Hybrid physics-informed and data-driven model for torque estimation—toward real-time sensorless control of ultrasonic motors

Abstract

Precise torque control in ultrasonic motors is important in applications such as robotics and medical devices. Sensorless torque control is especially crucial for achieving system miniaturization and integration. However, existing methods either rely on physically large and costly torque sensors or are insufficient in terms of accuracy, real-time performance, and adaptability, making it difficult to meet the requirements of high-performance and integrated applications. To overcome these problems, this study proposes a hybrid ultrasonic motor torque estimation model that combines physics-informed and data-driven approaches for sensorless control on edge devices. In the physics-informed approach, the stator-rotor contact model is combined with the equivalent circuit model. Electrical signals are used to replace the internal state variables, which are difficult to obtain directly, and nonlinear compensation is applied to the measured vibration amplitude. In the data-driven approach, with real-time multi-source information taken as input, two parallel lightweight neural networks are designed, namely a physics parameter correction network, which corrects the key parameter of the physical model in real time to adapt to changes in working conditions, and a residual correction network, which learns and compensates for the predicted residuals of the physical model. An attention mechanism guided by SHAP analysis is introduced to improve estimation performance. The evaluation results show that the proposed model has good estimation accuracy and generalization ability. Finally, precise, stable, and low-latency online sensorless torque control is achieved on a microcontroller.