چكيده به لاتين
Turnouts, in addition to being very vital and important elements in rail transportation, are accident-prone areas with high interaction forces due to the discontinuity of the rail. Today, with the increasing need to promote the speed and axial load, and considering the high cost of supply and maintenance of turnouts, constant attention to the health of the turnouts and the safe running of the vehicles at the maximum reliable speed is among the necessities of the railway industry.
Developing models based on monitoring system data to predict turnout vibration conditions can help to identify harmful trains, determine the safe speed of the needle, and finally reduce the cost of maintaining and supplying the needle. The models available in the technical literature have several limitations. In this study, while stating these limitations, a new, accurate and practical method based on actual measured data is presented.
The aim of the thesis is to create a model to predict the behavior of the turnout based on the mechanical and geometric parameters recorded during the passage of different rolling stocks through data mining of the database of the continuous monitoring system of the turnout.
For this purpose, the data of the needle monitoring system of the Yatri station on the Tehran-Mashhad axis has been used, which detects the daily passage of more than 30 trains, including more than 7 million data per day.
In this thesis, along with the initial processing and evaluation of the output of all the sensors of the monitoring system, the modeling of the vibration behavior of the turnout during the passage of passenger trains and its prediction at higher speeds in the crossing panel has been done. Determining the safe running speed of the train on the main line of the crossing is another goal of this thesis.
The safe running speed of the train from the turnout depends on various factors such as the geometric and mechanical characteristics of the turnout and the state of maintenance of the turnout and the vehicles (such as the age of the turnout and the vehicles, the maintenance history of the turnout and the vehicles, etc.), and the set of these factors affects the dynamic response of the turnout. In the crossing panel, using a recursive elimination method among the parameters effective in the vibration response of nose rail, independent input parameters including speed, number of wagons, axle load of the locomotive and wagons were selected to be supplied to the soft computing prediction model.
The local linear model (Lolimot) has been used to predict the vibrations of the crossing panel based on the data related to the passage of 146 passenger trains. The results have been verified by two validation methods in comparison with the measurement values, and the maximum prediction error of the model is 6% and 5% respectively for cross-validation and extrapolation.
The results of the study in the crossing panel led to two distinct outputs of passenger train classification and determining the safe speed of trains from the main line of the turnout. Lolimot divided the trains into four categories according to the pattern of the turnout response model, among which categories 1 and 2 were identified as critical categories according to the results of sensitivity analysis. Using the prediction results of the model and the definition of the speed increase risk index, the severity of the risk was estimated for all trains up to a speed of 130 km/h based on novelty detection. The speed of 130 km/h has been chosen to assess the risk of trains according to the speed limit in the collected data, and by increasing the speed of trains to 130 km/h, the tests can be done again up to 160 km/h. According to the results, 19 cases of trains in critical categories 1 and 2 were identified as having risk, of which only 10 cases have medium and high risk for increasing the speed from 100 to 130 km/h.
The data of blade movements can be used practically to monitor the amount of vibration and the effect of wheel shocks caused by the passing of a train. The classification of trains is based on the measurement of switch blade lateral displacements (BLD) due to different train/turnout interaction behaviors.
Data with distance from normal data were identified by novelty detection method and assigned to second class trains. The main difference in the blade vibration caused by the detected trains was the free vibration in the intermediate areas between two bogie pairs passing, which does not necessarily occur in trains with maximum blade vibration.
Also, trains were classified using rail strain measurements, and 80% of the identified trains matched the results of the novelty detection method.
Also, all the trains detected by this method were the subset of trains that were identified by novelty detection. Finally, strain patterns and BLD signals for two representative trains were compared and significant differences in values and patterns were observed.
The importance of this part of the research is in providing a new solution for track-side monitoring and identifying different vibration patterns of the blade rail in an unsupervised way. This study is an initial attempt to measure and process line responses using a displacement sensor in a railway switch panel.
The innovations of this thesis can be summarized in the main axes, including the development of a soft computing model for predicting the vibration of the crossing panel, identifying the safe speed of passing through the crossing panel, and identifying and classifying the status of trains from the switch blade displacements using the novelty detection method.
