Ballast track bed sand-bearing state prediction method based on deep learning

Abstract

The invention provides a method for predicting the sand-bearing state of a ballast track bed based on deep learning. The method comprises the steps that a discrete element simulation model of a sandstorm track bed is established, and rigidity parameters of the sandstorm track bed in different positions and depth areas are obtained through analysis; based on the rigidity parameters, a vehicle-desertification ballast track coupling finite element simulation model is established, and the vibration acceleration of the sleeper in different sand-containing states is obtained through analysis; and inputting the vibration acceleration as training data into the constructed sand-bearing state prediction model to train the sand-bearing state prediction model to obtain a trained sand-bearing state prediction model, and performing sand-bearing state prediction by using the trained sand-bearing state prediction model to obtain a sand-bearing state prediction result. The sand-bearing state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network. According to the invention, the sand-bearing state of the ballast bed can be predicted with high precision. The method has important reference value for guiding line maintenance work and guaranteeing stable operation of the train.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for predicting the sand content state of a ballasted track bed based on deep learning. Background Art

[0002] In recent years, with the continuous intensive improvement of the railway network in the eastern part of my country, railway construction has also been increasingly inclined and developed towards the western region, and it is inevitable that railway lines will be built in extreme climate zones such as deserts. Due to the large number of gaps inside the loose track bed, the railways built in these areas are bound to face the threat of wind and sand to varying degrees. In a harsh wind and sand environment, the continuous intrusion of sand into the track bed will change the porosity and basic mechanical parameters of the track bed, especially the uneven distribution of sand in the track bed, which gathers at different depths and positions, resulting in local changes in the mechanical properties of the track bed. The gathered sand will cause uneven lifting of the rails and sleepers, causing line arches, forming triangular pits and low joints, etc., which seriously affect the track smoothness and endanger driving safety. Therefore, in-depth research on the mechanical characteristics of railway track structures in windy and sandy areas, especially the distribution of sand in the track bed, is of great significance for guiding line maintenance and repair operations and ensuring the smooth operation of trains.

[0003] At present, my country's research on railways in desert areas mainly focuses on windbreak and sand fixation along the railway, such as planting vegetation and setting up grass grids. Most of the experimental studies are based on indoor scaled tests, and there are few field tests, which makes it difficult to fully reflect the internal working mechanism and evolution law of the multi-scale overall roadbed particle system during the process of wind and sand intrusion. Traditional roadbed sand content detection methods mainly rely on on-site sampling and laboratory analysis, which mainly obtain the overall sand content of the roadbed and cannot provide specific distribution information of sand particles inside the roadbed, including the intrusion depth and spatial position. In addition, these methods have the disadvantages of being time-consuming, costly, and destructive to the track structure, and are difficult to meet the actual needs of on-site maintenance and repair.

[0004] Therefore, there is an urgent need for an efficient and non-invasive method for detecting the sand content in the roadbed, which can accurately predict the intrusion depth and spatial distribution of sand particles in the roadbed, provide refined data support for railway maintenance, and ensure the smooth and safe operation of railways in sandy areas. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the sand content state of ballasted track bed based on deep learning, which can accurately predict the penetration depth and spatial distribution position of sand particles in the roadbed.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] In a first aspect, the present invention provides a method for predicting the sand content state of a ballasted track bed based on deep learning, comprising:

[0008] Establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different locations and depths;

[0009] Based on the stiffness parameters, a vehicle-sanded ballasted track coupling finite element simulation model is established and the vibration acceleration of the sleeper under different sand content conditions is analyzed and obtained;

[0010] The vibration acceleration is used as training data and input into the constructed sand content state prediction model to train the sand content state prediction model to obtain a trained sand content state prediction model, and the trained sand content state prediction model is used to predict the sand content state. The sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

[0011] Furthermore, the establishment of a discrete element simulation model of the wind-sand roadbed includes:

[0012] Acquiring point cloud data of ballast particles, and generating a discrete element simulation model of the ballast particles based on the point cloud data;

[0013] The ballast structure is divided into multiple layers according to thickness by using the “layered generation – step-by-step filling – particle replacement” method;

[0014] After the discrete element simulation model of each layer of ballast particles is balanced and stable, the particle replacement method is used to replace the sand particle size with the required smaller particle size;

[0015] Import the sleeper model for assembly, set the basic physical property parameters and contact parameters of the discrete element simulation model of the ballast particles, and apply the gravity field;

[0016] The wall unit is used to simulate the roadbed, and the elastic characteristics of the roadbed are simulated by reasonably setting the stiffness of the wall unit;

[0017] Gravity is applied to make the discrete element simulation model of ballast particles reach equilibrium, and finally a discrete element simulation model of a sandy roadbed with different sand content states is obtained.

[0018] Furthermore, based on the stiffness parameter, a vehicle-sanded ballasted track coupling finite element simulation model is established, including:

[0019] Modeling vehicle components and meshing the ballasted track based on the stiffness parameters;

[0020] Based on the preset boundaries and constraints, the constructed vehicle components and the meshed ballasted track are adjusted to obtain a vehicle-desertified ballasted track coupled finite element simulation model.

[0021] Furthermore, the constructed sand content state prediction model includes a convolutional neural network and a long short-term memory network;

[0022] The step of using the vibration acceleration as training data and inputting it into the constructed sand content state prediction model to train the sand content state prediction model includes:

[0023] Inputting the vibration acceleration into a convolutional neural network to extract local spatiotemporal features;

[0024] The local spatiotemporal features are input into a long short-term memory network to extract long-term time-dependent information, and a prediction result of the sand content state of the wind-blown roadbed is obtained based on the long-term time-dependent information and the local spatiotemporal features. The constructed sand content state prediction model is trained and updated based on the prediction result.

[0025] Furthermore, the basic physical property parameters include density, Poisson's ratio, and Young's modulus of the material, and the contact parameters include coefficient of restitution, coefficient of static friction, and coefficient of dynamic friction.

[0026] Furthermore, the modeling of vehicle components and meshing of ballasted track based on the stiffness parameters include:

[0027] Use discrete rigid body units to simulate the car body and bogie, use analytical rigid body units to simulate the wheelset, and use solid beam units to model the rails and sleepers, and obtain the car body model, bogie model, wheelset model, rail model, and sleeper model accordingly;

[0028] Based on the stiffness parameters, the ballasted track bed is modeled using solid elements, the track bed is divided into multiple regions, corresponding material property parameters are assigned to each region based on the calculation results of discrete elements, a transition layer is set at the junction of the regions to make the material properties change smoothly, and eight-node linear hexahedral elements are used for meshing to obtain the ballasted track after meshing.

[0029] Furthermore, before inputting the vibration acceleration into the convolutional neural network, the method further includes:

[0030] The vibration acceleration is filtered using a Butterworth bandpass filter to obtain a processed vibration acceleration:

[0031]

[0032] Where H(f) is the vibration acceleration after processing, f is the frequency of vibration acceleration, and f c is the cut-off frequency and n is the order of the filter.

[0033] In a second aspect, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the program.

[0034] In a third aspect, the present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0035] In a fourth aspect, the present invention further provides a computer program product, comprising a computer program, wherein the computer program implements the above method when executed by a processor.

[0036] Beneficial effects of the present invention: The present invention provides a method for predicting the sand content state of ballasted track bed based on deep learning. The present invention establishes three core models: one is a discrete element model of wind-sand roadbed, the second is a vehicle-desertified ballasted track coupling finite element model, and the third is a deep learning model based on the combination of convolutional neural network and long short-term memory network. The discrete element model provides physical parameter basis for subsequent models, and the finite element model provides training data for the deep learning model. Through supervised learning, the deep learning model, with the support of a large amount of simulation data and combined with some experimental test data, learns the mapping relationship between the vibration acceleration signal of the sleeper and the sand content state. Through the synergistic effect of these three models, the high-precision prediction of the sand content state of the roadbed is successfully achieved. It has important reference value for guiding line maintenance and repair operations and ensuring the smooth operation of trains.

[0037] The method of the present invention is simple, easy to implement, flexible, efficient, and highly adaptable, and is applicable to different types of ballasted track structures and various train models. By adjusting the model parameters and training data sets, it can flexibly adapt to the detection requirements of the roadbed sand content under different lines and working conditions, and has good versatility and scalability. The proposal of the present invention can provide technical support for railway maintenance and repair in the northwest desert areas of my country, especially in areas affected by wind and sand.

[0038] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0040] Figure 1One of the flow charts of the method for predicting the sand content state of ballasted track bed based on deep learning provided in an embodiment of the present invention;

[0041] Figure 2 A second flow chart of a method for predicting the sand content state of a ballasted track bed based on deep learning provided in an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of a discrete element model of a wind-blown roadbed with different sand content states provided by an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of a vehicle-sanded ballasted track coupling finite element model provided in an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of the structure of a sand content prediction model provided by an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of a sleeper vibration acceleration curve when a train passes by provided in an embodiment of the present invention;

[0046] Figure 7 A schematic diagram of a curve showing the relationship between the sand content of the ballast and the change in stiffness provided in an embodiment of the present invention;

[0047] Figure 8 A schematic diagram of the sand distribution state inside the ballast provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0049] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "an" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0050] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0051] In view of the railway lines built in extreme climate zones such as the western deserts of my country, sand will continue to invade the roadbed in harsh windy and sandy environments. At the least, it will fill the gaps in the ballast and change the elasticity of the roadbed. At the worst, it will cause uneven tracks, increased wheel wear, and reduced service performance of the track structure, which will in turn affect the stability and safety of train operation. There is a lack of scientific basis for when to clean and screen, and scientific, reasonable and effective maintenance is urgently needed. However, there are still many shortcomings in the research on the invasion of wind and sand into ballasted roadbeds at home and abroad. The traditional roadbed sand content state detection method mainly relies on on-site sampling and laboratory analysis, which mainly obtains the overall sand content of the roadbed and cannot provide the specific distribution information of sand inside the roadbed, including the invasion depth and spatial position. In addition, these methods have the disadvantages of being time-consuming, costly, and destructive to the track structure, and it is difficult to meet the actual needs of on-site maintenance and repair. Based on this, the present invention proposes a method for predicting the sand content state of ballasted track roadbed based on deep learning from a technical perspective.

[0052] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0053] Example 1

[0054] See also Figure 1 and Figure 2 A method for predicting the sand content state of ballasted track bed based on deep learning comprises the following steps:

[0055] S101, establishing a discrete element simulation model of the wind-blown roadbed and analyzing to obtain the stiffness parameters of the wind-blown roadbed at different positions and depths.

[0056] In this step, by simulating different sand contents and spatial distribution of sand particles, the stiffness parameters of the ballast at different positions and depths are calculated, and the influence of sand intrusion and different distribution states on the mechanical properties of the ballast are analyzed. Figure 8 shown.

[0057] Specifically, point cloud data of ballast particles are obtained, and a discrete element simulation model of the ballast particles is generated based on the point cloud data. More specifically, laser scanning technology can be used to obtain point cloud data of the outer surface of ballast particles of different shapes; using a particle stacking method, spherical units are filled into the reconstructed three-dimensional profile of the ballast particles to generate a discrete element simulation model of the ballast particles.

[0058] The roadbed structure is divided into multiple layers according to thickness using the "layered generation - step-by-step filling - particle replacement" method. It should be noted that when modeling, according to the uneven distribution of sand particles, the required quality of ballast and larger sand particles are generated by the rain method in each layer and at different lateral positions according to the set sand content.

[0059] After the discrete element simulation model of each layer of ballast particles is balanced and stable, the particle replacement method is used to replace the sand particle size with the required smaller particle size.

[0060] The sleeper model is imported for assembly, the basic physical property parameters and contact parameters of the discrete element simulation model of the ballast particles are set, and the gravity field is applied.

[0061] The wall unit is used to simulate the roadbed, and the elastic characteristics of the roadbed are simulated by reasonably setting the stiffness of the wall unit.

[0062] Apply gravity to make the discrete element simulation model of ballast particles reach equilibrium, and finally obtain the following Figure 3 The discrete element simulation model of the aeolian roadbed with different sand content states is shown.

[0063] The basic physical property parameters include the density, Poisson's ratio and Young's modulus of the material, and the contact parameters include the coefficient of restitution, the coefficient of static friction and the coefficient of dynamic friction.

[0064] Density is obtained by consulting relevant literature on the material's own properties. Poisson's ratio, Young's modulus, etc. are obtained through direct shear tests and uniaxial compression tests.

[0065] The coefficient of restitution is solved by the inclined plane collision method;

[0066]

[0067] Where e is the coefficient of restitution, L is the horizontal distance between the object and the collision point when it falls to the horizontal plane, H1 is the release height of the object, H2 is the vertical height between the collision point and the horizontal plane, and θ is the inclination angle of the inclined plane.

[0068] The static friction coefficient is solved by the inclined plane method;

[0069] μ=tanα

[0070] Where μ is the static friction coefficient and α is the inclination angle of the inclined plane when sliding just begins.

[0071] The dynamic friction coefficient is obtained through the repose angle test and based on the numerical simulation model:

[0072]

[0073] In the formula, θ r is the repose angle, H is the stacking height, and D is the disk diameter.

[0074] In this embodiment, the stiffness parameters are obtained through the discrete element simulation model of the wind-sand roadbed, as shown in FIG. Figure 7 As shown, it shows the relationship curve between the sand content of the roadbed and the stiffness change, reflecting the impact of sand intrusion into the roadbed on the contact state between ballast particles and the mechanical properties of the overall roadbed structure, and thus provides a basis for the selection of the elastic modulus of the roadbed material in the finite element model.

[0075] S102, based on the stiffness parameter, establish Figure 4 The vehicle-sanded ballasted track coupling finite element simulation model shown is used to analyze the vibration acceleration of the sleepers under different sand content conditions.

[0076] In this step, a vehicle-desertified ballasted track coupling finite element model is established, the roadbed is divided into multiple areas, and corresponding material property parameters are assigned to each area to reflect the spatial distribution of sand particles and analyze the vibration acceleration response of the sleeper under train running conditions.

[0077] Specifically, vehicle components are modeled, and the ballasted track is meshed based on the stiffness parameters.

[0078] More specifically, discrete rigid body units are used to simulate the car body and bogie, analytical rigid body units are used to simulate the wheelset, and solid beam units are used to model the rails and sleepers, and the car body model, bogie model, wheelset model, rail model and sleeper model are obtained accordingly. In this embodiment, discrete rigid body units are used to simulate the car body and bogie, and parameters such as mass and moment of inertia are set. Analytical rigid body units are used to simulate the wheelset; the wheel-rail contact normal force is solved using Hertz nonlinear elastic contact theory, and the contact state between the wheel and the rail is assumed to be an elliptical contact spot. The wheel-rail contact tangential force only considers the relative sliding force caused by friction between the wheel and the rail; the rails and sleepers are modeled using solid beam units to simulate their elastic characteristics; the fastener and spring bar system is simulated using linear springs and viscous damping elements, and the vertical, lateral and longitudinal stiffness and damping coefficients along the line are set.

[0079] Based on the stiffness parameters, the ballasted track bed is modeled using solid elements, the track bed is divided into multiple regions, corresponding material property parameters are assigned to each region based on the calculation results of discrete elements, a transition layer is set at the junction of the regions to make the material properties change smoothly, and eight-node linear hexahedral elements are used for meshing to obtain the ballasted track after meshing.

[0080] Based on the preset boundaries and constraints, the constructed vehicle components and the meshed ballasted track are adjusted to obtain a vehicle-desertified ballasted track coupled finite element simulation model.

[0081] In this embodiment, the sandy ballasted track bed is modeled using solid elements. The track bed is divided into multiple areas. Corresponding material property parameters are assigned to each area based on the calculation results of discrete elements. A transition layer is set at the junction of the areas to make the material properties change smoothly. The model is meshed using eight-node linear hexahedron (C3D8R) elements. According to the actual situation on site, parameters such as train speed and axle weight are set, track unevenness is applied, and the boundaries and constraints of the model are set to obtain a vehicle-sandy ballasted track coupled finite element model.

[0082] In this embodiment, the following is obtained through dynamic analysis of the vehicle-track coupling finite element model: Figure 6 The sleeper vibration acceleration signal shown in the figure shows obvious periodic fluctuations when a train passes by, and the amplitude changes with the train load.

[0083] S103, using the vibration acceleration as training data and inputting it into the constructed sand content state prediction model to train the sand content state prediction model, obtain the trained sand content state prediction model, and use the trained sand content state prediction model to predict the sand content state, wherein the sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

[0084] Among them, a nonlinear mapping relationship between the sleeper vibration acceleration and the roadbed sand content state is established through the sand content state prediction model.

[0085] Specifically, the vibration acceleration is input into a convolutional neural network to extract local spatiotemporal features;

[0086] The local spatiotemporal features are input into a long short-term memory network to extract long-term time-dependent information, and a prediction result of the sand content state of the wind-blown roadbed is obtained based on the long-term time-dependent information and the local spatiotemporal features. The constructed sand content state prediction model is trained and updated based on the prediction result.

[0087] In addition, before inputting the vibration acceleration into the convolutional neural network, the method further includes:

[0088] The vibration acceleration is filtered using a Butterworth bandpass filter to obtain a processed vibration acceleration:

[0089]

[0090] Where H(f) is the vibration acceleration after processing, f is the frequency of vibration acceleration, and f c is the cut-off frequency and n is the order of the filter.

[0091] In addition, the training data is preprocessed by the following preprocessing methods:

[0092] The DC component and trend changes in the signal are eliminated by removing the mean and trend. The signal amplitude is normalized to a distribution range with a mean of 0 and a variance of 1 to eliminate the influence of dimensional differences and numerical scales on model training. Wavelet transform is used to obtain the time-frequency characteristics of the signal.

[0093] The preprocessed training data are divided into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15%, respectively.

[0094] The training set is input into the constructed sand content state prediction model, which includes a convolutional neural network (CNN) and a long short-term memory network (LSTM). The convolutional neural network includes a convolution layer, an activation layer, and a maximum pooling layer.

[0095] like Figure 5 As shown in the figure, the convolutional neural network performs convolution operations on the input training set to extract local spatiotemporal features. The number of convolutional layers and the kernel size are adjusted according to the model performance, and the activation function is ReLU. The maximum pooling layer is used to reduce the feature dimension and prevent overfitting.

[0096] The LSTM layer extracts features from the feature sequence output by the convolutional neural network to capture the temporal dependency of the feature sequence.

[0097] Finally, the features of LSTM output are mapped to the output layer through the fully connected layer. The output layer includes three nodes, which are used to predict the sand content of the roadbed, the depth of sand intrusion and the distribution position. The activation function of the output layer is a linear function. In this embodiment, the mean square error (MSE) is used as the loss function:

[0098]

[0099] In the formula, n is the number of samples, y is i is the true value, is the predicted value.

[0100] In addition, during the training process, the Adam optimizer is used for optimization, and a Dropout layer is added after the fully connected layer to prevent overfitting; an early stopping strategy is used to stop training when the validation set loss no longer decreases after several consecutive iterations. The actual collected sleeper vibration acceleration signal is preprocessed and input into the trained deep learning model to predict the sand content of the roadbed and achieve non-invasive, high-precision detection.

[0101] The method for predicting the sand content state of the ballasted track bed based on deep learning provided by the embodiment of the present invention firstly uses the discrete element method to establish a ballasted track bed model with different sand content states and sand distribution, analyzes the influence of sand intrusion and different distribution states on the mechanical properties of the ballasted track bed, and calculates the stiffness parameters of the track bed in different positions and depth areas. The discrete element simulation model of the wind-sand track bed shows the relationship curve between the sand content state of the track bed and the change of stiffness, reflects the influence of sand intrusion on the contact state between ballast particles and the mechanical properties of the overall track bed structure, and then provides a basis for the value of the elastic modulus of the track bed material in the finite element model. Secondly, based on the discrete element calculation results, a vehicle-sanded ballasted track coupling finite element model is established, and the vibration acceleration response of the sleeper under the train running conditions is calculated and analyzed. Then, the simulated vibration acceleration signal is subjected to data processing and feature extraction, and a deep learning training data set containing the sand distribution information inside the track bed is constructed. A hybrid model combining convolutional neural network (CNN) and long short-term memory network (LSTM) is used to extract the spatiotemporal features and time series features of the sleeper vibration acceleration signal, and establish a nonlinear mapping relationship between the sleeper vibration acceleration and the sand content of the ballast bed. Finally, in practical applications, high-precision acceleration sensors are installed on the sleepers to collect vibration signals, which are input into the trained deep learning model after preprocessing to predict the distribution of sand inside the ballast bed. This method realizes non-invasive and high-precision detection of the sand content of the ballast bed, can analyze the accumulation location and invasion depth of sand in the ballast bed, improves the precision and accuracy of detection, and is of great significance to the maintenance and safe operation of railways in sandy areas.

[0102] In the embodiment provided by the present invention, in the discrete element model of the wind-sand roadbed with different sand content states, the size range of ballast particles is 16 mm to 63 mm, the size range of sand particles is 0.3 mm to 3 mm, the shape of ballast particles is approximated by combined ball units, and the shape of sand particles is approximated by single ball particles.

[0103] The contact between ballast particles and between ballast particles and sand, sleepers, etc. are all solved and calculated based on the Hertz-Mindlin (No-slip) contact model.

[0104] The vehicle part of the finite element model includes the primary and secondary suspensions. The suspension parameters are set according to the actual train model and can accurately simulate the dynamic characteristics of the train on the sandy track.

[0105] In addition, the present invention can also modify specific parameters and model sizes according to the form of track structure, the type of ballast particles and the type of running train, and adjust the model parameters according to specific circumstances, and has wide applicability.

[0106] Example 2

[0107] On the basis of Example 1, this Example 2 provides a device for predicting the sand content state of a ballasted track bed based on deep learning. The device for predicting the sand content state of a ballasted track bed based on deep learning corresponds to the above-mentioned prediction of the sand content state of a ballasted track bed based on deep learning, and specifically includes:

[0108] The first model building module is used to establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different positions and depths;

[0109] The second model building module is used to establish a vehicle-sanded ballasted track coupling finite element simulation model based on the stiffness parameters and analyze and obtain the vibration acceleration of the sleeper under different sand content conditions;

[0110] The third model training module is used to use the vibration acceleration as training data and input it into the constructed sand content state prediction model to train the sand content state prediction model, obtain the trained sand content state prediction model, and use the trained sand content state prediction model to predict the sand content state, wherein the sand content state prediction model is a deep learning model based on convolutional neural network and long short-term memory network.

[0111] For specific details, please refer to the description of the ballasted track bed sand content prediction method based on deep learning, which will not be repeated here.

[0112] Example 3

[0113] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, the processor calls the program instructions to execute a method for predicting the sand content state of a ballasted track bed based on deep learning, and the method includes the following process steps:

[0114] Establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different locations and depths;

[0115] Based on the stiffness parameters, a vehicle-sanded ballasted track coupling finite element simulation model is established and the vibration acceleration of the sleeper under different sand content conditions is analyzed and obtained;

[0116] The vibration acceleration is used as training data and input into the constructed sand content state prediction model to train the sand content state prediction model to obtain a trained sand content state prediction model, and the trained sand content state prediction model is used to predict the sand content state. The sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

[0117] Example 4

[0118] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, a method for predicting the sand content state of a ballasted track bed based on deep learning is implemented. The method includes the following process steps:

[0119] Establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different locations and depths;

[0120] Based on the stiffness parameters, a vehicle-sanded ballasted track coupling finite element simulation model is established and the vibration acceleration of the sleeper under different sand content conditions is analyzed and obtained;

[0121] The vibration acceleration is used as training data and input into the constructed sand content state prediction model to train the sand content state prediction model to obtain a trained sand content state prediction model, and the trained sand content state prediction model is used to predict the sand content state. The sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

[0122] Example 5

[0123] Embodiment 5 of the present invention provides a computer program product, including a computer program. When the computer program is executed by a processor, a method for predicting the sand content state of a ballasted track bed based on deep learning is implemented. The method includes the following process steps:

[0124] Establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different locations and depths;

[0125] Based on the stiffness parameters, a vehicle-sanded ballasted track coupling finite element simulation model is established and the vibration acceleration of the sleeper under different sand content conditions is analyzed and obtained;

[0126] The vibration acceleration is used as training data and input into the constructed sand content state prediction model to train the sand content state prediction model to obtain a trained sand content state prediction model, and the trained sand content state prediction model is used to predict the sand content state. The sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

[0127] In summary, the method for predicting the sand content state of ballasted track bed based on deep learning provided by the embodiment of the present invention establishes three core models: one is the discrete element model of the wind-sand roadbed, the second is the vehicle-desertified ballasted track coupling finite element model, and the third is a deep learning model based on the combination of convolutional neural network and long short-term memory network. The discrete element model provides physical parameter basis for subsequent models, and the finite element model provides training data for the deep learning model. Through supervised learning, the deep learning model, with the support of a large amount of simulation data and combined with some experimental test data, learns the mapping relationship between the vibration acceleration signal of the sleeper and the sand content state. Through the synergistic effect of these three models, the high-precision prediction of the sand content state of the roadbed is successfully achieved. It has important reference value for guiding line maintenance and repair operations and ensuring the smooth operation of trains.

[0128] The method of the present invention is simple, easy to implement, flexible, efficient, and highly adaptable, and is applicable to different types of ballasted track structures and various train models. By adjusting the model parameters and training data sets, it can flexibly adapt to the detection requirements of the roadbed sand content under different lines and working conditions, and has good versatility and scalability. The proposal of the present invention can provide technical support for railway maintenance and repair in the northwest desert areas of my country, especially in areas affected by wind and sand.

[0129] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0130] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the method or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The method and system embodiments described above are merely schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0131] The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting the sand content of ballasted track bed based on deep learning, characterized in that: include: Establish a discrete element simulation model of the wind-sand roadbed and analyze and obtain the stiffness parameters of the wind-sand roadbed at different locations and depths; Based on the stiffness parameters, a vehicle-sanded ballasted track coupling finite element simulation model is established and the vibration acceleration of the sleeper under different sand content conditions is analyzed and obtained; The vibration acceleration is used as training data and input into the constructed sand content state prediction model to train the sand content state prediction model to obtain a trained sand content state prediction model, and the trained sand content state prediction model is used to predict the sand content state. The sand content state prediction model is a deep learning model based on a convolutional neural network and a long short-term memory network.

2. The method according to claim 1, characterized in that The discrete element simulation model of the wind-sand roadbed is established, including: Acquiring point cloud data of ballast particles, and generating a discrete element simulation model of the ballast particles based on the point cloud data; The ballast structure is divided into multiple layers according to thickness by using the "layered generation - step-by-step filling - particle replacement" method; After the discrete element simulation model of each layer of ballast particles is balanced and stable, the particle replacement method is used to replace the sand particle size with the required smaller particle size; Import the sleeper model for assembly, set the basic physical property parameters and contact parameters of the discrete element simulation model of the ballast particles, and apply the gravity field; The wall unit is used to simulate the roadbed, and the elastic characteristics of the roadbed are simulated by reasonably setting the stiffness of the wall unit; Gravity is applied to make the discrete element simulation model of ballast particles reach equilibrium, and finally a discrete element simulation model of a sandy roadbed with different sand content states is obtained.

3. The method according to claim 1, characterized in that The vehicle-sanded ballasted track coupling finite element simulation model is established based on the stiffness parameter, including: Modeling vehicle components and meshing the ballasted track based on the stiffness parameters; Based on the preset boundaries and constraints, the constructed vehicle components and the meshed ballasted track are adjusted to obtain a vehicle-desertified ballasted track coupled finite element simulation model.

4. The method according to claim 1, characterized in that: The constructed sand content state prediction model includes a convolutional neural network and a long short-term memory network; The step of using the vibration acceleration as training data and inputting it into the constructed sand content state prediction model to train the sand content state prediction model includes: Inputting the vibration acceleration into a convolutional neural network to extract local spatiotemporal features; The local spatiotemporal features are input into a long short-term memory network to extract long-term time-dependent information, and a prediction result of the sand content state of the wind-blown roadbed is obtained based on the long-term time-dependent information and the local spatiotemporal features. The constructed sand content state prediction model is trained and updated based on the prediction result.

5. The method according to claim 2, characterized in that: The basic physical property parameters include density, Poisson's ratio, and Young's modulus of the material, and the contact parameters include coefficient of restitution, coefficient of static friction, and coefficient of dynamic friction.

6. The method according to claim 3, characterized in that The modeling of vehicle components and meshing of ballasted track based on the stiffness parameters include: Use discrete rigid body units to simulate the car body and bogie, use analytical rigid body units to simulate the wheelset, and use solid beam units to model the rails and sleepers, and obtain the car body model, bogie model, wheelset model, rail model, and sleeper model accordingly; Based on the stiffness parameters, the ballasted track bed is modeled using solid elements, the track bed is divided into multiple regions, corresponding material property parameters are assigned to each region based on the calculation results of discrete elements, a transition layer is set at the junction of the regions to make the material properties change smoothly, and eight-node linear hexahedral elements are used for meshing to obtain the ballasted track after meshing.

7. The method according to claim 4, characterized in that Before inputting the vibration acceleration into the convolutional neural network, the method further includes: The vibration acceleration is filtered using a Butterworth bandpass filter to obtain a processed vibration acceleration: Where H(f) is the vibration acceleration after processing, f is the frequency of vibration acceleration, and f c is the cut-off frequency and n is the order of the filter.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: The device stores a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images