Resource-Aware Multi-Fidelity Multi-Objective Multidisciplinary Design Optimization

Abstract Modern simulation-based design optimization problems have evolved from being single-objective, single-discipline, and single-fidelity to being multi-objective, multi-disciplinary, and multi-fidelity. This evolution has been mainly driven by the elaborate design requirements and the availability of multi-disciplinary, multi-fidelity simulation models for complex products in industrial environments. In this paper, we formulate a more practical design optimization problem that considers the computation resources on top of the existing multi-objective, multi-disciplinary, and multi-fidelity problem. In the context of multi-disciplinary and multi-fidelity design optimization, the evaluation costs of simulation models can vary significantly across different disciplines and fidelities. Most existing algorithms for solving these problems only assume sequential or simply parallelized evaluation of the simulation models, which may be inefficient due to the mosaic cost structures of different simulation models. In our problem formulation, we incorporate the model evaluation environment and grant the optimization algorithm flexible control over the simulation model evaluations, so that it can potentially achieve higher efficiency than in traditional setups. Along with our new problem formulation, we propose an intuitive algorithm that solves the new problem and capitalizes on the flexible control over the model evaluations. We use a simulated case study to demonstrate the features of our problem formulation and algorithm.