Diagnosis and prognosis of rolling element bearings using audio signals and deep learning
Date of Award
Master of Engineering Science
School of Engineering
Condition monitoring and fault diagnosis of industrial equipment have become increasingly important in the wake of the Fourth Industrial Revolution. Unexpected failures of common industrial equipment components, such as rolling-element bearings, can cause production downtime and the resulting financial losses can be significant for large industries including mining. Proactive maintenance strategies can help mitigate unexpected shutdown events. Real-time condition monitoring of industrial equipment plays a critical role in proactive maintenance.
This study presents an intelligent diagnostic system for real-time condition monitoring of rolling-element bearings based on their acoustic signatures. In the literature, there is no comprehensive audio repository available corresponding to various types of bearing faults. Consequently, a special test rig was built for improvising various types of bearing faults and data collection. A comprehensive data set of audio samples corresponding to various bearing faults was prepared. This data set was then used to develop an intelligent diagnostic system, which shows promising results for bearing fault diagnosis. The proposed deep neural network architecture provides an overall macro accuracy of 92.2% in classifying five modes of operation.
Similar to bearing diagnosis, the prognosis of bearings, especially related to critical machines and plants, has received immense attention. The ability to predict the remaining useful life (RUL) of rolling-element bearings is beneficial to heavy industries, as the predictions provide early warnings to mitigate unexpected shutdowns. The study proposes a deep convolutional neural network and log-scaled mel-spectrogram-based intelligent system to estimate the stage RUL of bearings. The proposed method is validated using experimental data. and 9 stage classification shows the highest accuracy of 98%.
Access to this thesis is embargoed until 08 09 2023. At the expiration of the embargo period, access to the thesis will be restricted to current ECU staff and students. Email queries to email@example.com
Darshana, W. G. (2021). Diagnosis and prognosis of rolling element bearings using audio signals and deep learning. https://ro.ecu.edu.au/theses/2447