The Recognition And Prediction Of High-speed Railway Track Slab Based On The Temporal-spatial Data Mining Of Track Geometry

High-speed railway slab track is currently the most widely used ballastless track structure in China.Under the influence of environmental factors such as wheel-rail dynamic load and temperature,the track slab structure is prone to diseases such as arching,warping and separation between layers,thereby reducing the track regularity,riding comfort and safety of train.The traditional analysis methods for track substructures including track slabs are mainly physical simulation modeling and wayside monitoring.However,the factors affecting the deformation of track slabs are complicated and physical model has certain limitations under real operating conditions.The wayside monitoring method can provide multivariate data for analysis.However,due to the cost constraints,the sensors cannot cover the entire railway line.Besides,the high-precision sensors are extremely vulnerable to external interference.The sensors may fail and even be damaged under extreme conditions,which make it difficult to achieve long-term continuous monitoring of track structures and large amounts of data collection.In view of the fact that track dynamic inspection is currently the most common detection method for railway infrastructures,and there is a certain mapping relationship between the the track slab deformation and the track irregularity,the author proposes an analysis method of track slab deformation based on the data collected from track geometry car.The time and frequency domain indicators of track irregularity are extracted to reflect the position and degree of track slab deformation.Then the author used the analysis results of multiple inspection data and explored the information of track slab deformation in temporal and spatial dimension.The main research work and related conclusions are as follows:1.The three-year dynamic inspection data of CRTS Ⅰ,Ⅱ,and Ⅲ track slab lines in China were obtained through investigation.The moving window waveform matching algorithm was proposed to eliminate sections with different changes pattern of the left and right irregularities.The author filtered the track surface center irregularity,and calculated the characteristic wavelength distribution of the track slab deformation on the railway line.2.The time-frequency analysis method of continuous wavelet transform is used to introduce the evaluation index of track slab deformation(Track Irregularity Degradation Index,TIDI),and the reasonableness of the proposed index is verified by using the virtual surface irregularity.Based on the wavelet energy calculation results of multiple inspection runs,a spatial-temporal distribution matrix of the track slab deformation index was proposed.Using the KDE(Kernel Density Estimation),DBSCAN(Density-Based Spatial Clustering of Applications with Noise)and LSTM(Long Short-Term Memory Neural Network),a spatial-temporal model for track slab deformation and degradation prediction are established.3.Analyze the track inspection data of CRTS Ⅰ,Ⅱ,Ⅲ slabs of bridges and roadbed sections respectively.The main research conclusions are as follows:(1)The deformation degree of the track slab is subject to periodic changes in temperature.Among them,the deformation degree of type Ⅰ and Ⅱ slab and Ⅲ-type slab of the roadbed are positively related to temperature,and the warping and deformation of the track slab may occur.The deformation of the Ⅲ-type slab on the bridge has a negative correlation with the temperature,and it is inferred that frost heave may occur.(2)The overall trend of the deformation of the Ⅰ-type slab does not change significantly with time,the deformation of the Ⅱ-type slab shows an overall upward trend with time, and the deformation of the Ⅲ-type plate has an overall downward trend with time.It is inferred that the residual deformation of the Ⅱ-type slab increases with time,and the roadbed settlement of the type Ⅲ slab may occur,thereby covering the arching on the track slab.(3)When using the LSTM neural network to predict the deformation degree of the track slab,adding historical data cannot significantly improve the prediction performance.As the number of prediction days’ increases,the prediction performance generally decreases. However,using this method can achieve short-term prediction of different track slab deformation within the next 15-30 days.(4)Different track slabs have different degrees of deformation and abnormal deformation length.Among them,the Ⅰ-type slab has the smallest deformation and the shortest abnormal deformation sections,the Ⅱ-type slab has the largest deformation and the longest abnormal deformation section.After using the TIDI value to identify the deformation position of the track slab,it is recommended to investigate the nearby sections at the deformation position.The recommended lengths of the deformation position detection sections for Ⅰ,Ⅱ,and Ⅲ slabs are 40 m,60m,and 80 m,respectively.