Toward scalable benchmark problems for multi-objective multidisciplinary optimization
Victoria Johnson, João A. Duro, Visakan Kadirkamanathan, Robin C. Purshouse · 2022 IEEE Symposium Series on Computational Intelligence (SSCI) · 2022
Scalability in disciplines is an important consideration for multidisciplinary design optimization (MDO). Very few benchmark problems for multi-objective MDO exist in the literature, none of which are readily scalable. In this study, we introduce a new scalable benchmark problem that extends an existing well-known multi-objective benchmark problem. We show that scaling the number of disciplines in the problem, implying an increasing number of decision variables, does produce substantive changes in convergence ability. We also show that the accuracy of the multidisciplinary analysis (MDA) solver has an impact on the convergence ability of the multi-objective optimization algorithm, which is particularly noticeable when moving from 7 to 14 disciplines. Modification of standard (non-MDO) multi-objective benchmark problems is a promising approach to developing scalable multi-objective MDO benchmarks.