Identification Method Of Subgrade Settlement Of Ballastless Track Based On Vehicle Response And Machine Learning

Under the long-term high-frequency train load and complex environment,the subgrade settlement of high-speed railway will inevitably occur,which seriously affects the safety and stability of high-speed train.The existing settlement monitoring methods are generally applicable to typical construction sites and local sections,with low efficiency and high cost.There is a lack of a simple identification method that can reflect the subgrade settlement state in real time and accurately.Therefore,this paper taken the settlement disease of double block ballastless track subgrade as the research object,studied the vibration law of vehicle system by establishing the vehicle-track-subgrade vertical coupling dynamic model considering subgrade settlement,and realized the effective identification of ballastless track subgrade settlement disease by combining feature extraction,support vector machine,convolution neural network.The main research results and conclusions are as follows:(1)This paper analyzes the current research status of subgrade settlement monitoring and identification methods and ballastless track intelligent identification,and puts forward a ballastless track subgrade settlement identification method based on vehicle system vibration response: study the vibration law of vehicle system under subgrade settlement,consider the additional influence of track structure layer damage or deformation on settlement,and obtain the vibration sensitive characteristics of vehicle system,based on the concept of feature extraction,two subgrade settlement recognition algorithms,PSO-SVM and CNN-SVM,are proposed to recognize subgrade settlement.(2)The vehicle-track-subgrade coupling dynamic model considering subgrade settlement is established,and the vibration response law of vehicle system under different subgrade settlement conditions is studied.The settlement wavelength is 10 m,and the amplitude increases from 0mm to 30 mm.When the vehicle passes through the settlement section,the maximum vertical acceleration of the vehicle body,bogie and wheel set gradually increases,with the corresponding maximum increases of 754.5%,134.8% and 19.4% respectively.The vertical vibration acceleration of vehicle body and bogie is relatively sensitive to subgrade settlement.At the same time,the changes of vehicle passing speed and track irregularity spectrum have relatively little impact on the vibration law of these two indexes.(3)The additional influence of damage or deformation of track structure layer is studied,and the vibration characteristics of vehicle system sensitive to the settlement of ballastless track foundation are put forward.Considering the failure of one pair of fasteners,the separation of the support layer from the joint(the separation length is 5m and the height is 2mm)and the cosine curve settlement(the wavelength is 10 m and the amplitude is 10mm),the increase of the vertical acceleration of the vehicle body is 22.2%,88.9% and 588.9% respectively,and the increase of the vertical acceleration of the bogie is 5.1%,11.6% and 75.4% respectively;The influence of each working condition on vehicle body vibration is mainly concentrated at 1 ～ 10 Hz.At this time,the influence of subgrade settlement on vehicle body and bogie vibration is much greater than that of other working conditions;The vertical acceleration of vehicle body is most affected by subgrade settlement and can be used as a sensitive feature for identifying the degree of subgrade settlement.When judging whether the subgrade settlement exists under the influence of additional factors such as fastener failure and bearing layer separation,the vertical vibration acceleration of vehicle body and bogie can be used as a sensitive feature for identifying.(4)Based on the vibration sensitive characteristics of vehicle system,on the basis of support vector machine theory,artificial feature extraction and convolution neural network automatic feature extraction are compared and considered respectively,and the effective identification of ballastless track subgrade settlement is realized.The recognition rate of PSO-SVM and CNNSVM algorithms for subgrade settlement is 100%,the minimum recognition rate of settlement wavelength is 97.56% and 100% respectively,and the minimum recognition rate of amplitude is84.78% and 97.56%.Both algorithms can effectively judge the existence,wavelength and amplitude of subgrade settlement,but the accuracy of CNN-SVM algorithm is always better than PSO-SVM algorithm.When the vehicle passing speed and track irregularity spectrum are changed,the interference effect on settlement amplitude identification is obvious.At this time,the maximum difference of amplitude recognition rate between PSO-SVM and CNN-SVM algorithm is 2.44%and 5.94% respectively,indicating that the robustness of CNN-SVM algorithm is stronger than PSO-SVM algorithm.