Implementation of ANN-based sensorless induction motor drives

Abstract

The paper discusses the DSP implementation of various speed-sensorless induction motor drives which incorporate artificial intelligence (AI). The first drive is a medium performance induction motor drive and contains a minimal configuration neural-network-based speed estimator. Although the neural network was trained by using simulation results only, it was successfully implemented in a drive employing a 3 kW cage induction motor. However, the same speed estimator ANN was also successfully used in a drive with a 2.2 kW motor. Speed estimators using feedforward multilayer and recursive artificial neural networks (ANNs) are also compared. In addition to an ANN-based speed estimator, the second drive contains a simple fuzzy-logic-based system with a minimal rule-base, which improves the low-speed performance. The third drive is an improved speed-sensorless DTC drive employing a simple predictive torque error minimization technique. The experimental results show that the implemented AI-based drives give satisfactory performance in a wide speed range. The drive schemes are simple to implement and the memory requirements are modest. The DSP used is the TMS320C30.