Research On The Gear Box Fault State Diagnosis Technology Based On Manifold Learning And LVQ

Gear box occupies an important position in the field of mechanical industry and is widely used in all kinds of mechanical equipment,especially the important transmission parts of rotating machinery.The gear box which can work normally relates to the partial or overall working condition of the whole machine or equipment,and the fault of rotating machinery is caused by a large proportion of the gear box whose components could fail,the vibration and noise signal is the carrier about characteristics of gearbox fault,so the vibration signals of gearbox for accurate diagnosis of the early fault and to predict the development trend of fault have important significance to improve the overall reliability of rotating machinery and to make a decision in which the fault or potential fault components are replaced as quickly as possible.This paper firstly introduces the related concepts of the main failure forms,fault characteristics and vibration mechanism of the gear box.According to the traits of the cross modulation components of vibration signals of the gear emerging defects,the feature selection and extraction method which rely on the wavelet denoising,EEMD and ISOMAP concerning high dimensional characteristics of fault samples is presented,a gear fault diagnosis model which rely upon the ISOMAP and LVQ is established on this basis;According to the traits of the multicomponent FM phenomenon which is caused by the bearing emerging different fault at the center of the low-order to the high-order of all natural frequencies existing in the outer ring and the vibrational energy exhibiting decay,the feature selection and extraction method which rely on the wavelet denoising,LMD and distance adaptive LLE concerning high dimensional characteristics of fault samples and the important parameter-K selection of the LLE algorithm based on the concept of discreteness are presented,a bearing fault diagnosis model which rely upon the distance adaptive LLE and LVQ is established on this basis.The experimental results show that the methods which rely on the wavelet denoising,EEMD(or LMD)and manifold learning can effectively reduce the noise's serious interference for the low dimensional topology information of manifold learning.The experimental results of gear fault simulation also show that the diagnostic accuracy of the model based on the ISOMAP and LVQ is higher than the model based on the BP which is widely used in the field of fault diagnosis;And on the comparisons of the PCA,LLE and distance adaptive LLE,the experimental data of bearing fault simulation verifiy the feature separability based on dimension reduction with the distance adaptive LLE whose important parameter-K selection depends on the concept of discreteness,and the diagnostic accuracy of the model based on the distance adaptive LLE and LVQ is higher than the model based on SVM.