Data-Driven Model-Free Intelligent Roll Gap Control of Bar and Wire Hot Rolling Process Using Reinforcement Learning

Abstract

Modern bar and wire manufacturing plants are constantly seeking to achieve lower costs, higher product quality, higher efficiency, and greater flexibility, which in turn require a significant increase in the degree of automation. The lack of accurate models and measurements of process essential parameters affects the realization of innovative control strategies. Data-driven approaches have received a lot of attention from control engineering researchers owing to their outstanding performance as function approximators. Supervised, unsupervised, and reinforcement learning approaches have been successfully employed in system identification problems and control systems design. Reinforcement learning holds advantages over the other approaches owing to its ability to learn without having the desired ground truth state. In this paper, a data-driven model-free reinforcement learning algorithms are developed for model parameters identification and roll gab control of a bar and wire hot rolling process. The reinforcement learning algorithms are based on the Deep Deterministic Policy Gradients algorithm with the actor-critic structure. The validation results of the developed solutions showed high performance and the agents were able to generalize to unseen scenarios.