Abstract:

This paper deals with the fault pattern recognition of gears based on energy loss. In the investigation,first,a modified LLE ( Locally Linear Embedding) algorithm,which combines the supervised learning with the local principal component analysis,is proposed to effectively extract the low-dimension manifold structure and the classification feature of data. Then,the energy loss of gear tribological system and its fault pattern recognition method are analyzed. Finally,by taking the test rig for energy loss monitoring of gear box as an example,the variations of input power loss under different kinds of gear faults are analyzed,the dimensionality reduction and pattern recognition are performed by using the modified LLE algorithm,and the classification performance of the algorithm is evaluated according to the recognition rate of the multi-class support vector machine. The results show that the modified LLE algorithm is of high recognition rate and is effective in fault pattern recognition of gear energy loss.

Key words: fault diagnosis, multi-fault classification, locally linear embedding, manifold learning, energy loss, pattern recognition