Frequency Band Selection Based on a New Indicator: Accuracy Rate
Local defects in rotating machinery give rise to periodic impulses in vibrations. In order to acquire the information of these faults, various diagnostic methods have been proposed in the past decades. Most methods used the squared envelope spectrum (i.e., the spectrum of the squared envelope) as the final diagnostic tool, but different preprocessing steps were used before obtaining the envelope signal. The key problem is to obtain the center frequency and bandwidth of the fault signal, then analyze the envelope (squared envelope spectrum) of the band-pass filtered signal. The framework of accuracy rate method was proposed by means of cross validation of the nearest neighbor classifier in this paper: a) obtain the piecewise signal through original signal segmentation; b) calculate the feature of each piecewise signal; c) then an accuracy rate is calculated based on cross validation of the nearest neighbor classifier; and d) repeat the above steps in different frequency band, then find a frequency band with the maximum accuracy rate. Through this algorithm, we can obtain a fault frequency band, and then we can find out the type of the fault by the spectrum of the squared envelope. At the end of this paper, the proposed method is validated by two examples and compared with the other two diagnostic methods: conventional envelope analysis and Fast kurtogram. Through the comparison of results, the validity and superiority of this method has been proved.