| Title |
A Study of Vibration and Current Data Characteristic Analysis for Motor Mechanical Fault Level Determination by Deep Learning |
| Authors |
한지훈(Ji-Hoon Han) ; 박상욱(Sang-Uk Park) ; 홍선기(Sun-Ki Hong) |
| DOI |
https://doi.org/10.5370/KIEE.2021.70.10.1509 |
| Keywords |
Deep learning; Motor fault diagnosis; DT-CNN; Vibration signal; Current signal; Fault level determination |
| Abstract |
In the classic motor fault diagnosis system, a method of analyzing the differences between the normal and collected state signals of the motor to be diagnosed was used and the method can diagnose only the limited situations because the diagnosis is based on the frequency of the mechanical failure. In order to compensate for this, some studies on a system that performs more specialized fault diagnosis through deep learning algorithms were carried out. However, the level of failure cannot be determined because these studies consider only the signals that have a great influence on motor operation. To solve this problem, the characteristics of vibration and current signals are analyzed to develop a deep learning algorithm suitable for fault level determination. The characterized signals are used for fault diagnosis and prediction. Fault diagnosis based on vibration signal is carried out through DT-CNN (Decision Tree Convolutional Neural Network). In addition, it is checked whether the current signal in the initial failure state, which is relatively insensitive to failure, can be classified through a deep learning algorithm. The proposed data utilization performance was evaluated through an induction motor and the analyzed signal-based fault diagnosis system is expected to enable a more precise diagnosis compared to the existing system. |