Fault diagnosis method based on wavelet transform and CNN for rotating machinery

초록

Fault diagnosis technology and condition monitoring of facilities used in major production processes have become an important issue in recent industrial development. This paper proposes fault diagnosis and abnormal condition monitoring method by analyzing vibration signals generated from bearing part of rotating equipment based on wavelet transform (WT) and convolutional neural network (CNN). The vibration signals of the bearings installed in the rotating equipment usually include dynamic information of the rotating system, which is very useful for defect shape extraction. Vibration signals are acquired by sensors attached to the rotating machine. Fault diagnosis methods focus on distinguishing signals under different conditions, during which an effective feature extraction method is very important. Classic fault diagnosis techniques generally rely on features generated through traditional signal analysis methods, such as frequency domain analysis or statistical analysis. However, at the signal processing stage, it requires various technologies and know-how such as noise signal filtering and defect frequency domain analysis by various operating environments. Wavelet analysis is especially suitable for non-linear and non-stationary signals, while it can effectively increase the signal-to-noise ratio. In this paper, a noble approach for fault diagnosis and feature extraction was proposed. The WT is applied to the measured vibration signal and divides it into different scales. The frequency spectrum of the different signals extracts features through the CNN model. In addition, the classifier is eventually used as an error classification method. The integrated structure allows to classify the vibration signal into different working conditions. Since this procedure can be proposed as a method for analyzing a vibration signal measured in real time to detect an abnormal state.