Hybrid deep reinforcement learning-assisted adaptive TID controller for sensorless BLDC motor drives with torque ripple minimization and dynamic stability enhancement

This paper proposes a hybrid Deep Reinforcement Learning (DRL) assisted adaptive Tilt-Integral-Derivative (TID) controller for sensorless Brushless Direct Current (BLDC) motor drives with enhanced dynamic stability and torque ripple minimization. The proposed framework integrates an adaptive back-EMF observer for sensorless rotor position estimation with a Deep Deterministic Policy Gradient (DDPG)-based online gain tuning mechanism for real-time optimization of TID controller parameters. Unlike conventional fixed-gain controllers, the proposed method continuously adapts control gains to operating conditions, improving transient response, robustness and disturbance rejection. A multi-objective reward function incorporating speed tracking error, torque ripple,and RMSE is formulated to optimize dynamic and steady-state performance simultaneous. Lyapunov-based stability analysis is presented to guarantee asymptotic convergence of the closed-loop system under bounded disturbances. The proposed controller is validated through MATLAB/Simulink simulation and real-time DSP-based hardware implementation using a TMS320F28379D controller platform. Comparative analysis with PI, PID, FOPID, TID-GWO and adaptive TID controllers demonstrates that the proposed DRL-TID controller achieves a 79% reduction in settling time, 85.5% reduction in torque ripple and 92.3% improvement in ITAE compared with the conventional PI controller. Experimental results further confirm stable sensorless operation, fast speed recovery, low harmonic distortion and robust performance under dynamic load and parameter uncertainty conditions.