The Research Of Gear Fault Diagnosis Based On Image Recognition

Gear is one of the most important parts in the mechanical system, its running state is good or not will directly determine the health of the whole mechanical system, the most unfavorable situation is may result in the whole production line stop, cause huge economic losses. So once the gear malfunctioned, discovered in time and eliminate of these faults is particularly important. In the past,the gear fault diagnosis is accomplished by analyzing the spectrum characteristics of vibration signals, this paper puts forward a kind of gear fault diagnosis method based on image recognition, the method combined with the advanced image recognition technology and the mature signal spectrum analysis technology, through analysis of gear fault signal bispectrum image texture feature, and then use the SVM classification algorithm to identify the gear fault types. Finally, experimented on the wind turbine power transmission fault diagnosis(WTDS) experiment platform, and validated the feasibility of this method. The main research content is as follows:1. Multi-resolution decomposition technique based on wavelet packet denoising ability has been widely recognized, in addition, the bispectrum analysis in processing with non-gaussian signal has a good advantage, combining these two points of wavelet packet and bispectrum analysis method has excellent ability to filter out the noise. Filter the background noise by the wavelet packet and bispectrum analysis method, and obtaint the stable texture feature of bispectrum and wavelet packet image.2. Fusion of different channel bispectrum image can Integrate fault information of different channel, it will be more conducive to gear fault diagnosis. This paper adopt the image fusion technology based on wavelet transforms, the multi-channel fault information was fused on the iamge level.3. Extract GLCM matrixs on the four direction of an image with wavelet packet bispectrum analysis, then using the fusion method based on weighted average to fuse these GLCM matrixs on the four directions, at last, generate feature vector with 12 second order statistics of the GLCM matrix.4. Support vector machine(SVM) is one of the best algorithms to solve small sample learning, and has good generalization ability.With the eigenvector which is generated by GLCM matrix, by the multi-classification SVM classification algorithm has achieved the gear fault identification on the iamge level.5. Through the experiments on WTDS experiment platform validated the feasibility of Gear fault diagnosis which is based on Image recognition technology. Experiments prove that under the polynomial kernel function, the recognition rate of this method can reach about 85%.