| In recent years,with the development of the rail transit in large and medium-sized cities across the country,the total mileage of rail transit operations and the total number of vehicles have been rising,at the same time,the complexity and integration of train vehicles have been continuously improved.This puts higher demands on the safety of railway transportation equipment and the efficiency of maintenance.However,the current train maintenance and repair methods cannot fully follow the pace of rail transit development.The current maintenance method not only consumes human resources,but also causes equipment failures due to untimely maintenance and other reasons,resulting in economic losses and even jeopardizing passenger safety.Therefore,it is of great significance to study the reliable prediction and accurate maintenance of faults in advance by using historical Train-State-Information to protect passenger safety,reduce economic losses and reduce the workload of maintenance personnel.Based on the 440 series Train-State-Information of Deutsche Bahn,this thesis proposes a Hidden Markov Model and combines the Condition-Based-Maintenance(CBM)and Prognostics and Health Management(PHM)technical characteristics to establish the train state prediction and fault maintenance model.According to the historical Train-State-Information,the model uses the machine-learning algorithm to predict and obtain the health rating of the equipment at the next moment through the likelihood transformation.The maintenance personnel pre-process the train equipment with reference to the predicted health rating,and the health rating generated by the actual operation after the pre-processing is fed back to the prediction module.The thesis establishes a two-level prediction mode for train state prediction: the first-level prediction uses the time-delay neural network for the whole train equipment states;for the complex characteristics of the train equipment state and the state of the train equipment with poor prediction effect,the second-level prediction uses different neural networks to predict the different train equipment states.It is difficult to predict the abnormal state of the train(small probability event),and the small probability event has obvious heterogeneity.The combined neural network can solve this problem better.From the perspective of the site operator,this thesis analyzes the maintenance interval and maintenance cost of the train equipment fault prediction and maintenance method.Finally,the thesis extends the fault prediction and maintenance method based on Train-State-Information to similar railway equipments,and combines with the regular maintenance and repair methods to initially design the railway equipment maintenance management system.The system can take advantage of rail transit big data information to make equipment maintenance work more scientific.Based on the analysis of Train-State-Information the thesis establishes a method for fault prediction and maintenance of train epuipment.For the safety requirements of railway equipment,after numerical simulation analysis,the two-stage prediction mode combined neural network has higher accuracy than the single neural network,and the computational power requirement is not significantly improved.Through qualitative and quantitative analysis,train equipment fault prediction and maintenance method have advantages over traditional planned maintenance methods in terms of maintenance intervals and maintenance costs.At this stage,train equipment fault prediction and maintenance method can be used as a supplement to the current traditional maintenance strategy,which can reduce maintenance costs,and reduce the number of failures in the actual operation of the train. |
