Study On The Evaluation Technology Of The Track Maintenance Quality In High-Speed Railway

Track irregularity of high speed railway directly determines the running safety and comfort of track-vehicle system. At present, the track irregularity management mainly uses the amplitude value and the mean value management. From the point of development trend, it will be hot and difficult problem to comprehensively evaluate track status by use of dynamic inspection data including track irregularity, car-body acceleraton data from track inspection car or comprehensive inspection car, car-body acceleraton data from operating car, car-body acceleraton data from locomotive, human sensory data. In this paper, these methods of signal processing, data mining, neural network technology are combined to investigate the unit evaluation model of railway line maintenance of high speed railway. The main research contents of this paper are as follows.(1) Research the automatic preprocessing methods for massive track geometry inspection data ofhigh speed railway. By virtue of maximum correlation principle, Fourier transform, the absolute average value, some new automatic prepcosssing algorithms are provided and realized, which include kilometer correction, trend filtering, single abnormal valure filtering, and distributed abnormal value filtering. These new methods low efficiency and diversity problems derived from elimating the invalid data by manual editing, which can provide reliable data support for overrun judgment andthe calculation of track state characteristic parameters.(2) Research the selection and analysis methods of deviation data of feature parameters from car-body acceleraton data of operating car and car-body acceleraton data of locomotive and human sensor data. Divide the deviation data of every month into four parts with respect to four weeks, and choose these deviation data which satisfy repeatability principle for track unit status evaluation. The analysis results show that, compared with the selection deviation method at random, the selection method based on repeation principle is more reasonable and more scientific, and the evaluation results are more reliable.(3) Research and present a comprehensive evaluation method for track unit status by the combination of LVQ neural network, pairwise comparison matrix, and fuzz clustering method. The feature parameters weight coefficients are calculated from pairwise comparison matrix, which will be used to calculate the quantitative scoring indexs of the track units. Fuzzy clustering method is introduced to determine the unit level on basis of random sample from a vast amount of measured data. Finally the LVQ neural network is applied to establish the comprehensive evaluation model of track unit status with the quantitative scoring indexs as model input and track unit stauts level as model output.(4) On the basis of the principles of unified application display, multi-source data access at the same time and to be scalable, the high speed railway track unit management system is designed in detail and developed, which is for the use of the comprehensive evaluation of track unit status.