Classification of Induction Motors by Fault Type with bidirectional Long-Short Term Memory Method

Abstract

It is important to determine the initial level of failures of induction motors used in many industrial applications. The sudden stops of the system can be prevented with the pre-detection of the fault. The experiment mechanism was established to detect mechanical unbalance and short circuit faults in the induction motors. Current values were measured and saved at fault time. As a result, 9.000 data were obtained consisting of 3 phase currents. In this study, a Long-Short Term Memory (LSTM) deep neural network has been developed that classification of induction motors by fault type. In the training of the neural network, 3 input parameters and 3 classification types of 1 output parameter are used. It was reserved for training 60% of data and 40% for testing the model in the dataset. As a result of the fault type classification with the LSTM model, 98.5% accuracy and 1.12 average absolute error value were obtained. It has been shown that the proposed bi-LSTM network can be used for fault detection of asynchronous motors.