Cascaded Hybrid Soft Computing Controllers using Deep Reinforcement Learning and Fuzzy SARSA Learning for Enhanced BLDC Motor Performance

Abstract

BLDC motors have efficiency, miniaturization, and low maintenance requirements that make them a crucial aspect of modern electric drive systems. Inherent nonlinearities and parameter uncertainties, together with high torque and speed transitions, generally worsen the performance of conventional controllers in a BLDC system. In this regard, this work proposes a Cascaded Hybrid Soft Computing Controller that merges DRL with Fuzzy SARSA(λ) learning for intelligent, adaptive, and data-driven control of BLDC motors. A cascaded architecture combines the global policy optimization of DRL with the fuzzy rule-based adaptability of fuzzy reinforcement learning to guarantee real-time stability with quicker convergence and higher learning accuracy. Simulation output shows the superior transient and steady-state performance of the developed hybrid controller. The developed method registers a settling time of 0.28 s, steady-state speed of 1498 RPM, torque ripple of less than 2.95 N·m, and remarkably stable DC-link voltage around 300 V when compared to Fuzzy SARSA(λ)-only and MPC control strategies. These results confirm that the controller has superior robustness, adaptability, and accuracy under speed reversals and sudden load disturbances. The resulting cascade scheme therefore offers a strong direction for future intelligent motor drives and autonomous energy-efficient actuation systems