Data-Driven Model Predictive Control for Speed Temperature Drift Compensation of Rotary Traveling Wave Ultrasonic Motors

Abstract

This paper proposes a data-driven model predictive control (MPC) framework for high-precision speed control of rotary traveling wave ultrasonic motors (RTWUSMs) under temperature drift. To address the strong nonlinearity and time-varying thermal characteristics of RTWUSMs, a Koopman–convolutional neural network–long short-term memory (Koopman–CNN–LSTM) prediction model with radial basis function (RBF) observable features is constructed. The model maps the nonlinear electromechanical coupling and friction-driven dynamics of the motor into a linear invariant subspace, achieving high prediction accuracy while maintaining low computational complexity. On this basis, a data-driven MPC scheme is designed, which eliminates the dependence on accurate analytical plant models and compensates for thermal-induced speed drift through online driving frequency adjustment. The experimental results show that under 900 s of continuous operation, the proposed scheme achieves a relative steady-state speed error of 0.37%, which is significantly better than typical temperature drift compensation methods. This scheme can also provide stable tracking performance under load torques of 0.5 N·m and 1.0 N·m, providing a practical solution for the long-term stable speed regulation of RTWUSM in precision drive applications.