Power Semiconductor Selection and Drive Circuit Design for Brushless DC Motors in Unmanned Aerial Vehicles

Abstract

Brushless direct-current (BLDC) motors are widely used in unmanned aerial vehicle propulsion, where power-semiconductor selection directly affects inverter loss, junction temperature, and electrical operating margin. However, conventional rating- or figure-of-merit-based selection provides limited representation of variable motor loads, while purely data-driven approaches may overlook explicit electrical and thermal constraints. This study proposes PCTO-BLDC, a physics-constrained computational framework that combines synchronized current–speed feature encoding, electro-thermal loss modeling, a lightweight neural surrogate, feasibility aggregation, and joint optimization of power devices and driver parameters. The surrogate is trained using physics-derived semiconductor-loss and junction-temperature targets, while candidate solutions remain subject to deterministic voltage, current, and thermal limits. Across ten repeated experiments, PCTO-BLDC achieved an average semiconductor loss of 4.55 ± 0.19 W, a peak junction temperature of 80.7 ± 2.5 C, and a robust feasibility rate of 95.1 ± 2.1%. External evaluation on DUDU-BLDC 1.5 retained a feasibility rate of 89.7 ± 3.6%. These results indicate that integrating measured operating profiles with physics-constrained computational optimization can support reproducible and interpretable power-stage design under defined BLDC operating conditions.