Feature extraction based on bispectral entropy for gear weak fault

Researched on the vibration signals of the gear test-bed. According to the characteristic of the signals non-Gaussian changes caused by gear weak fault, used bispectral entropy to quantitatively describe the distribution of non-Gaussian components in bifrequency domain. Finally extracted fault information based on bispectral entropy, and the feature trend of crack expansion period could be obtained. Results show that bispectral entropy is less influenced by the non-fault factors and not based on the energy information. Bispectral entropy can inhibit Gaussian noise efficiently, meanwhile it is very sensitive to weak fault. So bispectral entropy provides a new effective method for follow-up fault diagnosis and trend prediction.