A data-driven dynamic modeling method for servo actuators

Abstract. Servo actuators are widely used in fields such as aerospace, manufacturing, and robotics. Nonlinear factors, including friction, backlash, and transmission error, significantly affect their servo performance. Traditional modeling methods for these nonlinear factors rely on simplified analytical models, which struggle to meet the increasing demands for dynamic model accuracy in control systems. Therefore, this study proposes a data-driven modeling method for nonlinear factors. A back propagation (BP) neural network is employed to perform nonlinear regression analysis of friction and backlash. Based on order spectrum analysis and the principle of dominant order invariance, a multi-order harmonic superposition model is established to describe transmission error. The proposed modeling method has been experimentally validated and demonstrates significantly improved accuracy in nonlinear modeling. Compared with traditional models, the developed data-driven model achieves a goodness of fit exceeding 0.92 with the actual system, an average improvement of over 7 %. Moreover, it accurately captures velocity fluctuations caused by transmission errors, velocity dead zones induced by friction, dynamic backlash variations under load, and uneven friction torque at the same velocity. The proposed data-driven dynamic modeling method can provide valuable insights for accurate modeling of servo systems and controller design.