M&R Periods Prediction Model And M&R Decisions Optimization Models For Track Facilities

Tracks are one of the main technical facilities of railways and subways and are the basis of train operation.To keep the track facilities in good condition so that trains can run safely and steadily and to extend the service lives of the facilities as far as possible,currently,periodic maintenance and corrective maintenance are mainly performed on China’s railways.However,with the development of high-speed and heavy-load trains,higher demands are made on the health condition of tracks.At the same time,progressively expanding assets and continuously increasing working hours quickly increase maintenance and renewal(M&R)costs.Therefore,the track maintenance mode is gradually changing from periodic maintenance and corrective maintenance to preventive maintenance.To achieve this transition,track managers need to depict the track degradation rules more accurately,predict M&R periods more precisely,and optimize M&R decisions more reasonably.The prediction of the M&R periods and the optimization of M&R decisions are studied in this dissertation by focusing on the preventive maintenance for track facilities.The individualized prediction model of M&R periods and the optimization model of the lifecycle M&R decisions are developed.The main contents of the dissertation are as follows.(1)An individualized prediction model of M&R periods for track facilities based on the heterogeneity factors discrete-state Weibull distribution(HFDS-Weibull)is proposed.Because of the heterogeneity,uncertainty,multistage nature,and comovement in the track facility degradation process,the linear and continuous track facilities on a railway line are divided into multiple adjacent track segments,and the deterioration process of each segment is divided into multiple degradation stages in this model.Based on this,the model takes each stage of each segment as the object,and the heterogeneity factors that affect their degradation processes are precisely quantified.The model also individually describes the deterioration rule of each stage of each segment using the HFDS-Weibull method,accounts for the effects of heterogeneous factors on the deterioration process,and captures the variations of these effects at different deterioration stages.Using this information,the M&R periods of each track segment at different spatial positions are calculated precisely.The historical inspection data records of rail wear detection from the period of September 2004 to September 2015 for Beijing Metro’s sharply curved rails,together with data on various heterogeneous factors,were used to validate the effectiveness of the model.The results demonstrated that the accuracy of the proposed model enables management personnel to analyze M&R requirements accurately at high space-time resolution and provide adequate decision-making support for track preventive maintenance while considering the effects of the heterogeneity,uncertainty,multistage nature,and comovement in the track facility degradation process.(2)A facility-level lifecycle M&R decision optimization model based on an adaptive-learning Markov decision process(AL-MDP)is proposed.For the optimization of the lifecycle M&R decision for an individual track facility,the model takes each segment as the object based on the division of track segments,and the facility-level MDP state transition probability matrix,which considers the effects of the deterioration heterogeneity and uncertainty,is used to describe the deterioration process.At the same time,by setting up an adaptive-learning mechanism based on the HFDS-Weibull method in the model,the MDP state transition probability matrices can be updated according to the segment’s latest condition state data at each decision moment to achieve the.adaptive learning of the segment’s deterioration process.Finally,to minimize the segment’s lifecycle cost,consisting of inspection costs,M&R costs,and salvage value,with the constraints of the segment condition state,the model uses a Dynamic Programming(DP)based Backwards Induction Value Iteration Algorithm to obtain the optimal lifecycle M&R decisions for each segment.The proposed method was verified with data on the sharply curved rail segment on the down direction of Beijing Subway Line 2 at K11+913.705-K12+084.605.The facility-level AL-MDP model proposed in this dissertation and the traditional facility-level MDP model without adaptive learning are used to optimize M&R decisions for a 10-year planning period based on the state simulation data obtained by a Monte Carlo simulation approach.Comparing and analyzing the optimization results show that,in contrast to the MDP model,the AL-MDP model proposed here can effectively improve the quality of M&R decisions,and it can save M&R costs more effectively while ensuring security.(3)A network-level life-cycle M&R decision optimization model based on AL-MDP is proposed.Because of the optimization of the lifecycle M&R decision for the track facility network,the model takes a network consisting of multiple track segments as the object based on the change process of the network condition state,described by the network-level MDP state transition probability matrix,which considers the effects of uncertainty.At the same time,by setting up an adaptive-learning mechanism based on the Maximum Likelihood Estimation(MLE)method in the model,the MDP state transition probability matrix can be updated according to the latest state data of the network at each decision moment to achieve adaptive learning about the deterioration process.Finally,the model sets 0-1 variables for the M&R actions that can be chosen for each segment of each facility type in the network at each decision moment.To minimize the lifecycle cost of the network(consisting of inspection costs,M&R costs,and salvage value),with the network constraints of the condition of the network,M&R budgets,state transition,and so on,the model uses an Linear Programming(LP)based Mixed-integer Linear Programming Formulation Algorithm to obtain the optimal lifecycle facility-specific policies of the network.The proposed method was verified with the network data for the 156 sharply curved rail segments of Beijing Metro.The network-level AL-MDP model proposed in this dissertation and the traditional network-level MDP model without adaptive learning are used to optimize M&R decisions for a 10-year planning period based on the state simulation data obtained by a Monte Carlo simulation approach.After comparison and analysis of the optimization results,the superiority and practicability of the M&R decisions of the network-level AL-MDP model proposed in this dissertation compared with the MDP model are demonstrated.