Developing an artificial neural network controller for accelerating the hot deformation of the titanium aluminide TNM-B1 using reinforcement learning and finite element simulations

Abstract This work presents a framework for interfacing reinforcement learning algorithm with finite element model in order to develop an artificial neural network controller. The goal of the controller is accelerating hot compression process titanium aluminide TNM-B1. interacts by exploring different die velocities and receiving input measurements (the velocity, displacement force die) while collecting rewards if constant stress state workpiece achieved. Synthetic stochastic material behavior was used simulate observed variations deformation same setup reward function able adapt two example environments; simple cylinder between flat dies more complex bone dies. performance environment comparatively reduced less consistent. In addition, training times instability were significantly increased. Furthermore, results suggest that can be as tool find optimizations or alternative routes. demonstrates concept provides groundwork fundamentals transferring method physical setup.