Motor State Prediction and Friction Compensation for Brushless DC Motor Drives Using Data-Driven Techniques

In order to provide robust, reliable, and accurate position and velocity control of motor drives, friction compensation has emerged as a key difficulty. Non-characterised friction could give rise to large position errors and vibrations which could be intensified by stick-slip motion and limit cycles. This paper presents an application of two data-driven nonlinear model identification techniques to discover the governing equations of motor dynamics that also characterise friction. Namely, the extraction of low-power data from time-delayed coordinates of motor velocity and sparse regression on nonlinear terms was applied to data acquired from a Brushless DC (BLDC) motor, to identify the underlying dynamics. The latter can be considered an extension of the conventional linear motor model commonly used in many model-based controllers. The identified nonlinear model was then contrasted with a nonlinear model that included the LuGre friction model and a linear model without friction. A nonlinear grey box model estimation method was used to calculate the optimum friction parameters for the LuGre model. The resulting nonlinear motor model with friction characteristics was then validated using a feedback friction compensation algorithm. The novel model showed more than 90% accuracy in predicting the motor states in all considered input excitation signals. In addition, the model-based friction compensation scheme showed a relative increase in performance when compared with a system without friction compensation.