Degradation State Identification And Trend Prediction Of Switch Machine Based On Combined Cluster Analysis

With the development of China’s railway industry entering the new normal,under the traction of the era demand of rapid railway development and the maintenance and guarantee demand of railway machinery and equipment,relevant railway equipment will also face new challenges,and its future development direction will further integrate with information technology and move towards intelligence with the development of science and technology.As an important part of railway safety operation equipment,signal equipment is one of the main maintenance equipment of the joint Department of industry and electricity.The maintenance ability of signal equipment determines the level of train operation to a very great extent.With the increasing transportation tasks of railway,electro-hydraulic switch machine gradually replaces electric switch machine.Therefore,higher requirements are put forward for the application quality of electro-hydraulic switch machine.As a key component of switch equipment,the failure rate of switch machine is very high because it works outdoors for a long time.However,in the process of using the switch machine,the performance of the switch machine is not only in the two states of fault and normal,but will gradually experience a series of different degradation,from intact to complete failure.Once the switch machine is degraded,its operation reliability will be sharply reduced.In serious cases,the switch will not work normally,which will bring certain potential safety hazards to the safe and normal operation of the train.In order to ensure the safe and efficient operation of the switch machine equipment,it is of great practical significance to carry out the research on the identification of the performance degradation state of the switch machine,predict the development trend of the degraded performance of the switch machine,and carry out targeted and timely maintenance.The signal generated in the switching process of switch machine is generally non-stationary and nonlinear.And the characteristics of various degradation States experienced in the degradation process are less different,which leads to the problems of low recognition accuracy and inaccurate prediction state of switch machine degradation state.Therefore,taking ZYJ7 switch machine as the research object,this paper studies a method that can accurately identify the degradation state of switch machine and accurately predict the development trend of degradation performance state of switch machine.The main research contents are as follows:(1)Degenerate feature selection.Aiming at the small difference of characteristics in different degradation states of switch machines,a multi domain feature extraction method based on Kernel Principal Component Analysis(KPCA)is proposed.Firstly,preprocess the power curve data of the switch machine;Then,by synthesizing the extraction indexes of time domain,frequency domain and time-frequency domain,the high-dimensional feature vector reflecting the degradation state of switch machine from multiple angles is obtained;KPCA is used for feature reduction and fusion,and the degraded feature space of switch machine is constructed.(2)Degradation state identification.In order to improve the recognition effect of the degradation state of the switch machine,by analyzing the data set of the daily work of the switch machine collected from the microcomputer monitoring system,the degradation state of the degradation characteristic data of the switch machine is analyzed by using the combined clustering method based on density propagation based adaptive multi density clustering algorithm(DPAM)and Gath-Geva clustering(GG),Combining the advantages of partition clustering and density clustering,different state modes of switch machine under performance degradation are excavated,which reduces the heavy work caused by manual analysis on site and improves the work efficiency of signal equipment maintenance.The actual power curve data of a road bureau is used to verify the effectiveness of the combined clustering algorithm.From the final clustering effect,the selection of the number of clusters by the combined clustering algorithm is more scientific and does not need to be set manually,which greatly reduces the time;Moreover,the Classification Coefficient(CC)is the highest and the Average Fuzzy Entropy(AFE)is the smallest,and the aggregation of the same operating condition of the switch machine is closer;From the recognition results,compared with the corresponding single clustering model,the accuracy of the combined clustering recognition model is the best,which is 96.33%.(3)Prediction of development trend of degraded performance state.The key to prediction is the accuracy and stability of prediction.In order to improve the prediction effect of the development trend of the degradation state of the switch machine,the classical small sample machine learning algorithm Support Vector Regression(SVR)is used for prediction and analysis.SVR algorithm has a large number of kernel functions,which can be flexibly used to solve various nonlinear regression problems.It has been widely used in the fields of target recognition,detection,prediction and analysis.However,the penalty factor,insensitivity coefficient and kernel width of SVR algorithm have a great impact on its accuracy and stability.Therefore,this paper introduces Grey Wolf Optimizer(GWO)to solve the problem of joint optimization of penalty factor,insensitivity coefficient and kernel function width when constructing SVR prediction model.Compare the prediction results of SVR,BP(Back Propagation),ELM(Extreme Learning Machine),HGWO-SVR(Improved GWO based on differential evolution to optimize SVR algorithm)prediction models and the GWO-SVR prediction model established in this study,The prediction performance of the first two principal component degradation performance indexes obtained by KPCA fusion is the best.The Absolute Percentage Error(MAPE)of the prediction model is 0.5571 and the Root Mean Square Error(RMSE)is 0.0046,which has high prediction accuracy.To sum up,combined with the historical data of the power curve of the switch machine in the microcomputer monitoring system of a railway bureau,this paper uses the DPAM-GG combined clustering model to establish the degradation state identification model of the switch machine,and uses the GWO-SVR combined prediction model to predict the degradation trend of the switch machine,so as to provide theoretical and technical support for the on-site maintenance strategy of the switch machine.