A novel bearing fault detection approach using a convolutional neural network

Abstract

Bearing fault detection is an important part of mechanical equipment and rotating machinery. Bearing failure should be detected early because it can lead to property and safety losses. This study proposes convolutional neural network (CNN) based models for bearing fault detection. Since the main advantages of the proposed methods apply to different types of warehouse data, failure can be detected in a short time and applied directly to raw data. These new models achieve comparable or better performance compared to the existing models in the literature. Although the structure of the proposed models is simpler and the number of parameters used is smaller, these new models achieve successful empirical results. Data sets from CWRU and IMS were used to test the models. This study compares the proposed models with the existing models in the literature. It also compares the new models with the machine learning algorithms and obtains better empirical results.