Sensorless Speed Estimation for Line-connected Induction Motor Based on Recurrent Multilayer Neural Network

A new method for rotor sensorless speed estimation of line-connected induction machines is proposed in this paper. Considering that it is difficult to install speed sensor for submersible induction motor in some specific situations such as special high temperature working environment, the proposed technique of recurrent multilayer neural network is used to estimate the speed of sensorless submersible induction motor. The stator current measured by data collector is analyzed by wavelet, thus the influence of high frequency noisy caused by high temperature is filtered off. Since the useful signal is extracted as sample input, and the speed signal collected by speed sensor as sample output, a neural network is trained on the principle of "training off-line, estimating on-line ", so that the network can estimate the speed only using stator current. Experimental results prove that the proposed method has very high precision and good dynamic quality. Furthermore, the estimation results can provide powerful security for closed-loop control and fault diagnosis.