Method and device for determining aerodynamic load of train and train

Abstract

The invention discloses an aerodynamic load determination method and device of a train and the train, and relates to the field of aerodynamics. The method comprises the steps that geometric shape data of the train with the load to be determined are obtained; inputting the geometric shape data of the train with the to-be-determined load into the prediction model to obtain the pressure and shear stress of the train with the to-be-determined load, which are output by the prediction model; and according to the pressure and the shear stress of the train with the to-be-determined load, determining the resistance, the lift force and the lateral force of the train with the to-be-determined load. The prediction model is trained through the data of the train with the known load, the trained prediction model only needs to input the geometric file of the train object to be calculated, the pressure and the shear stress can be obtained, the threshold in the professional field is not involved, and therefore all workers have the operable ability. Simulation calculation is replaced by the prediction model, the prediction model does not need to be subjected to simulation calculation again, the pressure and the shear stress are directly output, the calculation time is shortened, and the calculation efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of aerodynamics, and in particular to a method, apparatus, and train for determining aerodynamic loads. Background Technology

[0002] Currently, the evaluation of train aerodynamic performance typically employs two methods: testing and simulation. Testing requires providing a full-scale vehicle or scaled-down prototype model, resulting in high manufacturing costs. Furthermore, the testing process and methods are limited by experimental conditions, significantly restricting the content, procedures, and results of the tests, thus failing to provide comprehensive aerodynamic performance evaluation throughout the entire train development, design, and manufacturing cycle. While simulation is an effective means of evaluating train aerodynamic performance, conventional fluid dynamics simulation methods are challenging for trains, whose geometric dimensions span millimeters to hundreds of meters, making geometric modeling and simplification difficult and time-consuming. Conventional simulations usually utilize commercial software or open-source code for model processing, mesh discretization, and equation solving. Both packaged software and open-source code place high demands on users, creating a high barrier to entry. Moreover, for high-speed trains, conventional fluid dynamics simulations are time-consuming and computationally inefficient. Summary of the Invention

[0003] The purpose of this invention is to provide a method, apparatus, and train for determining aerodynamic loads on a train. By training a prediction model using data from a train with known loads, the trained model only requires the input of the geometric file of the train object to be calculated to obtain pressure and shear stresses. This method does not require specialized expertise, making it accessible to all personnel. By replacing simulation calculations with a prediction model, the model eliminates the need for further simulation calculations, directly outputting pressure and shear stresses, thus reducing calculation time and improving computational efficiency.

[0004] To solve the above-mentioned technical problems, the present invention provides a method for determining the aerodynamic load of a train, comprising:

[0005] Obtain the geometric shape data of the train with the load to be determined, wherein the geometric shape data is a grid discretized from the surface of the train;

[0006] The geometric shape data of the train with the load to be determined is input into the prediction model to obtain the pressure and shear stress of the train with the load to be determined output by the prediction model. The prediction model is obtained by training the prediction model using the geometric shape data, pressure, shear stress and operating parameters of the known load train. The operating parameters include the train's operating speed.

[0007] The resistance, lift, and lateral force of the train under the load to be determined are determined based on the pressure and shear stress of the train under the load to be determined.

[0008] On the other hand, the process of acquiring the geometric shape data of the known load train includes:

[0009] The surface of the known load train is divided into multiple discrete grids, each grid being a triangular grid consisting of three vertices, and the side length of each triangular grid is positively correlated with the length of the train;

[0010] The mesh is modified to form an equilateral triangular mesh;

[0011] Determine the three coordinates of the center point of each grid and the projected area of ​​each grid in the x, y and z directions.

[0012] On the other hand, after dividing the surface of the known-loaded train into multiple discrete grids, the method further includes:

[0013] Delete grids that intersect or overlap;

[0014] Delete grid cells whose area is outside the preset area range;

[0015] Delete grid cells whose interior angles are outside the preset angle range;

[0016] The mesh is corrected, including:

[0017] Correct the mesh after the deletion operation.

[0018] On the other hand, after dividing the surface of the known-loaded train into multiple discrete grids, the method further includes:

[0019] The origin of the x-direction coordinate is taken as the length center of the known load train, with the train head facing -x. The origin of the z-direction coordinate is taken as the track surface of the known load train, with the height direction being +z. The y-direction is determined based on the x-direction, the z-direction, and the right-hand rule.

[0020] The grid is normalized and aligned according to the x, y, and z coordinates obtained from the division.

[0021] The mesh is corrected, including:

[0022] The grid after coordinate normalization and alignment is corrected.

[0023] On the other hand, the mesh is modified, including:

[0024] Adjust the positions of the three vertices of the mesh to change the shape of the mesh;

[0025] And / or merge multiple adjacent grids with side lengths below a side length threshold into a polygon, and redivide the polygon into multiple equilateral triangles;

[0026] And / or for a quadrilateral formed by two adjacent grids, if the side lengths of the two grids generated by the current diagonal are both lower than the side length threshold, then the diagonal is flipped, and the quadrilateral is reconstructed into two triangles.

[0027] On the other hand, the process of obtaining the pressure and shear stress of the known load train includes:

[0028] The center point of a plurality of discrete grids into which the surface of the known load train is divided is determined as the point of application of the pressure, and the direction perpendicular to the center point of the grid is the direction of application of the pressure.

[0029] The center point of the grid is determined as the point of application of the shear stress, and the direction of the shear stress is perpendicular to the direction of the pressure.

[0030] The pressure and shear stress at the center point in the x, y, and z directions are obtained by simulation or experiment based on the geometric shape data.

[0031] On the other hand, the process of obtaining the operating parameters of the known load train includes:

[0032] Determine the operating parameters and feature parameters corresponding to the geometric shape data;

[0033] The operating parameters include the angle between the train's direction of travel and the wind direction of the surrounding environment, the neighborhood radius, the Reynolds number of the air around the train, the wind speed of the wind around the train, the train's speed, and the air density. The characteristic parameters include the train's reference length, train's reference width, train's reference height, and train's reference area.

[0034] On the other hand, the training process of the prediction model includes:

[0035] The geometric shape data, pressure, shear stress, and operating parameters of the known load train are classified according to working conditions and train type;

[0036] The prediction model is trained using the geometric shape data, pressure, shear stress and operating parameters of the known load train after classification, to obtain the predicted pressure and shear stress output by the prediction model.

[0037] The prediction model is modified using a loss function, the expression of which is:

[0038] ;

[0039] in, Let be the value of the loss function, α be the weight of the pressure, i be the i-th geometric shape data, and n be the total number of geometric shape data. The pressure corresponding to the i-th geometric shape data. Let β be the predicted pressure corresponding to the i-th geometric shape data output by the prediction model, β be the weight of the shear stress, j be the direction of the shear stress, and x, y, and z be the coordinates in the three directions. Let be the shear stress corresponding to the i-th geometric shape data in the j-th direction. The predicted shear stress in the j-direction corresponds to the i-th geometric shape data output by the prediction model.

[0040] On the other hand, determining the resistance, lift, and lateral force of the train under the load to be determined based on the pressure and shear stress of the train under the load to be determined includes:

[0041] The pressure and shear stress of the train with the load to be determined are integrated along the running direction of the train to be determined to obtain the resistance of the train with the load to be determined.

[0042] The relationship between the drag coefficients of the aforementioned resistance is as follows:

[0043] ;

[0044] in, Let A be the drag coefficient, and A be the projected area of ​​the train with the load to be determined. air density, The speed of the train carrying the load to be determined. The surface geometry of the train with the load to be determined. The pressure of the train with the load to be determined. Let be the normal vector of the train with the load to be determined. Let be the vector in the opposite direction to the travel direction of the train with the load to be determined. Let be the shear stress of the train under the load to be determined, and s be the area of ​​the geometric surface mesh element of the train under the load to be determined;

[0045] The pressure and shear stress of the train with the load to be determined are integrated along the height direction of the train to be determined to obtain the lift of the train with the load to be determined.

[0046] The pressure and shear stress of the train under the load to be determined are integrated along the width direction of the train to be determined to obtain the lateral force of the train under the load to be determined.

[0047] To address the aforementioned technical problems, the present invention also provides a device for determining the aerodynamic load of a train, comprising:

[0048] Memory, used to store computer programs;

[0049] A processor is used to implement the steps of the above-described method for determining the aerodynamic load of a train when executing the computer program.

[0050] On the other hand, it also includes:

[0051] An input interface, connected to the processor, is used to acquire externally imported computer programs, parameters, and commands, and save them to the memory under the control of the processor.

[0052] The display unit, connected to the processor, is used to display the data sent by the processor;

[0053] The network port is connected to the processor and is used for communication with external terminal devices.

[0054] To solve the above-mentioned technical problems, the present invention also provides a train and the aforementioned aerodynamic load determination device.

[0055] On the other hand, it also includes a prompting device;

[0056] The prompting device is connected to the aerodynamic load determining device and is used to acquire the drag, lift, and lateral force output by the aerodynamic load determining device, and to provide a prompt when the drag, lift, and lateral force exceed the corresponding threshold.

[0057] This application provides a method, apparatus, and train for determining aerodynamic loads on a train, relating to the field of aerodynamics. The method includes: acquiring geometric data of the train with the load to be determined; inputting the geometric data of the train with the load to be determined into a prediction model to obtain the pressure and shear stress of the train with the load to be determined, output by the prediction model; and determining the drag, lift, and lateral force of the train with the load to be determined based on the pressure and shear stress. By training the prediction model with known load train data, the trained prediction model only requires the input of the geometric file of the train object to be calculated to obtain the pressure and shear stress, without requiring specialized expertise, making it operable by all personnel. By replacing simulation calculations with the prediction model, the prediction model eliminates the need for re-simulation calculations, directly outputting pressure and shear stress, reducing calculation time and improving computational efficiency. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the prior art and embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart of a method for determining the aerodynamic load of a train provided by the present invention;

[0060] Figure 2 A schematic diagram of the structure of a prediction model provided by the present invention;

[0061] Figure 3 This is a schematic diagram of the structure of a train aerodynamic load determination device provided by the present invention. Detailed Implementation

[0062] The core of this invention is to provide a method, device, and train for determining the aerodynamic load of a train. By training a prediction model with data from a train with known loads, the trained prediction model only requires the input of the geometric file of the train object to be calculated to obtain the pressure and shear stress. This does not involve any professional field, so that all staff members have the ability to operate it.

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Figure 1 A flowchart of a method for determining the aerodynamic load of a train provided by the present invention, the method comprising:

[0065] S11: Obtain the geometric shape data of the train with the load to be determined. The geometric shape data is a grid of the train's surface discretized.

[0066] The geometric shape data contains point cloud files of all geometric shape data of the train, stored in STL format. It is known that the train generates a large amount of aerodynamic historical data during operation. This type of data is currently underutilized, but has great potential for development. Therefore, this application uses standardized cleaning methods to form a standard dataset from a large amount of dormant data in the industry. The standard dataset is then used in combination with geometric neural operators and Fourier neural operators for model training.

[0067] The geometric shape data of a train is not a simple regular geometric body; it includes streamlined curved surfaces at the front, body waistlines, door and window recesses, rear diffuser sections, bogies, and other curved surfaces. By dividing the geometric shape data into multiple grids, analyzing each grid individually, and then summing the results, the stress analysis problem of the entire train under a given load can be transformed into a stress analysis of multiple smaller grids. Smaller grids can more accurately determine the forces on the train's curved surfaces, reducing the probability of distortion.

[0068] S12: Input the geometric shape data of the train with the load to be determined into the prediction model to obtain the pressure and shear stress of the train with the load to be determined output by the prediction model. The prediction model is obtained by training the prediction model using the known geometric shape data, pressure, shear stress and operating parameters of the train with the load. The operating parameters include the train's operating speed.

[0069] The cell area, pressure data, and surface shear stress data of each point in the point cloud are stored in a CSV file. This file must include the cell area of ​​each point in the point cloud (the area projected onto the cell in the x, y, and z directions must also be given), the three coordinates of each point, the surface pressure at each point, and the surface shear stress in the x, y, and z directions at each point.

[0070] The various operational parameters and train-specific characteristic parameters of the train are stored in JSON files. The operational parameter files include train speed, air density, ambient wind speed, and environmental risk. The train-specific characteristic parameters include characteristic length, characteristic height, characteristic width, and reference area. It should be noted that all of these operational parameters affect the stress on the train during operation. Therefore, incorporating these parameters into model training allows the model to more accurately output pressure and shear stress.

[0071] By using the coordinates of the center point and the force at the center point in each grid (point cloud) as training data for the prediction model, the pressure and shear stress can be directly obtained from the geometric shape data of the train under the load after the prediction model training is completed. This eliminates the need for simulation calculations of the train under the load, greatly reducing the computation time.

[0072] S13: Determine the resistance, lift, and lateral force of the train with the load to be determined based on the pressure and shear stress of the train with the load to be determined;

[0073] Using the standardized dataset described above, large-scale deep learning and training were performed using geometric neural operators and Fourier neural operators. The trained model only requires inputting the geometric shape data of the train under load into the prediction model to obtain the pressure and shear stresses.

[0074] Surface pressure is the normal force exerted by air on the train surface, perpendicular to the surface. Shear stress is the tangential force exerted by air on the train surface, parallel to the surface, caused by air viscosity. Drag, lift, and lateral force are the components of the total surface force in specific directions, such as the direction of travel, the vertical direction, and the lateral direction.

[0075] The train surface is divided into several grids. The pressure and shear stress components in the target direction are calculated for each element, and then summed over all elements to obtain the total force. The surface force of each element is divided into the normal force generated by pressure and the tangential force generated by shear stress. These two forces are then projected onto the target direction. The components of all grid elements are summed to obtain the train's total drag, lift, and lateral force.

[0076] This application provides a method for determining the aerodynamic load of a train, relating to the field of aerodynamics. The method includes: inputting the geometric shape data of the train with the load to be determined into a prediction model to obtain the pressure and shear stress of the train with the load to be determined, output by the prediction model; and determining the drag, lift, and lateral force of the train with the load to be determined based on the pressure and shear stress. By training the prediction model with known load train data, the trained prediction model only requires the input of the geometric file of the train object to be calculated to obtain the pressure and shear stress, without requiring specialized expertise, making it operable by all personnel. By replacing simulation calculations with the prediction model, the prediction model eliminates the need for re-simulation calculations, directly outputting pressure and shear stress, reducing calculation time and improving computational efficiency.

[0077] Based on the above embodiments:

[0078] In some embodiments, the process of acquiring the geometric shape data of a known-load train includes:

[0079] The surface of the train under known load is divided into multiple discrete grids. Each grid is a triangular grid consisting of three vertices, and the side length of each triangular grid is positively correlated with the length of the train.

[0080] The mesh is corrected to an equilateral triangular mesh;

[0081] Determine the three coordinates of the center point of each grid and the projected area of ​​each grid in the x, y and z directions.

[0082] The train geometry file is an STL file, which discretizes the train surface into a triangular mesh file that approximates a point cloud. Each triangular mesh is approximately an equilateral triangle with a side length of 0.0024H~0.148H, where H is the characteristic length of the train, usually referring to the train's height or width. After the initial discretization of the train surface, each triangular mesh is checked and subjected to denoising, normalization, and coordinate alignment to correct and approximate each triangular mesh as an equilateral triangle. In the triangular mesh generation of the train surface, it is required that the triangles be as close to equilateral triangles as possible. The core purpose is to ensure the accuracy, stability, and convergence of numerical calculations and to avoid introducing additional calculation errors due to mesh shape distortion, resulting in the final train geometry file containing discrete point clouds and elements. The three coordinates of the center point of each triangular mesh, the projected area of ​​each triangular mesh in the x, y, and z directions, and the above four sets of data are read and recorded.

[0083] In some embodiments, after dividing the surface of the known-loaded train into multiple discrete grids, the method further includes:

[0084] Delete grids that intersect or overlap;

[0085] Delete grid cells whose area is outside the preset area range;

[0086] Delete grid cells whose interior angles are outside the preset angle range;

[0087] The mesh is corrected, including:

[0088] Correct the mesh after the deletion operation.

[0089] Remove triangular elements from the triangular mesh that are of poor quality or do not reproduce the original train shape well. For example, triangular elements that intersect or overlap with each other, or triangular elements that do not reproduce the train shape well or reproduce the shape poorly. These elements will cause shape reshaping deviations and cause large errors in subsequent training and inference.

[0090] Check for intersecting / interlacing triangles: By judging the spatial relationship of triangular facets, if the projections of two triangles overlap and their normal vectors contradict each other, they are considered intersecting units. Intersecting or interlacing triangles will disrupt the continuity of the train's shape. Calculate the triangle's aspect ratio (longest side / shortest height), interior angle size (whether there are acute angles less than 15° or obtuse angles greater than 165°), and area threshold (triangles with excessively large or small areas are usually caused by point cloud noise). Filter the triangle shape by interior angle size; triangles with excessively small interior angles are identified as narrower triangles. Extremely narrow acute-angled triangles or excessively large flat triangles will result in a rough model surface and blurred contours, failing to match the curved surface features of a real train.

[0091] In some embodiments, after dividing the surface of the known-loaded train into multiple discrete grids, the method further includes:

[0092] The origin of the x-direction coordinate is taken as the center of the length of the train with known load, and the front of the train is facing -x. The origin of the z-direction coordinate is taken as the track surface of the train with known load, and the height of the train is +z. The y-direction is determined based on the x-direction, z-direction and right-hand rule.

[0093] The grid is normalized and aligned according to the x, y, and z coordinates obtained from the division.

[0094] The mesh is corrected, including:

[0095] The grid after coordinate normalization and alignment is corrected.

[0096] This primarily addresses the train's external shape, formed by assembling different components. Since components may use different coordinate systems and origins during modeling, positioning deviations can occur during assembly. Therefore, when acquiring training data, it's agreed that the center of the train formation length is the X-axis origin, the train's nose facing -X and the track surface are the Z-axis origins, the train height is +Z, and the Y-axis is determined by the right-hand rule. All components are then normalized and aligned according to this coordinate system and origin to ensure error-free train assembly, mapping the train mesh's 3D coordinates to a unified numerical range and eliminating numerical differences caused by the train's actual dimensions and scaling.

[0097] In some embodiments, the mesh is modified, including:

[0098] Adjust the position of the three vertices of the mesh to change its shape;

[0099] And / or merge multiple adjacent grids with side lengths below the side length threshold into a polygon, and redivide the polygon into multiple equilateral triangles;

[0100] And / or for a quadrilateral formed by two adjacent grids, if the side lengths of the two grids generated by the current diagonal are both below the side length threshold, then the diagonal is flipped, and the quadrilateral is reconstructed into two triangles.

[0101] Referring to the denoising section, quality repair is performed on elements with overlapping quality. Standard triangular elements should be as close as possible to equilateral triangles; the aim is to make each triangular element approximate an equilateral triangle through correction. Common correction methods include:

[0102] (1) Change the shape of the triangle by adjusting the position of the three vertices of the triangle unit; fine-tune the three-dimensional coordinates of the triangle vertices, and optimize the shape of a single triangle to make it closer to equilateral without changing the mesh topology (connection relationship of points, edges and faces).

[0103] (2) Reconstruct all triangular units in a certain area and improve the shape by merging and re-dividing the triangular units; for example, merge adjacent narrow triangles into quadrilaterals or polygons, and then re-divide them to change their interior angles.

[0104] (3) Eliminate the narrow triangular units between units by changing the form of the nodes between units. For example, in quadrilateral ABCD, the original diagonal AC is divided into △ABC and △ADC, both of which are narrow triangles. By flipping the diagonal to BD, it is divided into △ABD and △BCD, which can be closer to equilateral.

[0105] In some embodiments, the process of obtaining the pressure and shear stress of the known load train includes:

[0106] The center point of a discrete grid divided into multiple grids on the surface of a train with a known load is determined as the point of application of the pressure, and the direction perpendicular to the center point of the grid is the direction of application of the pressure.

[0107] The center point of the grid is determined as the point of application of shear stress, and the direction of shear stress is perpendicular to the direction of pressure.

[0108] Obtain geometric shape data, and then use simulation or experimentation to determine the pressure and shear stress at the center point in the x, y, and z directions.

[0109] Train aerodynamic data (typically surface pressure and shear stress) obtained from simulation calculations, wind tunnel tests, or actual vehicle track tests are assigned to each triangular mesh (element) in the train's geometric shape STL file. The data is standardized to construct point-level feature vectors, establishing a one-to-one mapping between point features and STL point clouds. The point-level feature vectors are constructed by assigning pressure and shear stress vector components to each point, based on the center point of each triangular mesh. Both pressure and shear stress are vectors. The point of application of pressure is the center point of each triangular mesh, and its direction is perpendicular to the triangular mesh at that center point (i.e., the normal direction). The point of application of shear stress is the center point of each triangular mesh, and its direction is perpendicular to the pressure direction at that center point (i.e., the tangential direction). Since each triangular mesh is discretized from the complex shape of the train, the normal and tangential directions at the center point of each element are different. Such a large number of directions make it impossible to describe the pressure and shear stress in the same coordinate direction. Therefore, the pressure and shear stress at each node are uniformly decomposed into vectors according to the overall coordinate system of the train. In this way, the left and right pressure and shear stress components are all along the same xyz coordinate axis, which greatly improves the convenience and operability when performing calculations.

[0110] The train surface pressure and shear stress obtained from actual calculations, wind tunnel tests, or actual vehicle track tests are decomposed into vector projections in the x, y, and z directions according to the coordinate system in the STL file, and then corresponded one-to-one with each point in the STL point cloud file, so that each point has corresponding pressure and shear stress in the x, y, and z directions. That is, each point in the point cloud has 3 pressure values ​​and 3 shear stress values.

[0111] Record the three coordinates of the center point of each triangular mesh in the STL file, the projected area of ​​each triangular mesh in the x, y, and z directions, and the three pressure values ​​and three shear stress values ​​at each point. Record these 12 values ​​in a list to obtain a CSV file.

[0112] In some embodiments, the process of obtaining the operating parameters of a known load train includes:

[0113] Determine the operating parameters and characteristic parameters corresponding to the geometric shape;

[0114] The operating parameters include the angle between the train's direction of travel and the wind direction of the surrounding environment, the neighborhood radius, the Reynolds number of the air around the train, the wind speed of the wind around the train, the train's speed, and the air density. The characteristic parameters include the train's reference length, reference width, reference height, and reference area.

[0115] Edit the JSON file to define the boundary conditions required for training, including: train speed, ambient wind speed and direction, and the reference area and reference length for calculating aerodynamic and moment coefficients. A complete JSON file is shown below:

[0116] {"length":25,

[0117] "width":3.36,

[0118] "height":4.05,

[0119] "slant":5,

[0120] "radius":2.1,

[0121] "re":13817594,

[0122] "wind velocity":11,

[0123] "velocity":97.22,

[0124] "reference_area":11.94,

[0125] "density":1.225,}

[0126] In the above records, "length" refers to the train's reference length, "width" refers to the train's reference width, "height" refers to the train's reference height, "slant" refers to the angle between the train's direction of travel and the surrounding wind direction, "re" refers to the Reynolds number of the air around the train, "wind velocity" refers to the wind speed around the train, "velocity" refers to the train's speed, "reference_area" refers to the train's reference area, and "density" refers to the air density.

[0127] The angle between the train's running direction and the ambient wind direction is a key parameter for calculating lateral forces. The Reynolds number of the air surrounding the train characterizes the airflow state. The vector superposition of the ambient wind speed and the train's running speed yields the airflow velocity relative to the train, which is the dynamic basis for calculating aerodynamic loads. The train's running speed determines the airflow velocity relative to the train, which is the most critical parameter for aerodynamic load calculations. Air density reflects the mass inertia of air. Characteristic parameters are standardized representations of the train's inherent geometric properties, transforming the train's actual geometric dimensions into dimensionless parameters (such as Reynolds number, lift coefficient, and drag coefficient). This enables comparisons of the aerodynamic performance of different train models, generalized load calculations, and also determines the basic aerodynamic characteristics of the train's aerodynamic shape.

[0128] Figure 2 A schematic diagram of the structure of a prediction model provided by the present invention;

[0129] In some embodiments, the training process of the prediction model includes:

[0130] The geometric shape data, pressure, shear stress and operating parameters of the known load train are classified according to the working conditions and train type;

[0131] The prediction model is trained using the geometric shape data, pressure, shear stress and operating parameters of the known load train after classification, and the predicted pressure and shear stress output by the prediction model are obtained.

[0132] The prediction model is modified using a loss function, which is expressed as follows:

[0133] ;

[0134] in, Let be the value of the loss function, α be the weight of the pressure, i be the i-th geometric shape data, and n be the total number of geometric shape data. The pressure corresponding to the i-th geometric shape data. Let β be the predicted pressure corresponding to the i-th geometric shape data output by the prediction model, β be the weight of the shear stress, j be the direction of the shear stress, and x, y, and z be the coordinates in the three directions. Let be the shear stress corresponding to the i-th geometric shape data in the j-th direction. The predicted shear stress in the j-direction corresponds to the i-th geometric shape data output by the prediction model.

[0135] The structure of the large simulation model is as follows Figure 2 As shown, the model input consists of the center coordinates and area of ​​the high-speed train's geometric CFD (Computational Fluid Dynamics) grid, and the distance function (DF) of uniformly sampled points in the 3D space of the high-speed train's circumscribed cuboid. The output is the surface pressure and surface shear force corresponding to the CFD grid. The simulation model is divided into three network modules: two graph neural operator modules and one Fourier neural operator module. First, the first GNO (Graph Neural Operator) module maps the irregular high-speed train geometry to a regular 3D latent space. Then, the FNO (Fourier Neural Operator) module performs a fast Fourier transform on the latent space data, learns in the frequency domain, and then performs an inverse Fourier transform back to the latent space. Finally, the second GNO module maps the latent space data to the pressure and surface shear force distribution of each grid in the high-speed train geometry.

[0136] In some embodiments, determining the drag, lift, and lateral force of the train with the load to be determined based on the pressure and shear stress of the train with the load to be determined includes:

[0137] The pressure and shear stress of the train with the unknown load are integrated along the direction of travel of the train with the unknown load to obtain the resistance of the train with the unknown load.

[0138] The relationship between drag coefficient and resistance is as follows:

[0139] ;

[0140] in, Let A be the drag coefficient, and A be the projected area of ​​the train with the load to be determined. air density, To determine the speed of the train carrying the load, The surface geometry of the train with the load to be determined. The pressure of the train to be loaded is to be determined. Let the normal vector of the train with the load to be determined be... Let be the vector in the opposite direction to the direction of travel of the train with the load to be determined. Let s be the shear stress of the train under load to be determined, and s be the area of ​​the geometric surface mesh element of the train under load to be determined.

[0141] The pressure and shear stress of the train with the load to be determined are integrated along the height direction of the train to be determined to obtain the lift of the train with the load to be determined.

[0142] The pressure and shear stress of the train under the load to be determined are integrated along the width direction of the train to be determined to obtain the lateral force of the train under the load to be determined.

[0143] After predicting the surface pressure and shear force distribution, the drag coefficient of the high-speed train is obtained by integrating the pressure and shear force on the geometric surface of the train. Similarly, the same calculation method can be used when calculating lift and lateral force, and this application does not impose further limitations here.

[0144] It should also be noted that the pressure and shear stress provided in this application can be applied to calculations in other ways, including but not limited to drag, lift, and lateral force.

[0145] It should also be noted that to associate the JSON file with the corresponding STL and CSV files to form a complete sample, there are two methods for association:

[0146] (1) When uploading data, upload the JSON file along with the corresponding STL and CSV files together, and the binding is formed in one upload;

[0147] (2) When uploading STL and CSV files on the data upload page, enter the information of the JSON file on the interface, and the system will automatically generate the JSON file according to the input information to form data binding.

[0148] Boundary conditions are used as part of the model input for training large models.

[0149] The large model takes STL, CSV, and JSON files as input. A set of STL, CSV, and JSON files constitutes a dataset. A large dataset serves as the input for training the large model; it can be understood as the textbook for the model's learning process. After learning from these datasets, the large model gains the ability to predict the aerodynamic performance of new train shapes. When this predictive ability meets the accuracy requirements (currently, the required accuracy is a deviation of no more than ±5% from experimental values), the large model is considered usable. When a user inputs a train shape, the large model can predict the aerodynamic performance parameters of that shape.

[0150] Each set of data (stl, csv, json) is encapsulated into a uniform data sample;

[0151] A structured directory management system is adopted, such as classification by operating condition, train type, and simulation operating condition;

[0152] Operating conditions typically include single-car operation on open tracks, single-car operation on open tracks in strong winds, trains passing each other on open tracks, and train operation inside tunnels.

[0153] The aerodynamic performance parameters of the trains under the above operating conditions vary in different ways. However, even for different trains under the same operating condition, the characteristics of their aerodynamic performance parameters under the same operating condition are similar. Therefore, they can be classified according to the operating conditions. The datasets and configuration parameters used in the training of large models are similar and can be treated uniformly according to the classification.

[0154] It provides a data loading interface that supports data reading from mainstream frameworks such as PyTorch and TensorFlow, is compatible with different programming languages, and is suitable for all types of programmers and users.

[0155] Figure 3 This is a schematic diagram of the structure of an aerodynamic load determination device for a train provided by the present invention. The aerodynamic load determination device for a train includes:

[0156] Memory 31 is used to store computer programs;

[0157] The processor 32 is used to implement the steps of the above-described method for determining the aerodynamic load of a train when executing a computer program.

[0158] The method for determining the aerodynamic load of a train provided in this application is described in the above embodiments and will not be repeated here.

[0159] In some embodiments, it also includes:

[0160] The input interface, connected to the processor, is used to acquire externally imported computer programs, parameters, and commands, and saves them to memory under the control of the processor.

[0161] The display unit, connected to the processor, is used to display the data sent by the processor;

[0162] The network port is connected to the processor and is used for communication with various external terminal devices.

[0163] This input interface can be connected to an input device to receive parameters or commands manually entered by the user. This input device can be a touch layer covering the display screen, or it can be a button, trackball, or touchpad located on the terminal casing.

[0164] The display unit can be an LCD screen or an e-ink screen, etc.

[0165] The communication technology used in this connection can be wired or wireless, such as Mobile High Definition Link (MHL), Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), Wireless Fidelity (WiFi), Bluetooth, Bluetooth Low Energy, or IEEE 802.11s-based communication technologies.

[0166] The present invention also provides a train including the above-described aerodynamic load determination device.

[0167] In some embodiments, a prompting device is also included;

[0168] The alerting device is connected to the aerodynamic load determination device to obtain the drag, lift, and lateral force output by the aerodynamic load determination device, and to provide an alert when the drag, lift, and lateral force exceed the corresponding threshold.

[0169] The train described in this application is based on the above embodiments and will not be repeated here.

[0170] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0171] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0172] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining the aerodynamic load of a train, characterized in that, include: Obtain the geometric shape data of the train with the load to be determined, wherein the geometric shape data is a grid discretized from the surface of the train; The geometric shape data of the train with the load to be determined is input into the prediction model to obtain the pressure and shear stress of the train with the load to be determined output by the prediction model. The prediction model is obtained by training the prediction model using the known geometric shape data, pressure, shear stress and operating parameters of the train with the load. The operating parameters include the train's operating speed. The resistance, lift, and lateral force of the train under the load to be determined are determined based on the pressure and shear stress of the train under the load to be determined.

2. The method for determining the aerodynamic load of a train as described in claim 1, characterized in that, The process of acquiring the geometric shape data of the known load train includes: The surface of the known load train is divided into multiple discrete grids, each grid being a triangular grid consisting of three vertices, and the side length of each triangular grid is positively correlated with the length of the train; The mesh is modified to form an equilateral triangular mesh; Determine the three coordinates of the center point of each grid and the projected area of ​​each grid in the x, y and z directions.

3. The method for determining the aerodynamic load of a train as described in claim 2, characterized in that, After dividing the surface of the known-loaded train into multiple discrete grids, the process further includes: Delete grids that intersect or overlap; Delete grid cells whose area is outside the preset area range; Delete grid cells whose interior angles are outside the preset angle range; The mesh is corrected, including: Correct the mesh after the deletion operation.

4. The method for determining the aerodynamic load of a train as described in claim 2, characterized in that, After dividing the surface of the known-loaded train into multiple discrete grids, the process further includes: The origin of the x-direction coordinate is taken as the length center of the known load train, with the train head facing -x. The origin of the z-direction coordinate is taken as the track surface of the known load train, with the height direction being +z. The y-direction is determined based on the x-direction, the z-direction, and the right-hand rule. The grid is normalized and aligned according to the x, y, and z coordinates obtained from the division. The mesh is corrected, including: The grid after coordinate normalization and alignment is corrected.

5. The method for determining the aerodynamic load of a train as described in claim 2, characterized in that, The mesh is corrected, including: Adjust the positions of the three vertices of the mesh to change the shape of the mesh; And / or merge multiple adjacent grids with side lengths below a side length threshold into a polygon, and redivide the polygon into multiple equilateral triangles; And / or for a quadrilateral formed by two adjacent grids, if the side lengths of the two grids generated by the current diagonal are both lower than the side length threshold, then the diagonal is flipped, and the quadrilateral is reconstructed into two triangles.

6. The method for determining the aerodynamic load of a train as described in claim 1, characterized in that, The process of obtaining the pressure and shear stress of the known load train includes: The center point of a plurality of discrete grids into which the surface of the known load train is divided is determined as the point of application of the pressure, and the direction perpendicular to the center point of the grid is the direction of application of the pressure. The center point of the grid is determined as the point of application of the shear stress, and the direction of the shear stress is perpendicular to the direction of the pressure. The pressure and shear stress at the center point in the x, y, and z directions are obtained by simulation or experiment based on the geometric shape data.

7. The method for determining the aerodynamic load of a train as described in claim 1, characterized in that, The process of obtaining the operating parameters of the known load train includes: Determine the operating parameters and feature parameters corresponding to the geometric shape data; The operating parameters include the angle between the train's direction of travel and the wind direction of the surrounding environment, the neighborhood radius, the Reynolds number of the air around the train, the wind speed of the wind around the train, the train's speed, and the air density. The characteristic parameters include the train's reference length, train's reference width, train's reference height, and train's reference area.

8. The method for determining the aerodynamic load of a train as described in claim 1, characterized in that, The training process of the prediction model includes: The geometric shape data, pressure, shear stress, and operating parameters of the known load train are classified according to working conditions and train type; The prediction model is trained using the geometric shape data, pressure, shear stress and operating parameters of the known load train after classification, to obtain the predicted pressure and shear stress output by the prediction model. The prediction model is modified using a loss function, the expression of which is: ； in, Let be the value of the loss function, α be the weight of the pressure, i be the i-th geometric shape data, and n be the total number of geometric shape data. The pressure corresponding to the i-th geometric shape data. Let β be the predicted pressure corresponding to the i-th geometric shape data output by the prediction model, β be the weight of the shear stress, j be the direction of the shear stress, and x, y, and z be the coordinates in the three directions. Let be the shear stress corresponding to the i-th geometric shape data in the j-th direction. The predicted shear stress in the j-direction corresponds to the i-th geometric shape data output by the prediction model.

9. The method for determining the aerodynamic load of a train as described in any one of claims 1 to 8, characterized in that, Determining the resistance, lift, and lateral force of the train under the load to be determined based on the pressure and shear stress of the train to be determined includes: The pressure and shear stress of the train with the load to be determined are integrated along the running direction of the train to be determined to obtain the resistance of the train with the load to be determined. The relationship between the drag coefficients of the aforementioned resistance is as follows: ； in, Let A be the drag coefficient, and A be the projected area of ​​the train with the load to be determined. air density, The speed of the train carrying the load to be determined. The surface geometry of the train with the load to be determined. The pressure of the train with the load to be determined. Let be the normal vector of the train with the load to be determined. Let be the vector in the opposite direction to the travel direction of the train with the load to be determined. Let be the shear stress of the train under the load to be determined, and s be the area of ​​the geometric surface mesh element of the train under the load to be determined; The pressure and shear stress of the train with the load to be determined are integrated along the height direction of the train to be determined to obtain the lift of the train with the load to be determined. The pressure and shear stress of the train under the load to be determined are integrated along the width direction of the train to be determined to obtain the lateral force of the train under the load to be determined.

10. An aerodynamic load determination device for a train, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for determining the aerodynamic load of a train as described in any one of claims 1 to 9.

11. The aerodynamic load determination device for a train as described in claim 10, characterized in that, Also includes: An input interface, connected to the processor, is used to acquire externally imported computer programs, parameters, and commands, and save them to the memory under the control of the processor. The display unit, connected to the processor, is used to display the data sent by the processor; The network port is connected to the processor and is used for communication with external terminal devices.

12. A train, characterized in that, Includes the aerodynamic load determination device as described in claim 10 or 11.

13. The train as described in claim 12, characterized in that, It also includes a prompting device; The prompting device is connected to the aerodynamic load determining device and is used to acquire the drag, lift, and lateral force output by the aerodynamic load determining device, and to provide a prompt when the drag, lift, and lateral force exceed the corresponding threshold.

Images