Research On Fault Prediction Of Switch Machine Based On Hidden Markov Model

Turnout is the key equipment of railway signal system.With the continuous expansion of China’s railway network and the continuous acceleration of train operation speed,whether the turnout can work normally directly affects the train operation efficiency.More importantly,once the turnout fails,it may cause derailment accidents and cause casualties.Therefore,it is necessary to evaluate the health status of the turnout,so as to take timely measures to avoid the occurrence of failures.But at present,the fault prediction for turnouts is mainly carried out by maintenance personnel to regularly check and repair the turnouts,the maintenance workload is large,and there is a hidden danger of "over-repair".In view of the above problems,this thesis proposes a prediction method of turnout failure and remaining life based on HMM model.The main contents are as follows:(1)According to the feature extraction requirements of degradation state recognition,a multi-faceted feature space construction method is proposed.The features of the original power curve are extracted from the time domain,value domain,wavelet decomposition,EMD decomposition and information entropy angle.(2)Use the four degradation indicators of monotony,trend,robustness and predictability to evaluate and select the original degradation features.In view of the problem that the features still have redundancy after selection,t-SNE method is used to reduce the dimension of the selected features.(3)Aiming at the complex problem of the degradation state segmentation process,combined with the t-SNE dimensionality reduction feature,a method for automatic degradation state segmentation based on spectral clustering is proposed,and combined with the CH index and the distribution of the degradation state on the time axis,it is automatically determined Number of degraded states.(4)A CHMM-based turnout fault prediction model is proposed.According to the process of degradation feature extraction,turnout fault data is extracted,selected and dimension reduced,and a CHMM-based turnout fault diagnosis model is constructed.Combined with the severely degraded state of the degraded state recognition result,the turnout fault category is predicted.The prediction model was verified using ZYJ7 electro-hydraulic switch machine.(5)In view of the problem that the traditional HMM model does not have the display state residence time,the CHSMM model is applied to the prediction of the remaining life of the turnout,and the remaining life interval of each degradation state is calculated to provide a reference for the maintenance and repair plan of the turnout.