Extraction of weak transient signals based on adaptive window merging for rolling bearing fault diagnosis

The problem of local and weak defect detection in rotating machinery has been widely studied in many literatures. Its vibration signal is not easy to be recognized in the raw signal owing to its low energy level at the early stage of the fault and strong background noise. An early developed and still widely used technique for such a signal detection is envelope analysis. Despite the fruitful modifications and improvements on this technique, its effectiveness and application is still limited by its parameter selection, which is generally understood as the optimization on the structure and the parameters of the filter that are hardly ever known a priori. To solve this problem, this paper focuses on the adaptive selection of the filter parameters, i.e. the center frequency and the band width. By analyzing the energy distribution in frequency domain of the vibration signal, the center frequency is automatically determined by identifying the frequency with the maximal energy; and then, by comparing the energy changes of two adjacent windows and merging these windows if the changes are small enough, the appropriate band width is selected according to the window with the concentrated energy. An experimental vibration signal collected from a bearing with an outer race defect is used to verify the effectiveness of the proposed filter. A comparison between the proposed method and the fast kurtogram is also completed. The results indicate that the proposed filter can quickly identify the resonance frequency band induced by the faulty bearing and then extract weak transient signal for accurate fault diagnosis.