Bearing fault diagnosis based on recurrence plots and local non-negative matrix factorizatio

In view of non-stationary characteristics of bearing vibration signal and difficulty of extracting fault parameters in reality, a bearing fault diagnosis based on recurrence plots and local non-negative matrix factorization was proposed. Firstly, recurrence plots of the collected bearing vibration signal is analyzed and gray scale is generated. Then, characteristic parameters of the recurrence plots are extracted by the local non-negative matrix decomposition to obtain coefficient coding matrix. Finally, classifier is used for pattern recognition of coding matrix, so as to achieve automatic diagnosis of bearing failure. The method is applied to four kinds of typical bearing fault diagnosis cases, and the application results show that the method can calculate characteristic parameters adaptively for recurrence plots of different operating conditions and avoid influence of human factor on accuracy rate of diagnosis with better adaptivity and robustness.