Neural network-assisted design optimization with adaptive sampling for tow-steered composite structures

Abstract

Tow-steered composites offer significant potential for enhancing weight reduction and performance in aerospace structures. However, optimizing realistic tow-steered composite designs using finite element (FE)-based methods is often computationally prohibitive due to the expansive design space. Neural network (NN) models have emerged as a cost-effective alternative to FE-based optimization approaches. However, advanced NN models typically require substantial training data to achieve high accuracy, and the generation of this data through FE analysis of tow-steered composite structures remains computationally intensive. To address this challenge, this study introduces an adaptive sampling method that effectively reduces the required training data while enhancing the accuracy of NN-based design optimization. The proposed method is demonstrated on two tow-steered composite structures with different numbers of design variables, showcasing its ability to achieve improved optimization accuracy and reduced costs. The proposed method can be applied to other NN-based optimization problems, mitigating computational demands associated with generating training data.