Enabling Material Design within System-Level Optimization via Machine Learning

The focus of the proposed work is on new tools, primarily enabled with machine learning, to establish a stronger link between composite material selection, and design of materials, and vehicle-level design. This will be especially beneficial where extensive material test data is not readily available, such as novel materials for applications with extreme environments in propulsion structures. The effort outlined in this proposal starts with the structural analyst/designer viewpoint and would use machine learning to develop tools to build a bridge to materials scientists/engineers. The HyperX software, a tool for performing structural analysis and optimization at the vehicle/system level, will be used as the foundation of this approach. Structural optimization at a vehicle/system level can require between 1,000 to 100,000 candidate evaluations per component, repeated over 100s or 1,000s of components in a structure. Therefore, it is not practical to run something such as a micromechanics simulation during the evaluation of each candidate. However, advancements made in machine learning in the last decade creates the opportunity to embed surrogates of these multiscale material models within vehicle-level optimization while using significantly less computational resources. Machine learning has been successfully applied within each of those two domains; the work outlined in this proposal would be the first use of machine learning in a commercial software to link these two domains. Machine learning would be used to develop surrogates of the HyperX analysis and optimization, with composite lamina properties as inputs to the models. This would enable rapid design exploration and optimization with tailored material systems by providing accurate component performance and vehicle/system masses for each material candidate considered.