Novel AI based On-Line Sequential Learning Technique for High Performance DC Servo motor Control

Abstract

In this paper a neuro-fuzzy based adaptive tracking controller which is trained when the controller is operating in an online mode for high performance DC servo motor control is presented. The proposed structure consists of five layer feed-forward network which is trained using sequential learning method. Extreme Learning Machine (ELM), a recently developed novel method for the training of the single hidden layer feed forward neural networks (SLFNs) is used to initialize the training algorithm with a small chunk of training data. The membership function for each rule is determined using heuristics based methods and the consequent parameters of the Tagaki-Sugeno-Kang (TSK) type fuzzy inference are then determined in an online manner using the recursive least square method. The performance of the proposed technique in terms of the training time, training accuracy for tracking a reference trajectory is evaluated and is compared with the adaptive neuro-fuzzy based controller and other existing faster training algorithms such as ELM. The robustness of the proposed scheme is tested under DC motor parameters variations such as armature resistance, viscous friction and moment of inertia for all implemented controllers. Results obtained ensure the robustness of the proposed controller versus other implemented controllers.