Machine Learning and Metamodel-Based Design Optimization of Nonlinear Multimaterial Structures

Abstract

This study presents an efficient multimaterial design optimization algorithm that is suitable for nonlinear structures. The proposed algorithm consists of three steps: conceptual design generation, design characterization by machine learning, and metamodel-based multi-objective optimization. The conceptual design can be generated from extracting finite element analysis information or by using structure optimization. The conceptual design is then characterized by using machine learning techniques to dramatically reduce the dimension of the design space. Finally, metamodels are derived using Efficient Global Optimization (EGO) followed by multi-objective design optimization to find the optimal material distribution. The proposed methodology is demonstrated using examples from multiple physics and compared with traditional multimaterial topology optimization method.