Study On Damage Identification Of High Speed Turnouts Based On Rough Set Theory

High speed railway turnout is the basic equipment in the railway line.Compared with the basic rail,the composition of turnout is more complex and changeable.During the operation of turnout.the impact of the wheel passing on the track and other external factorsis are easy to cause the damage of turnout track,such as deformation,spalling,falling block,crack or even fracture.Therefore,the research on the damage monitoring of high-speed railway turnout plays an important role in maintaining and maintaining the switch timely and accurately and ensuring the safe operation of the train.When the train passing through the turnout,the vibration response signal generated by interaction between wheel and rail contains the important information of the current running state of the turnout.Therefore,the acquisition of vibration response signals with sensors is the basis of turnout monitoring.However,it is usually a non-stationary signal with complex components.Empirical Mode Decomposition can adaptively decompose non-stationary time-varying signals into different frequency bands on the frequency plane and showing features in time-frequency domain of the signal,but due to the high speed turnout of complex structure,the mechanism,mode and location of damage are changeable.How to identify and determine the precision of high speed turnout damage is still a hot issue for the experts all over the world.Rough set theory is a data analysis theory can effectively analyze and discover the implicit knowledge from the data,and plays an important role in optimizing the structure damage feature parameter sets and selecting damage feature parameter sets which are sensitive to damage operating mode.Neural network is one of the most important nonlinear modeling tools in pattern recognition.It is widely used in the research of structural damage identification.However,there are still some shortcomings in rough sets theory and neural networks,such as classical rough sets theory can only deal with symbolic data,the generalization ability of decision rules is insufficient,and some parameters of neural networks are difficult to set.In view of these problems,this dissertation is based on the study of rough sets theory.EMD theory is used to analyze the random response signals of turnouts.Combining the rough set theory and RBF neural network modeling method,the intelligent damage identification of high speed turnout based on vibration signal is studied.(1)A method based on IMF energy singular entropy for extracting the characteristics of high speed turnouts is proposed.Firstly,the advantages and disadvantages of the current nondestructive testing methods for high-speed rail are analyzed and compared,and the experimental method based on vibration response signal is given.Based on the EEMD theory,the relationship between the IMF component and the modal response is analyzed.The energy ratio characteristics,singular entropy features and energy singular entropy characteristics of the vibration signals of the high-speed turnouts are extracted.By comparing the calculation of the three different cases,it is proved that the energy singular entropy feature extraction method based on EEMD has little difference with other two methods,but the stability is higher than the former two.By adding different intensities of Gauss white noise to the measured signal and extracting the corresponding characteristic parameters,the robustness of the method to noise is proved.(2)A discrete method based on rough sets for the damage characteristics of high speed turnouts is proposed.For the discretization problems of continuous attributes in the theoretical study of Rough Sets Theory,in this dissertation,the characteristics of several common discretization methods are analyzed.Based on the improved Naive Scale algorithm,a global discretization method based on the positive domain is proposed.Using the improved global discretization method of rough sets,the damage characteristics of high speed turnouts in two experimental environments are discretized.(3)Based on attribute reduction theory of rough sets,the selection and optimization method of high speed turnout damage characteristics is proposed.In this dissertation,the shortcomings of the two common kinds of rough sets attribute reduction methods are analyzed.A reduction method of inconsistent decision table based on power graph is proposed.It solves the problem of attribute reduction method based on discernibility matrix which is not suitable for inconsistent decision table.By using the concept of positive domain,an equivalent fitness function is proposed,which solves the problem that the fitness function of the attribute reduction method based on random algorithm does not satisfy the equivalence.A solution space partitioning method is proposed to solve the problem of repeated iteration in attribute reduction problem based on stochastic optimization algorithm.On the basis of the above improvements,the discretization decision table of each measurement point in high speed turnout damage test are reduced and the minimal attribute reduction set are extracted.The method does not need structural dynamics analysis,and establishes the damage decision model directly according to the vibration signal.(4)A method of damage identification for high speed turnouts based on rough set and radial basis function neural network is proposed.Firstly,the energy singular entropy characteristics of high speed turnoutes are discretized.The rules are extracted based on attribute reduction of rough sets to obtain the initial decision model set.Then,the number of cluster centers and the location of initial cluster centers are determined.Finally,the initial location of the hidden center of the RBF neural network is determined by the improved rough k-mean algorithm,and the weights of radial basis function neural networks are optimized by artificial fish swarm algorithm.The method combines the advantages of the rough set theory,the rough k-clustering algorithm and the radial basis function neural network,and has strong adaptability.The experimental results of highspeed turnout show that the radial basis network designed by this method has simple structure and good generalization performance.It can improve the damage identification efficiency of high-speed turnout to a certain extent,and provides a common solution for non-stationary signal classification.