Auto-Tuning PID Controller on Electromechanical Actuators Using Machine Learning

The performance of an electromechanical actuator largely depends on the control strategy implemented and calibrated. The purpose of this paper is to analyze the shortcomings of existing methods and practices involved in the manual tuning of a Proportional-Integral-Derivative (PID) controller of a Brushless DC (BLDC) motor and therefore, to propose an improved auto-tuning method. The controller investigated is a 3-stage cascaded PID, where the outermost loop is a position-based control followed by speed and current loops, respectively. The tuning method is independent of the application and provides a fast, stable, and efficient response. The motor model is first approximated to a first-order system; then mechanical parameters of the motor such as inertia and coefficient of friction are estimated using the Recursive Least Square (RLSQ) optimization method. These mechanical parameters are used in conjunction with the steady state gain and time constant of each control loop to calculate PID gain. To further improve the performance of the controller, the load torque on the motor is also estimated using an Artificial Neural Network (ANN), which is used as a feed-forward command in the current controller to improve the response of the system. The tuning method was tested on an actuator which consisted of a BLDC motor and a gear reducer. The neural network was developed by taking extensive data on the brake fixture setup. The actuator was subjected to different loads and speeds irrespective of the application load profile. The PID gains for the actuator were then calculated using the tuning rules by running the actuator on the actual application. The results demonstrate a 90% improvement in overshoot, 45% in settling time, and a 12% improvement in current consumptions against the initial baseline manual calibration.