Research On Turnout Fault Diagnosis Based On Action Curve Compression

Turnout,whose state is closely related to traffic efficiency and safety,plays an important role in the system of railway signal.In pace with the development of urban rail transit information construction and the widespread application of sensor technology,the mass of the turnout monitoring data is generated with numerous sensors in a short period of time.More problems,including large storage space,difficult data-processing and redundant data information,are emerging in the wake of it.Facing the problems arising from such large-scale turnout monitoring data,it is of great necessity and significance in engineering practice that more attention is paid to the study of its compression and rapid diagnosis.The turnout monitoring data has the characteristics of diversity and imbalance,and the purpose that the storage space is saved and the breakdown are rapidly positioned can be accomplished by compressing data.As a result,this thesis has conducted a more in-depth study on how to adaptively adjust the turnout action curve and compress its monitoring data,as well as how to quickly diagnose its breakdowns.The major findings of this thesis are as follows:(1)By analyzing the characteristics of the turnout action curve and combining many algorithms that includes histogram threshold,Boxcar and Max-Min,an adaptive segmentation method based on the turnout action curve has been provided and researched,which can better preserve the original characteristics of the curve.In addition,a method that compresses and scales the turnout action curve has been designed,based on Cubic and SDT algorithm that are adjusted to a new threshold.It has been founded that the adaptive segmentation method has positive effects on different types of curve segmentation.What’s more,compared with other compression algorithms,the compression algorithm that are proposed in this study has more advantages of higher efficiency and better accuracy.(2)Facing the various turnout curve data and ensuring that this compression method can be universally applied for these data,its compressing effects of plentiful turnout curve data have been comprehensively analyzed from the perspective of subjective visions and objectively quantitative indicators in this thesis.In the meantime,a comprehensive evaluation criterion that compresses multiple objectives in the turnout action curve has been figured out,which can provide comprehensive evaluation on the quality of the compression algorithm from many aspects,such as subjective scoring,compression efficiency,and reconstruction error.(3)The research on how to rapidly diagnose the turnout breakdowns has also been carried out in this thesis.First of all,a large amount of data has been analyzed and nine typical failure modes have been extracted based on experts’ experience and data visualization.Secondly,a DTW-SMOTE weighting strategy has been designed for dealing with the imbalance monitoring data.Finally,the rapid fault diagnosis model which is on the basis of the LSTM-FCN,has been established.Among them,FCN(full convolution network)model has been used to extract the local features of curve,and LSTM(Long Short-Term Memory network)has been used to learn timing characteristics.As is shown in the experimental data that inputting compressive data of the same dimension has made the model has fewer parameters,shorter training time,faster convergence,and higher diagnostic efficiency.In the end,the research contents and future research are summarized.