Torque Ripple Minimization for A Bearing-Less Synchronous Motor Using Neural Network Based Control Technique.

Reliable torque production in bearing-less synchronous motors is hindered by torque ripple, which increases vibration, acoustic noise, and speed fluctuation, particularly at low speeds. This study develops a compact torque-ripple assessment and minimization framework for two BSM prototypes (37 kW/400 V and 11 kW/690 V) using a DSP-based experimental platform integrated with MATLAB/Simulink. The method identifies key ripple sources—cogging torque, harmonic distortion, and reluctance torque due to d–q axis interaction—and quantifies ripple severity via a torque ripple factor (TRF). A lightweight feed-forward neural network is trained to predict electromagnetic torque from d–q currents, rotor position, and rotor speed, with parameters optimized by minimizing mean squared error. Experimental results show periodic cogging torque with pulsations every 30° and pronounced effects below 500 rpm, a dominant 3rd torque harmonic of 0.2 Nm among five significant components, and total torque varying between 8.2 Nm and 11.8 Nm, with reluctance torque contributing about 15% of the ripple. Frequency-domain analysis reveals dominant ripple bands at 50 Hz, 100 Hz, and 150 Hz, with the 100 Hz component approximately 2.5 times stronger than the fundamental. The neural model converges within 40 epochs, reducing MSE from 0.52 to 0.08 and predicting torque within ±0.5 Nm over the 10–12.5 Nm range. The proposed low-complexity framework effectively characterizes torque ripple contributors, enables accurate torque estimation, and suggests prioritizing mitigation of dominant low-order harmonics, especially around 100 Hz.