Research On Classification Method Of Rotating Machinery Fault Data Based On Semi-supervised Dimensionality Reduction

Process monitoring and fault diagnosis analysis based on equipment can effectively control the development of faults and ensure the safe and stable operation of mechanical equipment.The use of equipment monitoring data for fault diagnosis is a common method,but due to the complex structure and operating environment of the rotating machine,the vibration signal often contains strong background noise and interference signals.Therefore,how to obtain effective information reflecting the fault becomes The key link.Data dimensionality reduction can eliminate redundant and irrelevant information in the original data and find valid data that reflects the nature of the fault.In practical applications,the fault feature set based on the multi-domain structure of the vibration signal tends to be too high in dimension and the category mark is insufficient.Using the traditional dimensionality reduction method,the ideal dimensionality reduction effect and classification intensive reading may not be obtained.In order to make full use of the fault information in the existing unmarked data,to solve the problem of supervised dimensionality reduction,the unsupervised learning model is inaccurate and to obtain higher diagnostic fault accuracy.This paper studied the fault Data set classification based on semi-supervised dimensionality reduction.The main work of this paper is as follows:(1)Aiming at the problem of low recognition rate of fault type and fewer marked fault samples of high-dimensional data sets that exhibit nonlinearities,a fault diagnosis method based on Kernerl Semi-supervised Local Fisher Discriminant Analysis(KSELF)data dimensionality reduction is proposed.The method first maps the original fault data set into the high-dimensional feature space through the RBF-type kernel function,and uses the SELF algorithm to calculate the optimal projection transformation matrix.Then based on the KNN classifier,the low-dimensional features obtained by dimension reduction are trained and learned,and the type of fault is identified.The proposed KSELF method avoids the over-learning problem caused by the insufficient label sample in the dimension reduction process by fully utilizing the information in the partial class label sample and a large number of unlabeled samples,effectively capturing the nonlinearity information in the data.Finally,the effectiveness of the proposed method is verified by a fault simulation experiment based on a double-span rotor test bench.(2)In the process of dimensionality reduction of high-dimensional nonlinear data,in order to maintain the global and local structure of the data and the prior information of the sample class label,a fault diagnosis method based on KPCA-SSLPP data dimensionality reduction is proposed.Firstly,based on the time domain,frequency domain and time-frequency domain analysis,the corresponding fault features are extracted from the original vibration signals to form a high-dimensional feature set.Then KPCA is applied to the high-dimensional feature set to obtain the main features,which reduces the correlation between features the correlation and maximizes maintains the global nonlinear structure in the feature set.Finally,the low-dimensional local essential features are mined using the SSLPP algorithm,and the obtained low-dimensional feature vector is input into the LSSVM for fault type identification.The effectiveness of the proposed method is verified by fault simulation experiment of the centrifugal pump test bench.The experimental results show that the proposed method has better dimensionality reduction effect in experiments than other dimensionality reduction methods,and can achieve higher diagnostic accuracy under a small number of labeled samples.(3)Based on the Lab VIEW platform,a vibration test system for the centrifugal pump test bench was designed.The system realizes the monitoring of the running state of the centrifugal pump,the data acquisition,processing and storage as well as the graph display of the time domain,frequency domain and axis trajectory.The fault classification is realized by embedding the dimensionality reduction algorithm in the system.