A Bayesian machine learning-driven control algorithm for one-phase winding permanent-magnet brushless DC machines

The one-phase brushless DC motor (BLDC) has become indispensable in-home appliances due to its high-power density, flexible control, and straightforward driving circuit, outperforming induction and universal motors. Additionally, it ensures higher efficiency across a wide range of speed-torque loads. This paper introduces a pioneering real-time control algorithm based on machine learning to enhance the BLDC motor’s overall performance compared to the traditional fuzzy-PID controller. A dynamic model of the BLDC motor is utilized to determine the EMF (electromotive force) and torque properties through finite element simulations conducted in the ANSYS/Maxwell environment. The targeted BLDC motor is driven by a space vector modulation inverter powered by a DC voltage source. The proposed machine learning-based control algorithm demonstrates superior performance over traditional methods under various load disturbances and reference speed variations, with overshoot/undershoot and settling time improvements of at least 60% and 46%, respectively. The enhanced performance was validated using a comprehensive dynamic model developed in the MATLAB environment and confirmed through an experimental setup.