Multi-objective topology optimization through GA-based hybridization of partial solutions

Alessandro Cardillo, Gaetano Cascini, Francesco Saverio Frillici, Federico Rotini

The embodiment design of a technical system can be considered as a multi-objective problem that the designer solves by finding geometrical solutions able to satisfy different conflicting requirements. The term ‘‘technical system’’ is intended as a set of independent elements linked together by relations, characterized by a set of properties, which allows to perform a certain function. In commercial software applications for structural topology optimization, the user has to build a model of the technical system to simulate its physical behavior (i.e., FEM with the material properties, loads, etc.) and to define the optimization problem. This activity requires the definition of the objective function embedding one or more physical properties of the technical system to be improved such as: global stiffness, natural frequency, etc., the constraints and the non-design areas that must be preserved during optimization.

Set of Pareto points obtained through the proposed method

In a recent project the authors have developed an approach to assist the identification of the optimal topology of a technical system, capable of overcoming geometrical contradictions that arise from conflicting design requirements. The method is based on the hybridization of partial solutions obtained from mono-objective topology optimization tasks. In order to investigate efficiency, effectiveness and potentialities of the developed hybridization algorithm, a comparison among the proposed approach and traditional topology optimization techniques such as Genetic Algorithms (GAs) and gradient-based methods is presented here. The benchmark has been performed by applying the hybridization algorithm to several case studies of multi-objective optimization problems available in literature. The obtained results demonstrate that the proposed approach is definitely less expensive in terms of computational requirements, than the conventional application of GAs to topology optimization tasks, still keeping the same effectiveness in terms of searching the global optimum solution. Moreover, the comparison among the hybridized solutions and the solutions obtained through GAs and gradient-based optimization methods, shows that the proposed algorithm often leads to very different topologies having better performances