Grid-based Decision-making Models For Railway Track Health Management

Railway track must be firm and stable, and has the correct geometry, to guarantee the operational safety and passenger comfort in trains. Due to the higher speed and heavier load of train, higher demands on the safety, stability and reliability of the track are made.So, the railway maintenance mode is gradually changing to the "Strict Inspection and Careful Maintenance", "Precise Maintenance", "Accurate Maintenance", "Preventive Maintenance","Centralized Maintenance",etc. Grasping the track health more comprehensively, and depicting the track degradation rules more accurately are important academic problems in railway infrastructure management field, and practical problems that need to be solved urgently in the railway management.Based on the research findings of many domestic and foreign experts, the following four parts are studied, using spatio-temporal analysis and Big Data.(1) A railway infrastructure gridding management theory is proposed. It is adapted to the features and degradation characteristics of railway infrastructure. Its core idea is that a continuous railway line is divided into adjacent "segments" of the same specific length. Based on the spatial location, a full life-cycle data integration is realized, and the model for condition variation rules for segments is customized. Moreover, railway maintenance activities are organized precisely, accurately, and efficiently.(2) A track-grid-based health evaluation method is proposed. Based on the railway infrastructure gridding management theory and Big Data, a condition evaluation index system and a health index are established for track grids (or components), to perfect the existing condition evaluation index system. On this basis, a health evaluation framework for railway track grids and components is developed. Employing PCA model, the condition indicators (or variables) of track grids (or components) are transformed into a new small set of condition variables, and the relevance among the multivariate data measuring track grids (or components) condition is reduced. Based on new condition variables, all of the possible health characteristics for track grids (or components) are identified, using the hybrid hierarchical k-means clustering model. And the range of the health indexes for track grids (or components) is defined. The tree-augmented naive Bayes classifier is employed to identify the possible underlying relationship between various condition indicators and a health index, and the health index for track grids (or components) is obtained directly. This method allows linear and continuous track grids(or components) health to be grasped in smaller ranges of space. The proposed method was verified with the condition data accumulated in March 2016, ranging from 548 km to 985.6 km in the up and down direction along the Lanzhou-Xinjiang Railway. Our evaluation is demonstrated that the proposed method outperforms the traditional method in China's railway management, and the success rate for the health index is 90.8%.(3) A bias correction model for railway track shaking is proposed. Track shaking measurements are the main data source of track grids (or components) health evaluation,coming from track geometry cars, track geometry trolleys, portable measuring instruments, etc. However, there are milepost bias and measurement bias in track shaking measurements. Owing to memorability of track grids (or components) degradation, a filtering model for railway track shaking measurements is developed, based on the railway infrastructure gridding management theory and Big Data. Using this model, a lot of the measurements data of the same inspection method at different inspection time are mined deeply. And the impact of milepost bias and measurement bias is reduced.Moreover, a defects type estimation model for railway track shaking measurements and a defects level estimation model are developed. Using this model, a lot of the measurements data of the different inspection method for long periods of time are mined deeply. And the reasons for track shaking are effectively diagnosed. This model reduces the workload of the railway inspectors. Track shaking raw measurements were accumulated from June 2016 to September 2016, ranging between 548 km to 985.5 km in the down direction along the Lanzhou-Xinjiang Railway to verify the proposed model.Track geometry defects data at the same spatial location were also accumulated from October 2015 to September 2016. Our evaluation showed that the success rate of filtering is 89.8%, and the success rate of diagnosis is 70.4%.(4) A track-grid-based condition prediction model is proposed. Based on the railway infrastructure gridding management theory, track grids (or components) are the objects of research. This model customizes degradation rules for each track grids (or components)to predict its short-term and long-term condition. This model allows linear and continuous track grids (or components) degradation rules to be grasped in smaller ranges of space.The condition rank of track grids (or components) is predicted between two adjacent maintenance activities after an inspection cycle, using the tree-augmented naive bayes model for railway track grids and components condition prediction. Due to heterogeneity,memorability, etc. of track grids (or components) degradation, the tree-augmented naive Bayes classifier is employed to identify possible underlying patterns or rules for predicting the condition rank of track grids (or components), based on historical measurements of track grids (or components). Track irregularities prediction is taken as an example to verify the proposed model. The measurement data obtained by track geometry cars, ranging from 428.2 km to 480.6 km in the down direction along the Beijing-Kowloon Railway, were collected between February 2008 and July 2010. Our evaluation shows that the success rate of the proposed model is 82.3%.Real service life of track grids (or components) is estimated using the Markov based estimation model for railway track grids and components service life. Due to heterogeneity, uncertainty, etc. of track grids (or components) degradation, the degradation evolution of track grids (or components) is discretized into several condition ranks. And the degradation process where the condition rank of track grids (or components) deteriorates from condition rank i to condition rank i+1 is depicted by the life distribution function. The degradation process in different condition rank is described by using different life distribution functions. A hazard model is built to evaluate the hazard rate of track grids (or components). Employing the theory of Markov stochastic processes, the uncertainties of rail degradation between two adjacent renewal activities after an inspection cycle are described. Rail service life estimation is taken as an example to verify the proposed model. Inspection data for rail defects were accumulated from January 2010 to May 2015, ranging between 1,397 km to 1,720 km along the Longhai Railway. Our evaluation demonstrates that the estimated rail service life is very close to the real rail service life and meets railway management requirements.