Optimization of Hot Forging Using Reinforcement-learning-based Slide Motion Control

Abstract

Intelligent slide motion control in hot forging that combines artificial intelligence was investigated. Focusing on the header processing of reinforcement materials, we investigated the effects of material composition and forging conditions on the strength of the molded product from both experimental and computer-aided engineering (CAE) perspectives. Next, we verified the accuracy of the results obtained by CAE and confirmed that the hardness distribution and product strength were qualitatively consistent with those obtained in the experiments. Using the results obtained by CAE, we constructed a surrogate model with forging conditions as the input and product strength and forging load as the output. By performing reinforcement learning using this surrogate model, we produced an intelligent model capable of providing slide motions for improving product strength with the considerations of decreasing the time required for product formation and the forging load. When implementing this intelligent model on a hydraulic press and confirming the effectiveness of the intelligent slide motion control, we found that both a reduced product formation time and an improved product strength could be achieved under randomly set forging conditions.