Intelligent Wear Condition Prediction of Ball Bearings Based on Convolutional Neural Networks and Lubricating Oil

• Published in: Journal of failure analysis and prevention Vol. 24; no. 4; pp. 1854 - 1864
• Main Authors: Sun, Jiasi, Bu, Jiali, Guo, Xiaopeng, Su, Changqing
• Format: Journal Article
• Language: English
• Published: Materials Park Springer Nature B.V 01.08.2024
• Subjects: Artificial neural networks; Back propagation networks; Ball bearings; Gas turbines; Lubricating oils; Network reliability; Neural networks; Preventive maintenance; Rotating machinery
• ISSN: 1547-7029; 1864-1245
• DOI: 10.1007/s11668-024-01972-0

Ball bearings play a pivotal role in rotating machinery, making it crucial to detect their wear condition in real time for preventive maintenance. This paper proposes a novel method, involving one-dimensional convolutional neural network (1DCNN) and lubricating oil, which is sensitive to early abnormal wear, to intelligently predict the wear condition of ball bearings. To achieve that, an industrial-level gas turbine is used in the test to simulate the practical working conditions. During the machine’s operation, the lubricating oil is sampled at different time intervals for spectral oil analysis (SOA), which is later utilized as the input data set. The 1DCNN model is then constructed and fine-tuned to extract its wear characters, and finally outputs a one-dimensional time-sequence feature vector of the SOA data, providing wear information for the upcoming hours. The predicted results have shown that the 1DCNN model can effectively predict the trend of wear condition with higher accuracy of over 97% and better generalization performance compared with the BP neural networks. Additionally, the validation test results have also verified the reliability of the 1DCNN model, which can be useful for manufacturing industries in reducing costly maintenance fees by early detection of potential failure.