Accelerated Gradient-Based Design Optimization via Differentiable Physics Informed Neural Operator for Composite Materials Processing

Composite materials have become indispensable in aerospace, automotive, and marine industries due to their exceptional mechanical properties. Composites are typically manufactured via autoclave curing processes, that demand precise control over temperature and pressure profiles. Optimizing the cure cycle as well as equipment design parameters is crucial for attaining the desired properties in the manufactured part. Traditional optimization methods require substantial computational time and effort due to the reliance on resource-intensive simulations like finite element analysis and the complexity of rigorous optimization algorithms. Data-agnostic AI-based surrogate models such as Physics-Informed Neural Operators (PINOs) offer a promising alternative for these conventional simulations with drastically reduced inference time, unparalleled data efficiency, and zero-shot super-resolution capability. Furthermore, their differentiable nature enables integrated gradient-based optimization within the simulation framework. This work presents an end-to-end accelerated AI-driven optimization framework for the manufacture of advanced composite materials. In particular, a novel Physics-Informed DeepONet (PIDON) architecture is proposed to accurately model the nonlinear behavior of composites' thermochemical evolution during the curing process for a high-dimensional design space, surpassing the performance of SOTA models. Leveraging PIDON's differentiability, we then employ a gradient-based optimization using Adam optimizer, achieving a 3x speedup in obtaining optimal design variables compared to gradient-free counterparts. The proposed framework delivers a scalable and efficient solution for optimizing curing processes and holds potential for broader applications in materials design.