A constitutive framework for non-isothermal plasticity through micro-mechanism informed artificial neural networks

Physics-informed neural network framework for constitutive modeling of aluminum alloys during non-isothermal hot forming. Combines metallurgical evolution equations (Kocks-Mecking dislocation dynamics, Hall-Petch grain refinement, JMAK precipitation kinetics) with deep learning through a dual-pathway architecture. Enforces thermodynamic consistency via penalty terms. Requires experimental data (500+ points) covering target process conditions. PyTorch implementation with MIT license.