Damage Identification Of CA Mortar Layer Based On BP Neural Network

Ballastless track is affected by long-term vehicle load,natural environment and other factors,which will inevitably produce a series of damage,affecting the performance and service life of track equipment.Among them,the damage of CA mortar layer is the most common,so it is of great engineering significance to seek the identification method of damage location and degree of CA mortar layer.In this paper,an BP neural network method is used to identify the large area damage and local minor damage of CA mortar layer,and the dynamic characteristics of ballastless track are taken as input to realize the damage identification of ballastless track CA mortar layer,and the anti-noise analysis is carried out to verify the feasibility of the BP neural network method in the field of ballastless track damage identification.Aiming at the large area damage of CA mortar layer,the finite element model is established and verified with the measured results of Hada Line.By comparing the sensitivity of frequencies and modes to structures before and after structural damage,this paper proposes that the difference of curvature modes of structural vibration and the new damage index(35)(37)_i/(35)w_j can be used as the input of neural network to identify the damage location and degree of Ballastless track,respectively.For damage location identification,the accuracy of the neural network is 100%when there is no noise.After adding 15%noise to the test sample,the damage location identification will be disturbed,but the specific damage location can still be identified.When the noise is up to 20%,the location can not be identified.For damage degree identification,the anti-noise ability of the neural network will increase with the increase of the damage degree,and when the damage degree is increased,the anti-noise ability of the neural network will be enhanced.When more than 50%,the neural network still recognizes the test samples mixed with 15%noise effectively.Aiming at the small area damage of CA mortar layer,the finite element model is established and verified by the test results.The damage distribution is studied by two-dimensional curvature mode difference,and the exact damage location is obtained.Taking(35)(37)_i/(35)w_j as the input of the neural network,it is concluded that the noise resistance of the neural network increases with the increase of the damage degree and area.When the damage area reaches 0.1m×0.1m and the damage degree reaches 50%,the neural network can effectively identify the test samples mixed with 15%noise.The results show that the combination of artificial neural network and modal analysis data can not only accurately estimate the location and severity of the damage,but also effectively identify the damage after adding large noise.In summary,it is reliable and effective to use the mode and frequency of structural vibration response as damage index and artificial neural network as damage detection tool for Ballastless track.