Anti‐Disturbance Model Predictive Control for Triaxial Servo System Based on Q‐Learning Algorithm Friction Identification

ABSTRACT This article proposes a novel model predictive control (MPC) strategy for a triaxial servo system with the capability of suppressing multiple disturbances. The discrete state‐space model is built for a triaxial servo system, considering nonlinear disturbances as constraints, and Q‐learning algorithm is used to identify the parameters of the nonlinear friction disturbance model. In the proposed method, a new three‐part cost function was designed. It combines tracking error and control input like the traditional cost function, and adds a disturbance term. This enables the consideration of disturbances when selecting control inputs. Additionally, the optimal control law is derived through differentiation, and its stability is proven. The experimental results show that, compared to the traditional MPC, our proposed approach offers a faster dynamic response and superior steady‐state accuracy. In three‐degree‐of‐freedom experiments, the speed ripple has been reduced by 61%, although computing time has increased by 12%.