Online identification and control of a DC motor using learning adaptation of neural networks

This paper tackles the problem of the speed control of a DC motor in a very general sense. Use is made of the power of feedforward artificial neural networks to capture and emulate detailed nonlinear mappings, in order to implement a full nonlinear control law. The random training for the neural networks is accomplished online, which enables better absorption of system uncertainties into the neural controller. An adaptive learning algorithm, which attempts to keep the learning rate as large as possible while maintaining the stability of the learning process is proposed. This simplifies the learning algorithm in terms of computation time, which is of special importance in real-time implementation. The effectiveness of the control topologies with the proposed adaptive learning algorithm is demonstrated. It is found that the proposed adaptive leaning mechanism accelerates training speed. Promising results have also been observed when the neural controller is trained in an environment contaminated with noise.