Vibration Sources Identification with Independent Component Analysis
Hongxian Ye, Shixi Yang, Jiangxin Yang, Huawei Ji · 2006
Independent Component Analysis (ICA) could recover the original sources in multichannel observations when the original sources and the mixing process are not known. We applied this technique to linear mixtures of simulated fault signals. The success separation inspired us to apply the technique to an experiment set-up where three known sources were used as an analogue of the real world in order to separate each source. Our goal was to remove the influence of the other machines and recur the vibration source in gearbox. The mixing model of signal transmission was investigated. The experiment results show that when the signal frequency is less than 100hz, it is appropriate to consider the mixing model of vibration sources as instantaneous model. The separation results by mean of FastICA algorithm show that we can separate the disturbing signal and inner source signals from the observation. The experiment results suggest that ICA is a promising method to enhance the accuracy of fault diagnosis.