DEEP LEARNING FRAMEWORK FOR WOVEN COMPOSITE DESIGN AND OPTIMIZATION

Abstract

Woven composites are produced by interlacing warp and weft fibers in a pattern or weave style. By changing the pattern or material, the mechanical properties of woven composites can be significantly changed. However, how a woven composite’s architecture (pattern, material) is related to its mechanical properties is still unclear. Thus, in this paper, we utilize Deep Neural Network to explore the relationship between 2D woven composite architectures and their corresponding mechanical properties. We propose Deep Convolutional Neural Network to predict in-plane modulus from a given weave pattern and material sequence. More importantly, we propose a Physics-Constrained Neural Network to predict woven composite architecture from in-plane modulus, which is an extremely complex problem. The results demonstrate that our proposed Deep Neural Network frameworks can effectively represent the relationships between woven composite architecture and corresponding in-plane modulus with much higher accuracy compared to existing frameworks. We anticipate our proposed frameworks will not only facilitate woven composite analysis and optimization process but also be a starting point to introduce Physics knowledge guided Neural Networks into woven composite analysis.