Fault Diagnosis of Rotating Machines Using Raw Vibration Signals and Deep Learning

Abstract

Rotating machines are important mechanical equipment used in many areas of industry. Component failures are inevitable as these equipment are often used under long and harsh conditions. Component failures that may occur can be predicted without stopping the machines, by collecting data on their operating conditions and status, and by monitoring the changes in this data over time. The most comprehensive data on the condition of the machine is obtained by vibration analysis. The patterns created by the vibrations change at the onset of the fault. When these patterns are interpreted correctly, future failure can be predicted. In this way, maintenance operations are applied to the machine before component failure and unexpected failure stops are prevented. In this article, a deep learning-based fault classification model is proposed using bearing vibration data of rotating machines. The proposed method is based on transforming raw vibration signals into images and classifying them with a convolutional neural network. In the application for the diagnosis of four different states of the rotating machine, an accuracy of 100% has been achieved. © 2021 IEEE.