Application of Blind Source Separation to Gear fault detection

The vibration signals collected by sensors installed on a complex machine are available as a combination of vibrational energy of target component and vibrational energy of other components in machine plus noise. In order to diagnosis a component of a complex machine, its vibration signal should be separated from other signals. Blind source separation method is commonly used to separate the signals and provide signals related to each component and finally system fault detection. The most important and applicable technique in blind sources separation is independent component analysis (ICA). There are a lot of algorithms for ICA and BSS, but in this research fixed point algorithm has been used for system optimization and finally source separation. Moreover, in this study, Pearson distribution has been used for generating score functions to be used in optimization algorithm. Accuracy of this method’s performance has been examined and accepted in lab through several tests and simulation on artificial signals and actual signals collected by sensors installed on a helicopter tail gearbox.