Research On PHM For Jointless Track Circuit Based On Data-Driven

In recent years,the railway has faced increased demands for its operational effectiveness and safety as a result of its importance as a key component of the nation’s infrastructure,the bellwether of economic development,and China’s social development.One of the "three major" outdoor railway signal equipment is the track circuit,which is used to transmit traffic information and check the occupancy and integrity of the rail track.The effectiveness of the track circuit’s work directly affects both the operational efficiency and safety of railway transport.At this stage,the railway bureaus of the National Railways Group have gradually transformed and installed centralised monitoring systems to collect and monitor the status data generated during the operation of the track circuit in real time,but intelligent analysis of the collected data is still lacking.The traditional threshold method is still used in the field for fault discrimination on track circuit,and its data analysis relies on a large number of manual participants,witch is prone to subjectivity and low discrimination efficiency.It is therefore important to consider the introduction of appropriate intelligent analysis methods to extract useful information from track circuit data so that field maintenance personnel can assess the type of fault in a timely and accurate manner and deal with it appropriately.The Prognostics and Health Management(PHM)technology,which is driven by data,analyzes the data gathered by the system and builds a fault prediction model by naturally integrating the correlations between the various devices in the system,effectively resolving the issue of defining complex mathematical models of components and systems.The incorporation of PHM technology into the operation and maintenance of track circuit is both a modern trend and a potent assurance of the efficiency and safety of railway transport in the framework of integrated railway maintenance.This dissertation takes the ZPW-2000 A jointless track circuit as the object of study and focuses on the following:(1)Introducing the structural components of track circuit and monitoring systems,reviewing relevant technical standards,acquiring and expanding data through on-site research and building a test platform for track circuit;researching and collating relevant materials and references,analysing and summarising several common failure modes of track circuit,and providing a theoretical basis for subsequent research.(2)To address the problems of complex fault types and low diagnostic accuracy of track circuit,a fault diagnosis method based on Deep Belief Network(DBN)combined with Least Squares Support Vector Machine(LSSVM)for jointless track circuit is proposed,and the Marine Predators Algorithm(MPA)is introduced to further optimize the penalty factors and kernel function parameters of the classifier model to establish an optimal fault diagnosis model.The DBN-MPA-LSSVM diagnostic model takes full advantage of the layer-by-layer extraction of DBN in the feature extraction process and the high-dimensional pattern recognition of LSSVM in the case of small samples.The experimental results show that the model has an average fault diagnosis accuracy of 97.98% and can achieve timely and effective fault type identification for track circuit.(3)To address the problem of insufficient timeliness of fault diagnosis in the steady-state environment of track circuit,a method is proposed to classify the degradation states of track circuit based on GG(Gath-Geva)fuzzy clustering and extract data degradation features using Convolutional Neural Network(CNN),by constructing an input data set including fault data and severely degraded data,combined with Bi-directional Gated Recurrent Unit(Bi GRU)for track circuit fault prediction.The method combines the Bi GRU with the input data set including fault data and severely degraded data to predict track circuit faults.The degradation process of a track circuit during operation is usually a normal to fault degradation process.A centralised monitoring system is used to obtain the normal operation data of each fault type in the track circuit for a certain period of time before the fault occurs,and the different degradation states are classified using GG fuzzy clustering.The fault data and the degradation data are then jointly input into a CNN-Bi GRU convolutional bi-directional gated recurrent neural network to predict the possible fault categories of the track circuit.(4)Combined with the above research,the App Designer toolbox in MATLAB was used to develop and design the intelligent diagnostic software for track circuit,and the testing and verification of relevant functions such as data import,fault diagnosis,state classification and fault prediction were also completed.