Research On Train-rail Contact Relationship Based On Image Processing

The study of the wheel-rail contact relationship is one of the necessary topics for the safe transportation of railway traffic.The development trend of high-speed and heavy-duty railways in China directly leads to the intensification of dynamic interaction between wheels and rails,which seriously affects the safety and stability of vehicle systems.Once a train derails,it will cause huge loss of life and property.Therefore,we must pay attention to the research of the wheel-rail contact relationship,and actively explore effective methods and paths to solve the derailment problem.Due to the complex wheel-rail contact relationship and the many factors that determine whether a train derails,this thesis used image processing technology and multi-sensor data fusion technology to track and process the wheel-rail contact status.The predicted dynamic derailment coefficient is used to guide the safe operation of the train,The main research content includes the following aspects:(1)According to the characteristics of the wheel-rail contact geometry,the wheel-rail contact model in the turnout zone and the no-turnout zone was established according to the displacement and the contact points.Combined the image characteristics of the wheel-rail contact area collected by the vehicle camera and the curve of the wheel lift over time in the track climbing experiment,the safety status classification of the wheel-rail contact state was made based on the difference in the relative displacement of the wheel-rail centerline,and the evaluation criteria for displacement derailment were established.(2)The geometric model of camera imaging is established,and the camera system is calibrated by solving the geometric model parameters through experiments and calculations.A two-step calibration algorithm that combines the factors of radial distortion,tangential distortion,and thin prism distortion are used.The actual on-site collection to distortion uncalibrated wheel-rail contact a recovery image frames which including non-fork area and turnout area,it can be seen from the bending degree of the sleepers that the correction effect is well.Introduce a regional energy term to improve the problem of insufficient curvature constraints of the T-snake model.(3)Combined mark control method is used to improve the problem of over-segmentation in the watershed algorithm.The morphological expansion and erosion operation is performed on the pre-processed image and the optimized gradient image is obtained by combining the image information entropy.The high-low hat transform is used to mark the threshold,and the result of the watershed segmentation is used as the initial contour of the T-snake transform.The experimental results showed that the watershed-optimized T-snake algorithm is ideal for image extraction of wheel-rail contact areas under complex backgrounds.(4)The multi-sensor datfactors considered are comprehensive.The wavelet neuralnetwork has strong nonlinear approximation and fault tolerance.The optimization of genetic algorithm can improve the accuracy and convergence speed of the prediction model.The relative displacement,speed,acceleration,and wheel load reduction ratio are input for data fusion to predict the derailment coefficient.Used K-fold cross validation,the results showed that the prediction model has well robustness and accuracy.