Processing parameters optimization in hot forging of AISI 4340 steel using instability map and reinforcement learning

Abstract

High-temperature compression tests of the AISI 4340 alloy were conducted at temperatures of 900–1200 °C, strain rates of 0.01–10 s−1, and true strain ranges between 0.01 and 1.00 to explore the hot deformation properties. The experiment result has been used to obtain stress-strain curves of the material. The instability map is built at each true strain based on the dynamic materials model and Prasad's instability criterion. On the top of the instability map, the deformation behavior obtained from finite element analysis (FEA) is schematically investigated in the deformation flow instability. A reinforcement learning-based optimization algorithm has been developed to derive the optimal processing parameter, including workpiece temperature and stroke speed in the manufacturing process. Q-learning was used to learn and explore the solution environment to achieve the optimum processing parameters that prevent defects from instability traits. The experiment results show that the algorithm successfully avoided the unstable domains. Optical photography and electron backscatter diffraction (EBSD) has verified the proposed approach's practicality with experimental observations of flow instability signs.