Application of the Stress Waves to Extract Multi-fault Features of the Low-Speed Machinery Based on Blind Source Separation

This paper proposed a non-linear signal processing method to separate the stress wave signals from different sources for extracting the multi-fault features without any a priori knowledge. Firstly, stress wave theories of low-speed rotating machines are discussed. And then this paper described the basic principles of BSS, and a kind of non-linear neural network is used in consideration of the complex behavior of vibration, either nonlinear or random. After that BSS was applied to the experimental simulations for diagnosing the rubbing and looseness defects. Finally the processing results of a real process show that the new method based on non-linear blind source separation studied in this paper can simply and effectively remove the noise contained in the vibration signals and extract the multi-fault features to avoid the grand accidence.