Realistic Constrained Multiobjective Optimization Benchmark Problems From Design

Multiobjective optimization is increasingly used in engineering to design new systems and identify tradeoffs. Yet, problems often have objective functions constraints that are expensive highly nonlinear. Combinations of these features lead poor convergence diversity loss with common algorithms not been specifically designed for constrained optimization. Constrained benchmark exist, but they do necessarily represent the challenges problems. In this article, a framework electro-mechanical actuators, called multiobjective actuators (MODAct), presented 20 test derived from specific focus on constraints. The full source code made available ease its use. effects analyzed through their impact Pareto front as well performance. A constraint landscape analysis approach followed extended three metrics characterize search spaces. MODAct compared existing suites highlight differences. addition, using NSGA-II, NSGA-III, C-TAEA suggests indeed difficult due particular, number simultaneously violated newly generated solutions seems key understanding challenges. Thus, offers an efficient analyze handle future algorithm design.