Online tribology ball bearing fault detection and identification
Document Type
Conference Proceeding
Publication Date
11-15-2007
Abstract
We present a feasibility analysis for the development of an online ball bearing fault detection and identification system. This system can effectively identify various fault stages related to the evolution of friction within the contact in the coated ball bearings. Data are collected from laboratory experiments involving forces, torque and acceleration sensors. To detect the ball bearing faulty stages, we have developed a new bispectrum and entropy analysis methods to capture the faulty transient signals embedded in the measurements. Test results have shown that these methods can detect the small abnormal transient signals associated with the friction evolution. To identify the fault stages, we have further developed a set of stochastic models using hidden Markov model (HMM). Instead of using the discrete sequences, our HMM models can incorporate the feature vectors modeled as Gaussian mixtures. To facilitate online fault identification, we build an HMM model for each fault stage. At each evaluation time, all HMM models are evaluated and the final detection is refined based on individual detections. Test results using laboratory experiment data have shown that our system can identify coated ball bearing faults in near real-time.
Publication Source (Journal or Book title)
Proceedings of SPIE - The International Society for Optical Engineering
Recommended Citation
Ling, B., & Khonsari, M. (2007). Online tribology ball bearing fault detection and identification. Proceedings of SPIE - The International Society for Optical Engineering, 6555 https://doi.org/10.1117/12.719416