Supervised Learning for Adaptive BLDC Motor Control: Integrating Classical PID with Neural Networks

Abstract

BLDC motors require a high level of precision and efficiency in controlling speed, which is needed in a wide range of applications, such as electric vehicle, robotics, and industrial automation. The classical proportional-integral-derivative (PID) controllers are often characterized by overshoot, a slow rate of convergence, and have a low flexibility to changing operating conditions. As a counter to this we have suggested in this paper a hybrid artificial-intelligence-based control scheme, a combination of a neural-network controller and the standard PID control. A large dataset based on controlled simulations, using PID control, is then applied in training a feed-forward neural network to estimate optimal control behaviour using a supervised learning approach. The neural network that results is an adaptive controller that adjusts dynamically the control voltage in real time as a function of the difference between the current speed and its derivative. The experimental evidence shows that neural-network controller performs better than the traditional PID control in terms of eliminating overshoot (0% vs. 12.13%), the settling time by up to 88% (0.8826 s vs. undefined for PID) and the Integral Absolute Error (IAE) by over 40% (86.96 vs. 151.64). Besides, the AI-based system produces more fluent controlvoltage curves, reducing mechanical forces, and promoting energy savings. This paper highlights the potential of hybrid neural-PID control systems to the high-performance motor control of BLDC motors and outlines future research opportunities in this area to address real-time operation and computational bottlenecks within embedded computer systems.