Research On Remaining Useful Life Prediction Method Of Point Switch Machine Based On Data-Driven

As a mechatronics equipment to realize the conversion and representation of railway turnouts,the point switch machine has many equipment and a high frequency of action in railway engineering,and its operation quality directly affects railway transportation efficiency and train driving safety.The fault rate of the point switch machine in the signal system is high,but there is no effective theoretical technology applied to the health state management of the point switch machine,and the operation and management of the point switch machine mainly relies on the staff to maintain and repair them according to the planned cycle or after the equipment failure.This operation and maintenance method has problems such as low maintenance efficiency,strong technical dependence on operators,and waste of materials.Therefore,it is necessary to carry out condition-based maintenance on the real-time operational status of the point switch machine in the high-speed development environment of railway transportation.Implementing condition-based maintenance requires proper detection of the cause of failure of faulty equipment,accurate assessment of the operating state of equipment that does not fail,and accurate prediction of the remaining useful life(RUL)of the equipment.In this thesis,the real-time monitoring and remaining life prediction methods of data-driven switching machines are studied and verified by using fault prediction and health management technology,combined with the actual application of switch machines in railway sites.The main work of the thesis is summarized as follows:Aiming at the problem that the characteristic information is not obvious when the overall analysis of the switch machine system is carried out by using a single power data,the threephase current data is considered and organically integrated with the power curve to enrich the information and improve the prediction accuracy.Based on this curve feature,the starting and ending parameters of the current curve in the conversion stage during the switch operation process are determined by calculating the mode of any phase current curve in the three-phase current curve,and the three stages of starting,conversion,and representation are more accurately divided.Based stage division,the Frecher distance between the three different stages between the switch machine action current curve and the power curve,and the reference curve is calculated respectively,and the operating state of each stage is evaluated sequentially as a health status evaluation index so that the fault causes of the switch machine equipment can be better analyzed and more targeted maintenance strategies can be formulated.Aiming at the problem that the amount of fault sample data collected by the switch machine at the maintenance base layer is small,the support vector machine(SVM)model is selected to analyze and detect the fault cause of the faulty equipment.Based on Multisim,the simulation model of the active circuit is established,the model and data are combined,the results of SVM diagnosis are used as the input of the circuit simulation model,the electrical parameters output by the model are compared and analyzed with the actual parameters,and the cause of the fault of the equipment is analyzed in reverse,to avoid the deficiency that a single data-driven method cannot identify the fault point of the circuit,refine the fault type,and enhance the interpretability of the detection results.To explore the deep features of the monitoring information of the switch equipment and quantify the uncertainty of the prediction results,an RUL prediction model of the switch machine based on multi-feature information fusion is proposed.The method includes the construction of health indicators combining a deep belief network(DBN)and local linear embedding(LLE)and RUL prediction based on a hidden semi Markov model(HSMM).Firstly,the unsupervised training DBN is used to extract the power curve and current curve features of the switch machine,and then the features with good tendency are screened as the input of LLE,and the comprehensive health indicators are constructed.Finally,the HSMM model is trained by using health indicators to realize the recognition of the degradation state and the prediction of the remaining life of the equipment.The effectiveness of the proposed method is verified by using railway field data,which provides a theoretical reference for the maintenance of the switch machine.