AI-supervised adaptive PID control of a DC motor drive using a bounded feedforward neural network under load disturbances and parameter uncertainties

This paper proposes a neural-network-supervised adaptive PID controller for an electric motor drive subjected to load disturbances and electrical and mechanical parameter uncertainties. The proposed controller combines a conventional PID structure with a feedforward neural network that adjusts the proportional, integral, and derivative gains according to the instantaneous tracking error, accumulated error, error variation, armature current, and motor speed. The selected neural supervisor contains five inputs, two hidden layers with 12 and 8 neurons, and three bounded outputs corresponding to the PID gains. The network is trained offline using optimal gain targets obtained under different speed references, load levels, and motor-parameter variations. During closed-loop operation, the trained network performs deterministic gain inference at every sampling instant, while gain projection, voltage saturation, and anti-windup compensation ensure safe controller operation. The proposed method is evaluated in MATLAB/Simulink using a separately excited DC motor model and is compared with a fixed-gain PID controller under nominal operation, reference changes, load disturbances, parameter uncertainties, and combined adverse conditions. The results demonstrate the effect of neural gain scheduling on transient tracking, disturbance rejection, steady-state accuracy, and control effort.