Research On Fault Diagnosis Of Rotating Machinery Based On Feature Dimension Reduction

With the deepening and popularization of rotating machinery in the industrial system,their reliability and health are related to the development of industrial safety and economy.The arrival of intelligent manufacturing also makes the mechanical equipment more accurate,automatic and complex,that brings a bulky pressure to the repair and diagnosis of its parts.In this context,the signal processing technology and dimension reduction technology are combined.The gear and bearing of the rotating machinery are used as the experimental carrier to establish a diagnosis framework of the rotating machinery with the feature dimension reduction as the core.Through the signal processing technology to solve the problem of nonlinear and non-stationary fault signals,a more complete failure information is obtained.The problem of feature space dimension is solved by manifold learning nonlinear dimension reduction technology,that bulky ease the difficulty of fault identification.The feature dimension reduction technology improves the intelligence and accuracy of the diagnosis results.The specific research contents are as follows:(1)The overall panel work and research value of fault diagnosis technology for mechanism equipment are clarified,and The fault diagnosis ways of rotating machinery based on characteristic dimension reduction are round up.It mainly discusses from the two aspects of feature processing technology and dimension reduction technology.(2)Based on the basic theory and related concepts of manifold learning,two nonlnear dimensions reduction algorithms are expound: the improved LLE and t-SNE algorithm.The results and performance of these two methods are verified by dimensioning the "Swiss volume" data and Fisher data.At the same time,the conception of clustering analy classifier is introduced,and the K-means and AP clustering methods are briefly explained.The advantages of the two classifiers are pointed out.The performance of the two methods is verified by using Gaussian random distribution number and Fisher data,which provides a theoretical support for the subsequent fault result analysis.(3)The high dimensional nonlinearity is aimed at the non-stationary of rotating machinery equipment information and fault characteristics.This paper studies an intelligence diagnosis method based on the VMD and improved LLE ways.The theory of variable mode decomposition(VMD)is expounded and its performance is verified.The initial signal decomposition of bearing data of Xishou University and gear data collected by our laboratory is carried out by VMD technology.The features of frequency domain,time domain and time-frequency domain are extracted to form a multi domain high-dimensional feature space.Then,the improved LLE method is used to reduce the dimension of features,and the low dimensional data is obtained for fault identification,which data will inputing the AP cluster techniques.The advantages of the this method are shown by comparing different methods.(4)In order to realize the higher accuracy of fault identification,this paper studies a new method based on adaptive feature selection and t-distribution random neighborhood embedding.The paper briefly expounds the wavelet packet(WPT)technology and the basic principle of adaptive feature selection.The original fault signal is resolved by WPT,and the features of frequency domain,time domain and time-frequency domain are extracted at the same time.The feature samples are selected by adaptive feature selection technology,and the sensitive features are selected and processed by t-sne.Finally,the data after dimension reduction is input into AP cluster.The method is verified by using the gear and bearing data collected by MFS mg and wtds,and the high fault discrimination rate is obtained.