Dsgn2Last

Dsgn2Last

GENSIZ optimization of a SUV chassis

Background

An SUV (Sports Utility Vehicle) chassis typically is a thin- walled structure, spot welded at multiple locations, providing for mounting of major subsystems such as the engine and fuel tanks, apart from seats and the body shell. To me, it seemed motivating to assess design efficiency of a typical SUV chassis.  I and my student got in touch with one of India's leading SUV makers and we soon took up work on a chassis in their product family. 

Objective

Seen here is the finite element model of the chassis and twenty property groups in distinct colors (partly!). The model included lumped masses representative of the engine, fuel tank and body shell, apart from other subsystems. The aim was to assess how far optimization could improve performance and reduce mass of the chassis. To this end, a simple objective was formulated to represent the ratio of norm of lowest modal eigenvalues, to the structural mass. GENSIZ did an unconstrained maximization, with search over a vector of allowable wall thicknesses. 

Convergence of search

On the right is a plot indicative of the population average objective's convergence with generations.  The objective, ratio of the fundamental modal frequency to structural mass, was of course maximized as all genetic algorithms do. The plot however shows the reciprocal of this ratio, namely weight- to- frequency.

Cloud plot

Seen here is a so- called cloud plot of all unique designs archived by GENSIZ, from start to convergence. The pair of lines just around 29 Hz and 117 kg respectively pertain to fundamental frequency and mass of the baseline design, then employed by the manufacturer. Clearly, among the thousands of designs encountered by GENSIZ, there exist quite a few hundreds of designs with higher frequencies AND structural masses lower than the baseline's (top left quadrant).  

Wall thicknesses comparison

Here is a plot comparing wall thicknesses  of the baseline and one of the optimum solutions. We could see that GENSIZ has re- distributed structural mass, moving out excessive mass (segment 1 for instance) to segments where it was deficient (like segments 14 and 15), in order to achieve design efficiency. One should note that the fundamental mode alone wouldn't suffice as a driver of the chassis design, nor would modal character alone do so either. Indeed, though not presented here, runs were made for distortion under cornering loads as well for norm of the lowest three modes. Typically chassis modes are of heave, pitch and roll in the 'rigid'- group while in the flexible group, the first- bending, first- shear and first- twist modes, which were all encountered, not pictured here.

Summary

The study convincingly shows an immense potential for design efficiency improvement using GENSIZ. As is typical with genetic algorithms, numerous multiple optima are produced by the search, offering as many options for exploring design needs not considered here, etc.

B. V. Vijay
X