Research On Assessment Method Of Track Geometry State Based On Vehicle Response

Scientific track geometry state assessment method is the foundation of ensuring safe and comfortable operation of railway vehicles.Vehicle response is an important indice to assess the safety and comfort of vehicle operation.However,due to the complex relationship between track geometry and vehicle response,assessment methods based on track geometry peaks and track quality index cannot objectively reflect the safety and comfort of vehicle operation.Therefore,it is necessary to study and establish a track geometry state assessment method based on vehicle response.On the basis of the correlation analysis between track geometry and vehicle response,vehicle response prediction models are proposed by using the deep learning method.Then vehicle response indices such as wheel-rail force,vehicle-body acceleration and comfort indice are predicted by using track geometry.Finally,track geometry indices and vehicle response indices are synthesized by using analytic hierarchy process and fuzzy comprehensive evaluation,and a track geometry state assessment method is proposed.The main contents and achievements are as follows:(1)By means of statistical analysis and time-frequency analysis,the correlation and the multi-band statistical characteristics between track geometry and vehicle response are systematically analyzed.In view of the complex relationship between track geometry and vehicle response,combining correlation analysis,coherence analysis and transfer function,the correlation between track geometry and vehicle response is studied,from the aspects of single-variable and multi-variable,linear and nonlinear,time domain and frequency domain.A multi-band component extraction method based on wavelet analysis is proposed,then the probability distribution characteristics and repeatability of multi-band components of track geometry and vehicle response are studied.All these above lays the foundation for the data modeling of vehicle response prediction.(2)Based on deep learning method,the instantaneous vehicle response prediction model and the vehicle response sectional state indice prediction model are proposed.In view of the deficiency of traditional machine learning models in feature learning and generalization ability,based on the long short-term memory LSTM with trend information learning ability in deep learning,the LSTM instantaneous vehicle response prediction model is established.In view of the shortage of the LSTM instantaneous vehicle response prediction model which is difficult to effectively learn the long-distance trend information and multi-band statistical characteristics of track geometry,the convolutional neural network CNN with the ability of shape feature learning and adaptive filtering is integrated with the LSTM,and the CNN-LSTM instantaneous vehicle response prediction model is proposed to improve the prediction accuracy.According to the logical relationship of ‘track geometry ? vehicle response ? sectional state indice',the multi-layer perceptron MLP is integrated with the CNN-LSTM,and the CNN-LSTM-MLP vehicle response sectional state indice prediction model is proposed,which uses the MLP to learn the complex mapping relationship between vehicle response and sectional state indice.Besides,a multi-objective loss function and a training method based on local gradient descent algorithm are established.The proposed models provide support for vehicle response prediction and track geometry state assessment based on vehicle response.(3)Considering the safety of train operation and the passenger ride comfort,the whee-rail force,vehicle-body acceleration and comfort indice are predicted by using track geometry,and the performance and predictions of the models are compared and analyzed.The wheel-rail force is predicted by LSTM instantaneous vehicle response prediction model,and then the derailment coefficient,wheel unloading ratio and other safety indexes are calculated.The vehicle-body acceleration is predicted by CNN-LSTM instantaneous vehicle response prediction model.And comfort indice is predicted by CNN-LSTM-MLP vehicle response sectional state indice prediction model.In order to optimize the performance of the models,structural parameters and training parameters of the proposed model are determined by parameter sensitivity analysis and prediction performance comparison.It is found that the absolute prediction error of wheel-rail force and vehicle-body acceleration is less than the inspection precision,and the accuracy grade of the proposed model in predicting comfort index is close to level 1,outperforming traditional machine learning models.In addition,the predicted wheel-rail force and vehicle-body acceleration are helpful to locate local track geometry defects that affect the safety and comfort of vehicle operation,and the predicted comfort indice is helpful to identify the track sections with poor ride comfort.(4)The existing track geometry indices and the predicted vehicle response indices are synthesized by using analytic hierarchy process and fuzzy comprehensive evaluation,and a fuzzy comprehensive assessment method of track geometry state is proposed.In view of the deficiency that the existing track geometry assessment method cannot objectively reflect vehicle response,a factor set is established with the predicted vehicle response indices as the main factor and the existing track geometry indices as the auxiliary factor.Besides,the existing assessment grade levels are refined,and a grade set and a comprehensive assessment system are established.According to the existing management standards,the multi-level management standards of track geometry indices and vehicle response indices are determined.In view of the shortcomings of the existing deduction method,the continuous deduction function is proposed,and then the calculation method of factor set based on the deduction function is established.Then,analytic hierarchy process is used to calculate the weight of the factor set,and fuzzy comprehensive evaluation is used to calculate the fuzzy membership of the grade set.On this basis,a numerical-continuous fuzzy comprehensive assessment indice and the corresponding management standard with fuzzy boundaries are put forward,thus establishing the fuzzy comprehensive assessment method of track geometry state.By comparing with the existing track quality index assessment method,it is found that the proposed fuzzy comprehensive assessment method can reflect the track smoothness,vehicle running safety and passenger ride comfort at the same time,and can make continuous,fuzzy and refined comprehensive assessment of the track geometry state.