CNN-based fault classification considered fault location of vibration signals

Abstract

Recently, with the development of the 4th industrial technology such as big data, cloud computing, and IoT, technologies that automatically perform specific tasks without human intervention are being applied in many industrial sites. In the case of rotating equipment diagnosis, features are extracted based on the shape and statistical information of the time and frequency signals of the vibration data for each location, and the acquired data is classified as normal or defective by learning it. However, since this method used the shape and statistical information of the vibration signals, the physical meaning is blurred and the information is not meaningful for actual diagnosis, resulting in inconsistent learning models even in the same facility. In this study, the possibility of classifying normal and fault condition were confirmed by generating images considering the fault component and sensor location of the vibration signal and applying to CNN-based deep learning technology. As a method of performing image processing, STFT is performed on the acquired vibration signal data for each position to generate an image. In addition, converting each sensor position attached to red, green, and blue to express location information, resynthesis was performed to configure learning data and create a classification model. In order to verify this method, verification was performed based on the data acquired for the gearbox system to confirm the possibility of classifying the fault condition. © 2023, The Korean Society of Mechanical Engineers and Springer-Verlag GmbH Germany, part of Springer Nature.