Research On Vibration Characteristics And Detection Of Vehicle-floating Plate Track Under Steel Spring Failure

With the rapid development of subway transportation in China,the importance of subway track maintenance has become increasingly prominent.As the main supporting element of floating plates,steel spring vibration isolators are prone to lose their working performance under repeated train loads.The disabling of steel springs can change the stress state of the track structure,thereby causing damage to other structural components,threatening the stability and safety of train operation.In order to ensure the safe operation of subway trains,it is necessary to promptly investigate and detect the disabled steel springs.However,there is relatively little research on the detection of the disabled steel springs at present.Therefore,studying convenient and efficient detection methods for disabled steel springs has important guiding significance for maintaining subway traffic operations.Based on the vehicle track coupling dynamics theory,this thesis establishes a vertical coupling dynamics model of the vehicle steel spring floating plate track system.Firstly,the number and location of the disabled steel springs and the dynamic response of the vehicle and floating plate track under different combinations are analyzed,and then the response data of the measurement points are extracted to obtain the difference between the acceleration of the disabled steel spring and the vehicle body under normal conditions.Based on the difference in vehicle body vibration acceleration,a dual cycle layer LSTM neural network model is used to study the detection of disabled steel springs.The research content and main conclusions are as follows:(1)Considering that the vibration acceleration of the vehicle body under the disabling of steel springs is determined jointly by the disabling position,the state of the previous operating section,and the operating state of the vehicle,the LSTM neural network model with a "memory structure" and a relatively complex internal structure was ultimately selected to detect the disabled steel springs.Then,Hypermesh,Ansys,and UM were used to jointly establish a vehicle floating slab track vertical coupling dynamic model,and the calculated results of the model were compared with the experimental results,laying a foundation for the subsequent construction of neural network model data sets.(2)The calculation conditions of steel spring energy loss under different combinations are designed,and the impact of steel spring energy loss on the vibration response of the vehicle floating slab track system under different conditions is analyzed.The results show that the deactivation of the steel spring results in a change in the stiffness of the lower support of the floating slab,which increases the vibration response of the vehicle floating slab track system.When the number of continuously disabled steel springs on one side is greater than or equal to 2,the maximum vertical displacement of the rail and floating slab exceeds the limit requirements specified in CJJ/T 191-2012 Technical Specification for Floating Slab Track;When there are two disabled steel springs on the same plate,the vehicle is more likely to roll at the one side disabled steel spring than at the two side disabled and intact steel springs;When the number of disabled steel springs is the same,the vibration response of the same floating plate is greater than the plate end position.(3)The disabling of steel springs can increase the dynamic response of various components of the vehicle and track system,but the extraction of the vertical acceleration of the vehicle body is faster and more convenient than the dynamic indicators of other components.Therefore,this dynamic indicator is used to detect the disabling of steel springs.Based on the difference in acceleration between disabled steel springs and normal vehicle bodies,the classification effect of LSTM on disabled and normal steel springs was studied using a two classification method.The results show that the constructed vehicle vertical acceleration data sample is input into the LSTM neural network model,and the loss function value is minimized through multiple adjustments of parameters.The optimal model is determined as follows: the number of training periods is 30;Adopt double circulation layer;The number of units in the first and second circulation layers is 32 and 64,respectively;The loop layer input has a Dropout rate of 0.5.The index values of Accuracy,FDR,and FAR,the three indicators used to evaluate the disability detection of steel springs,are 76.32%,74.18%,and 20.63%,respectively.