The body of a passenger car roughly constitutes 25–30% of its overall weight. Any reduction in the weight of the car’s body would not only mean less materials and fuel to be consumed, but also less exhaust emissions to be released and less non-biodegradable materials to be dumped or recycled. However, the automotive industry’s desire for an increasing weight reduction of passenger cars is inevitably limited by other design considerations such as mechanical strength, overall stiffness of the body, durability, safety and corrosion resistance. The problem of weight minimization can be expressed in the form of a constrained, multi-objective optimization problem in which the weight of the body and its fatigue life constitute the conflicting cost functions and values of such critical performance parameters as body’s natural frequency forms the constraint set. The above optimization problem poses a challenge to the designer, as the weight, fatigue life and natural frequency of the geometrically complex body cannot be readily evaluated and a comprehensive numerical model, such as a Finite-Elements (FE) one, has to be employed. This numerical model would nonetheless be highly time-consuming, especially considering the need for re-assessing the model dozens, and sometimes hundreds, of times per iteration of the optimization algorithm. To avoid this, we use a neural approximation of the FE model to reduce the time and computational cost. Results of a finite number of FE simulations are used to train the Multi-Layer Perceptron (MLP) neural network which will then be used as the evaluation engine of the optimization algorithm. An efficient computer code based on the improved Non-dominated Sorting Genetic Algorithms (NSGA II) is used to find the Pareto set of distinct solutions. The designer would then be able to choose from a set of non-dominated, feasible solutions based on economical and/or logistics requirements at an early stage of the design process.

This content is only available via PDF.
You do not currently have access to this content.