Study On Application Of Independent Component Analysis In Rotating Mechanical Fault Diagnosis

Independent Component Analysis (ICA) is a special method of Blind Source Separation (BSS) which was rapidly developed in 1990s. It is a hot research topic in the field of signal processing at present. Currently, ICA in mechanical vibration signal processing, biomedical signal processing, mixed speech signal separation, image denoising etc already was achieved better application effect.Under the source signal and the signal transmission channel parameters of known or completely unknown circumstances, independent component analysis is according to the input source signal's properties, only by observation signal (source signal with linear combination) through the specific optimization algorithm to extract all independent component signal process. ICA came from blind source separation method, so it is also named the blind separation. Independent component analysis method is introducted in rotating machinery fault vibration signal processing, it can effectively enhance the failure information, and solve the problem which is difficult to pick up early weak fault feature in strong noise background fault diagnosis.In this paper, we did a detailed analysis of the basic principles and algorithms of independent component analysis and the characteristics of rotating machinery fault signals, studied the application of the transient mixed-signal independent component analysis in rotating machinery fault signals , and used simulation example for the contrast verification. On this basis, this paper came up with single-channel Repeated Blind Source Separation (Re-BSS) method which based on independent component analysis, and it can solve the problem in the ICA application of source signal channel number less than the number of transmission channel and break through the limit of the number of acquisition channels. According to the single-channel Repeated Blind Source Separation (Re-BSS) method, we detailed studied the virtual channel signal characteristics (such as amplitude, frequency and phase, etc) as well as heavy noise and other factors on the heavy impact of Re-BSS methods. Single-channel data acquisition simulation and practical application examples show that the Re-BSS method is easy to operate. There is hope that we can use Re-BSS to pick up weak information and signal feature, and separate the source signal in the future.