Research On Rolling Bearing Fault Diagnosis Based On Data-driven

Rolling bearing is one of the key components of rotating machinery equipment,bearing faults can seriously affect the safe and stable operation of rotating machinery.Research on Rolling bearing fault diagnosis technology has important practical significance and engineering value for safeguarding equipment safety and stable operation,reducing the occurrence of safety accidents and reducing the cost of equipment maintenance.The rolling bearing vibration signal can provide the equipment status characteristic information in time and accurately,and can be monitored permanently or intermittently,so it is widely used in rolling bearing fault diagnosis.Bearing vibration signals have non-linearity and non-stationarity,and time-frequency analysis method is an effective method to analyze such signals.But the original feature set obtained by signal processing and feature extraction based on time-frequency analysis method will have interference and redundant information,it is necessary to extract sensitive features that are conducive to the recognition of fault conditions from the original feature set;for the high-dimensional feature set,it is necessary to reduce the dimension and obtain a low-dimensional feature set with better discrimination performance.In response to the above problems,according to the following four main steps in the process of rolling bearing fault diagnosis based on data-driven: signal processing,feature extraction,feature reduction,and pattern recognition,the following related research work is carried out in this paper:(1)The vibration signal processing method based on maximum overlap discrete wavelet package transform(MODWPT)is studied.MODWPT is used to decompose the vibration signal,the single-point reconstruction of the obtained terminal node is performed,and the reconstructed signals in different frequency ranges are obtained.The statistical parameters of the reconstructed signals and their Hilbert envelope spectrum are calculated and the original feature set is constructed.(2)Fault sensitive feature selection method based on ReliefF and mean deviation(FSRMD)is proposed.FSRMD uses the ReliefF algorithm to obtain feature weight value to represent the class discriminative degree of the feature,and calculates the mean deviation of statistical feature sample to represent the intra-class cohesion degree of the feature.The ratio of the feature weight value to the mean deviation is used as the evaluation index of the fault status sensitivity of the statistical feature,and the quantitative analysis of the fault status sensitivity of the statistical feature is realized.(3)A dimensionality reduction method for improving neighborhood preserving embedding(NPE)is proposed,supervised neighborhood preserving embedding based on maximum margin criterion(SNPEM).SNPEM combines the maximum between-class boundary obtained by maximum margin criterion(MMC)and the preservation of the the local manifold structure of data obtained by NPE,which achieves that the local manifold structure of data is preserved,while the boundary distance between different class samples is maximized and the discriminative performance of low-dimensional space samples is improved.(4)Fault diagnosis models are constructed based on support vector machine and K-nearest neighbor algorithm,and the bearing fault data of Case Western Reserve University and the bearing fault data of SQI-MFS test-bed are used for the experimental verification.The experiment sets two test cases under the same working condition and the different working condition respectively,and is used to verify the effectiveness and adaptability of the proposed method.The experiment is divided into two groups.One group does not introduce sensitive feature selection method,which is performed to test the effect of different dimensionality reduction methods on fault diagnosis;the other group uses sensitive feature selection method and different dimensionality reduction methods in the fault diagnosis model to conduct a series of comparative experiments.The experimental results show that,in this paper,the proposed fault sensitive feature selection method FSRMD can effectively select the sensitive features;feature reduction method SNPEM can improve the discriminative performance of feature set after dimensionality reduction,whihc are conducive to fault pattern recognition.FSRMD and SNPEM can improve fault diagnosis accuracy and have good adaptability.When the suitable number of sensitive features is selected,the fault diagnosis model using FSRMD and SNPEM can attain the best fault diagnosis performance under the same working conditions as well as under different working conditions.