Design of High-Performance Composites via "Self-constructible Finite Element Material Library" Driven by Reinforcement-based Machine Learning

Lightweight fiber-reinforced polymer composites offer a promising alternative to metal-based engineering solutions. However, understanding and predicting their complex nonlinear mechanical behavior poses challenges due to intricate microstructures and experimental limitations. Developing constitutive models for accurate Finite Element (FE) simulations demands significant expertise and time investment. This research proposes the integration of Artificial Intelligence (AI) into constitutive material modeling. Firstly, we establish a comprehensive database that combines fundamental experiments with FE-based data, enabling the categorization of elementary nonlinear thermo-mechanical features. Secondly, we develop a Neural Network (NN)-based architecture to identify nonlinear features in stress-strain responses. Lastly, we construct a self-consistent AI-based framework to determine the appropriate combination of physics-based rheological analogs needed to replicate the observed mechanical response of the material under various loading scenarios. This innovative approach harnesses existing experimental and simulation data, employing advanced AI algorithms to overcome traditional modeling challenges associated with composite materials. By digging into the existing experimental and simulated data, new material models are autonomously constructed. Furthermore, the prediction of mechanical behavior for materials under diversity of loading and environmental conditions are accomplished through this advanced method. The resulting framework serves as a valuable tool for guiding composite design and facilitating their integration across diverse engineering fields.