Rotor Spherical Region Location of Permanent Magnet Spherical Motor Based on Improved Random Forest

Real-time acquisition of rotor position information is a prerequisite for closed-loop control of spherical motors. The Hall effect sensor array based on the quadrature triangle model can locate the rotor position in the spherical intervals. Increasing the number of sensors can make the spherical area more precise, and thus the positioning accuracy is improved, but it is limited by the installation space and hardware circuit complexity. Therefore, without increasing the number of sensors, this paper introduces a Random Forests (RF) algorithm to improve the accuracy of identification. Based on the results of spherical interval identification, the RF algorithm can reduce the amount of data by one or two orders of magnitude. By combining these two methods, the speed of interval recognition and the high accuracy of machine learning algorithms are fully utilized. To reduce the interference of bad data and improve the data processing capability, an Improved Random Forests (IRF) algorithm is proposed. Finally, the proposed algorithm is experimentally validated. The maximum positioning error is only 0.6°, which is a low-cost and high-precision spherical motor position recognition solution.