RBFNN-Enabled Adaptive Parameters Identification for Robot Servo System Based on Improved Sliding Mode Observer

Abstract

Effective and accurate parameter identification, especially the identification of load torque, is one of the key factors to improve the control performance of the robot servo system. Sliding mode observer (SMO) has always been a common method for identifying load torque due to its advantages of simple implementation, strong robustness, and fast response. However, due to the discontinuity of the SMO switching function, the system will generate high-frequency chattering, which will reduce the accuracy of load torque identification and affect system performance. In this paper, an adaptive parameter identification method based on an improved sliding mode observer is proposed. A continuous deformation mode of saturation function based on boundary variation is proposed as the switching function to alleviate the chattering phenomenon. Meanwhile, the relationship between the sliding mode gain and the feedback gain of proposed SMO is defined so that it can be selected properly to improve the accuracy of identification, and the radial basis function neural network (RBFNN) is used to adaptively tune the boundary layer gain according to the speed change. Moreover, considering that the identification result of the load torque is related to the moment of inertia and the mismatch of the inertia will cause identification errors, the variable period integration method is proposed to identify the inertia and redefine the calculation period of the load torque and inertia. The effectiveness and superiority of the proposed method are verified by simulation experiments. Experimental results demonstrate that the improved SMO combines observer gain coefficient tuning and inertia matching can smoothly and accurately estimate the value of load torque, which is an adaptive identification method worthy of reference for robot servo system.