An Integral Reinforcement Learning Approach for Adaptive Chatter Control in Milling Processes

Chatter vibration in milling is a self-excited phenomenon that deteriorates surface finish, shortens tool life, and limits productivity. Conventional model-based controllers rely heavily on accurate identification of system dynamics and cutting force coefficients, which requires extensive modal testing and chip formation analysis. To overcome these limitations, this study proposes a partially model-free adaptive control approach based on Integral Reinforcement Learning (IRL) for chatter suppression in milling processes. Unlike traditional controllers, the proposed IRL-based design depends solely on the equivalent mass matrix, eliminating the need for stiffness, damping, or cutting force parameters. The control framework adopts an actor-critic learning structure in which the critic network estimates the value function, while the actor iteratively updates the control policy to minimize the Bellman error. To guarantee closed-loop stability in the presence of state delay, a Lyapunov-Razumikhin function is employed, ensuring asymptotic convergence of the estimation and control errors. The resulting control law drives both the control force and the state variables to operate within their optimal ranges, minimizing the quadratic performance index without prior model knowledge. Simulation results show that the proposed adaptive controller effectively suppresses regenerative chatter and substantially enlarges the chatter-free region in the Stability Lobe Diagram, allowing stable cutting at higher spindle speeds and greater axial depths.

کلیدواژه ها:

Active Chatter Suppression ، Partially Model-free Control ، Actor-Critic ، Milling