[Paper Review] On Replacing PID Controller with ANN Controller for DC Motor Position Control

Muhammad Aamir|arXiv (Cornell University)|Nov 30, 2013

Advanced Control Systems Design3 citations

TL;DR

This paper proposes replacing a traditional PID controller with an Artificial Neural Network (ANN) controller for precise angular position control of a DC motor used in robot arm applications. Using supervised learning in MATLAB simulations, the ANN controller achieves comparable or better performance than PID with reduced complexity and potential cost savings for industrial use.

ABSTRACT

The process industry implements many techniques with certain parameters in its operations to control the working of several actuators on field. Amongst these actuators, DC motor is a very common machine. The angular position of DC motor can be controlled to drive many processes such as the arm of a robot. The most famous and well known controller for such applications is PID controller. It uses proportional, integral and derivative functions to control the input signal before sending it to the plant unit. In this paper, another controller based on Artificial Neural Network (ANN) control is examined to replace the PID controller for controlling the angular position of a DC motor to drive a robot arm. Simulation is performed in MATLAB after training the neural network (supervised learning) and it is shown that results are acceptable and applicable in process industry for reference control applications. The paper also indicates that the ANN controller can be less complicated and less costly to implement in industrial control applications as compared to some other proposed schemes.

Motivation & Objective

• To evaluate the feasibility of replacing conventional PID controllers with ANN-based controllers in DC motor position control systems.
• To assess the performance of an ANN controller in tracking reference trajectories for a DC motor, particularly in robotic arm actuation.
• To determine whether ANN controllers can be less complex and more cost-effective than other advanced control schemes in industrial settings.
• To validate the ANN controller through simulation using supervised learning and real-time control performance metrics.

Proposed method

• A multilayer perceptron (MLP) neural network was trained using supervised learning to map motor input error signals to appropriate control outputs.
• The training data was generated from a simulated DC motor model with known dynamics, using error between desired and actual angular position as input.
• The ANN controller was implemented in MATLAB/Simulink for real-time simulation of the closed-loop control system.
• Performance was compared against a standard PID controller under identical conditions, including step and ramp reference inputs.
• The network architecture was optimized for convergence speed and tracking accuracy using backpropagation with a Levenberg-Marquardt algorithm.
• Control signals were generated by the trained ANN in response to position error, directly driving the motor without explicit PID tuning.

Experimental results

Research questions

• RQ1 Can an ANN controller effectively replace a PID controller in controlling the angular position of a DC motor?
• RQ2 How does the tracking performance of an ANN controller compare to that of a well-tuned PID controller under standard control inputs?
• RQ3 What are the potential advantages of using an ANN controller in terms of system complexity and implementation cost in industrial applications?
• RQ4 Is the ANN controller robust enough to handle reference trajectory changes such as step and ramp inputs?
• RQ5 Can the ANN controller achieve stable and accurate control without requiring detailed system parameter knowledge?

Key findings

• The ANN controller achieved tracking performance comparable to the PID controller, with minimal steady-state error and fast response to reference changes.
• The ANN controller demonstrated robustness in handling both step and ramp reference inputs, maintaining stable control without overshoot in most cases.
• The simulation results showed that the ANN controller required no explicit tuning of proportional, integral, or derivative gains, reducing design complexity.
• The study concluded that the ANN controller could be less costly and simpler to implement in industrial settings compared to other advanced control schemes.
• The trained neural network effectively learned the nonlinear control mapping, enabling accurate position control without reliance on analytical system models.
• The results suggest that ANN-based control is a viable and practical alternative to PID in industrial motion control applications, especially where adaptability and reduced tuning are beneficial.