Gear fault feature extraction based on the combination of Gabor reconstruction and bispectral analysis

As weak failures of gearboxes increase non-Gaussian characteristic of vibration signals, bispectral analysis can extract the fault features. However vibration signals of gearboxes usually contain strong periodic components. It causes great difficulties for fault feature extraction. This paper adopted the method based on Gabor reconstruction and bispectral analysis. At first, Gabor transformations were made for the vibration signals. In the next place filtered the strong periodic components in time-frequency domain, reconstructed the signals and then made bispectral analysis. After that, the effect of periodic high-energy non-fault factors had been eliminated. At last the effect of fault feature extraction became more obvious. Tested and verified the above method by use of gear´s practical measured vibration signals. Comparative analysis of the high order statistic difference in normal condition and in pitting failure condition had been made finally. The results showed that, bispectral analysis of reconstructing signals with Gabor filtering not only inhibited the Gaussian noise, but also eliminated the effect of non-fault factors. The extracted feature of non-Gaussian intensity based on above method was more sensitive to weak failures and helpful to the follow-up fault diagnosis.