Multi-objective optimization across multiple concepts

Abstract

Evolutionary multi-objective optimization (EMO) is often used to deal with practical problems in engineering design to identify products that exhibit the best possible compromise between multiple conflicting performance criteria. Much of the literature in EMO considers algorithmic developments and benchmarking problems involving a single concept only. However, in practice, there could be many applications where the solution of a given problem may be represented using multiple concepts, and optimizing each one of them individually to obtain the overall Pareto front may be inefficient. To address this gap, in this study, we firstly develop computer-aided models of multiple concepts for a simulation-based problem which can serve as a good application of multi-concept optimization. The problem involves the design of lattice structures with different types of unit cells (each constituting a concept), which can be customized to suit a range of real-world applications such as design of structures with minimal thermal expansion, low weight and high rigidity etc. Furthermore, we develop baseline evolutionary strategies to search across multiple concepts simultaneously. Empirical experiments show that the presented strategies are able to outperform conventional single concept-based approach. Moreover, some challenges are also uncovered which prompts the need for further developments.