Research on Motor Fault Identification Method Based on Vibration Current Multisource Data Fusion and Deep Learning

Traditional motor fault diagnosis techniques are usually based on a single type of state parameter. However, the monitoring range of a single type of motor state parameters is very limited, and it is difficult to meet the needs of comprehensive fault diagnosis. The purpose of this paper is to propose a fault diagnosis method for typical AC induction motors by fusing two types of parameters, i. e. vibration data and current data, in order to improve the accuracy of diagnosis. Based on the analysis of the basic faults of typical AC asynchronous motors, this paper studies the methods of motor fault feature extraction, including Fast Fourier Transform (FFT) and Hilbert-Huang Transform (HHT). Then, the method of motor multi-sensor parameter fusion is studied, and the realization scheme of data fusion is designed. Convolutional neural network is constructed as a feature fusion tool. The results of fast Fourier transform (FFT) of current data and Hilbert-Huang transform (HHT) of voltage data of different fault types are input into CNN as fault features, and the deep-seated features are extracted by using CNN's feature extraction ability. Finally, the CNN network series Multilayer perceptron (MLP) was constructed to realize fault classification, and the motor fault identification and fault diagnosis were realized based on the motor vibration and current parameters. The results show that the motor fault diagnosis technology based on data fusion can improve the accuracy of diagnosis results, reduce the uncertainty caused by single parameter, and improve the accuracy of motor fault diagnosis.