a p e r p r e s e n t e d a finite e l e m e n t b a s e d o p t i m i z a t i o n t e c h n i q u e in w h i ch the designer i n t eract i v el y chooses p e n a l t y f u n c ti o n s w h i c h lead to the desired o p t i m u m . T h e t e c h n i q u e was illustrated with t w o e x a m p l e s set up using MSC/NASTRAN.

G. Somayajula, J. Bernard / Design optimization of structures

289

Acknowledgments We appreciate the help of Navistar Corporation, especially Nick Mehta, Ed Wene, Ed Landis, and Craig Welch. References
[1] LEV, O.E. (ed.), Structural Optimization: Recent Developments and Applications, ASCE, New York, 1981. [2] LEVY,R. and O.E. Lv.v, "Recent developments in structural optimization", J. Struct. Eng. 113 (9) pp. 1939-1962, 1987. [3] RIZAI, M.N. and J.E. BERNARD,"An efficient method to predict design modifications", A S M E J. Mechanisms, Transmissions, Automation Des. 109, pp. 377-384, September 1987. [4] SCHMIT,L.A., "Structural synthesis--its genesis and development", AIAA 3". 19 (10), pp. 1249-1263, 1981. [5] STARKEY,J.M. and J.E. BERNARD,"A constraint function technique for improved structural dynamics," A S M E J. Acoust., Stress, Reliab. Des. 108, pp. 101-106, January 1981. [6] VANDEI~a'LAATS, G.N., "Structural optimization--past, present, and future", AIAA J. 20 (7) pp. 992-1000, 1982.

Vous aimerez peut-être aussi