Trend analysis for gear pitting fault based on the non-Gaussian characteristic

The gear vibration signal of industrial machinery is usually complex. It contains both Gaussian distribution and non-Gaussian distribution components. When early failure of the gear happens, weak fault information often hides in various complex components, which brings great difficulties for feature extraction and trend analysis. This paper took the measured gear vibration signals as the research objects. By means of time-domain analysis, frequency-domain analysis and bispectral analysis, the gear variation has been researched from normal state to pitting fault state, and then feature extraction and fault trend analysis were made in turn. The results showed that the traditional analysis methods were difficult to analyze the characteristics and trend of pitting fault. However, bispectral analysis method could not only effectively suppress Gaussian noise, but also analyze the nonlinear non-Gaussian changes caused by pitting fault from the standpoint of higher order statistical characteristics. Especially the non-Gaussian eigenvalue based on bispectrum had a high sensitivity and a stable performance to the pitting fault, and was able to obviously show the pitting fault trend. Its effects were far better than time-domain and frequency-domain characteristics, a new reliable feature was provided for the subsequent fault recognition.