Optimal design of selected features of exhaust system shields using different optimization methods and artificial neural networks

Abstract

The article presents problems related to methods of optimal design of heat shields used in exhaust systems of internal combustion engines. The optimization method proposed in this paper goes well beyond the scope of the standard design process. The paper uses a variety of local and global optimization algorithms, those both built into numerical simulation systems and in-house and external algorithms. An optimization criterion was defined and numerically implemented, together with constraints derived from the real requirements for this type of shielding. A simplified numerical model of finite element method providing the required accuracy adapted to the optimization task was developed. In addition, the work also presents a method for creating finite element surrogate models using artificial neural networks. The process of selecting the network topology and its learning allowed the development of a metamodel characterized by very good quality, for which, despite the relatively large number of design variables, the response errors are completely acceptable from a practical point of view. Numerical results were compared and developed for the used methods and algorithms.