Design of Static Neural Network-Based Controller for Brushed DC Motors Using Bayesian Inference and Quasi-Newton Method

This paper presents an effective speed control method for a brushed DC motor fed by a DC chopper using a two-layer feed-forward neural network, which is a popular type of static neural network. Based on the relationship between the voltage applied to the armature circuit and the rotor angular velocity, a dataset can be generated for training the neural network offline. To handle the uncertainty of the system and prevent the trained network from overfitting, a regularisation is performed with the use of Bayesian inference to find out the appropriate regularisation parameters. In addition, the quasi-Newton optimisation method is used to effectively minimise the cost function, which is a highly nonlinear function of weights and biases. The performance of the proposed control algorithm is validated on an experimental motor drive system in a university laboratory. The proposed controller is implemented using the MATLAB Package for Arduino. The experimental results obtained show that the neural network controller can significantly outperform the conventional proportional-integral (PI) controller as it can result in a shorter settling time and a lower overshoot compared to these when using the PI controller.