Digital twin simulation system for sleeper rail maintenance cycle

Abstract

The present invention relates to the field of digital twin simulation technology, specifically to a digital twin simulation system for the sleeper rail maintenance cycle, the system comprising: a multi-source data integration module, integrating laser scanner point cloud data with drone photogrammetry image sequences, extracting the rail profile point coordinates and sleeper position geometric parameters of the sleeper rail, and generating an integrated structural attribute set. In the present invention, by integrating laser scanner point cloud data with drone photogrammetry image sequences, the spatial coordinates of the rail profile and the sleeper geometric parameters are obtained, and a unified spatial coordinate system is constructed, thereby improving the accuracy and stability of data matching; and using the displacement vector and posture difference between the measured data and the digital model as the basis for adjustment, the adjustment operation of the rail node coordinates and the sleeper rotation matrix is made more consistent with the actual track facility status, achieving consistency between the simulated geometric structure and the actual track structure, and improving the credibility of the model simulation analysis.

Description

Technical Field

[0001] The present invention relates to the field of digital twin simulation technology, and in particular to a digital twin simulation system for sleeper rail maintenance cycles. Background Art

[0002] The field of digital twin simulation technology involves creating a complete digital mirror model of a real-world physical entity or system in a virtual space through information collection, data fusion, and dynamic modeling. This allows for real-time monitoring, simulation analysis, and prediction of the physical system. However, existing technologies generally use a single model with limited sophistication, which is insufficient in allocating computing resources across regions. This can lead to reduced simulation efficiency and increased computational overhead. Therefore, improvements are needed. Summary of the Invention

[0003] The purpose of the present invention is to solve the shortcomings of the prior art and propose a digital twin simulation system for the sleeper rail maintenance cycle.

[0004] In order to achieve the above objectives, the present invention adopts the following technical solutions: The digital twin simulation system for the sleeper rail maintenance cycle includes:

[0005] A multi-source data integration module integrates laser scanner point cloud data and drone photogrammetry image sequences to extract rail profile point coordinates and sleeper position geometric parameters, generating an integrated structural attribute set.

[0006] A geometric consistency calibration module calculates the spatial displacement vector between the reference point of the rail component corresponding to the digital twin model and the measured coordinates based on the preset geometric form of the digital twin model and the integrated structural attribute set, obtains a geometric deviation metric element, and adjusts the rotation matrix of the corresponding rail node coordinates and the sleeper component in the digital twin model based on the geometric deviation metric element, maps the geometric form changes, and generates a calibration geometric representation volume;

[0007] A topology evolution module retrieves the calibrated geometric representation and, in combination with maintenance records during the sleeper-rail maintenance cycle, determines changes in component connection relationships caused by sleeper replacement or addition or removal of fasteners, forms a connection change instruction set, and, based on the connection change instruction set, reconstructs the constraints between the rails and sleepers to construct an evolutionary topology network diagram.

[0008] The adaptive collaborative simulation module receives the evolutionary topology network diagram and the simulation analysis requirements of the sleeper rail maintenance cycle, selects matching track unit models based on the identified stress areas, combines them into a mixed-fidelity model configuration, performs simulation operations based on the mixed-fidelity model configuration, and obtains adaptive simulation output results.

[0009] Preferably, the steps of obtaining the integrated structure attribute set are:

[0010] Based on the rail surface point cloud data collected by the laser scanner and the image sequence obtained by drone aerial photography, the three-dimensional coordinates of the rail edge contour points and the geometric position parameters of the center point of the sleeper cross section in each frame are extracted. After uniformly projecting them into the WGS-84 coordinate system, all acquisition timestamps are aligned. The corresponding frame sequences are indexed and matched one by one to form a set of joint feature point pairs, resulting in a list of aligned joint feature point pairs.

[0011] Calculating a rail profile curve registration factor according to the aligned joint feature point pair list;

[0012] According to the rail profile curve registration factor and the three-dimensional offset value of each pair of feature points, point pairs with offset values greater than the rail profile curve registration factor are screened, and the remapping vector is recalculated and the coordinates are replaced in the global coordinate system. At the same time, the geometric parameters of the center point of the sleeper cross section are corrected by coordinate fitting. The remapped point coordinates and the corrected geometric parameters are combined to generate an integrated structural attribute set.

[0013] Preferably, the steps for obtaining the geometric deviation metric are:

[0014] Based on the standard geometric forms of rails and sleepers preset in the digital twin model, the coordinates of the rail edge contour points and the geometric parameters of the center point of the sleeper cross section in the integrated structural attribute set are called, and the corresponding rail component reference points and sleeper component center points in the digital twin model are matched to form a spatial point pair with a one-to-one correspondence between the model and the measured points;

[0015] Based on the coordinates of each pair of rail component reference points and the measured rail edge contour points in the spatial point pairs, the coordinate differences of the point pairs along the X, Y, and Z axes in three-dimensional space are calculated one by one, a complete three-dimensional spatial displacement vector is synthesized and the vector length is recorded, and the average spatial displacement vector length of the point pair set is calculated at the same time to generate a rail spatial displacement vector set;

[0016] Based on the geometric parameters of the center points of each pair of sleeper components in the spatial point pair and the center point of the measured sleeper cross section, the Euler angle difference value of the rotational posture change between the two points is extracted, and the posture angle differences in pitch angle, yaw angle and roll angle are recorded respectively to form a sleeper posture angle difference set, and the rail spatial displacement vector set and the sleeper posture angle difference set are combined to obtain a geometric deviation measurement element.

[0017] Preferably, the step of obtaining the calibration geometric representation is:

[0018] Based on the geometric deviation metric, each spatial displacement vector in the rail spatial displacement vector set is extracted, and the displacement components of the spatial displacement vector on the three coordinate axes of X, Y, and Z are analyzed respectively. For each corresponding rail node coordinate in the digital twin model, the analyzed displacement component values of each axis are superimposed on the original coordinate value of the node to form an adjusted rail node coordinate sequence;

[0019] Based on the adjusted rail node coordinate sequence, calling the sleeper attitude angle difference set in the geometric deviation measurement unit, analyzing the pitch angle difference, yaw angle difference and roll angle difference of each sleeper in the set one by one, recalculating the rotation matrix parameters of each sleeper component, and generating a revised sleeper rotation matrix sequence;

[0020] Based on the adjusted rail node coordinate sequence and the corrected sleeper rotation matrix sequence, the rail node coordinate sequence and the sleeper rotation matrix sequence are applied one by one to the original geometric configuration of the rails and sleepers in the digital twin model, and the three-dimensional morphological mapping is re-performed to update and form a calibrated geometric representation.

[0021] Preferably, the steps of obtaining the connection change instruction set are:

[0022] Retrieving the rail node coordinates and the sleeper component rotation matrix in the calibration geometric representation, and identifying the adjacency information and connection type identification in the original connection relationship between each node and each sleeper one by one, extracting all record sentences with the keywords "replacement", "removal" and "additional addition" in combination with the maintenance record text content during the sleeper-rail maintenance cycle, and parsing the associated sleeper numbers and fastener numbers to form an initial list of structural connection changes;

[0023] Based on the initial list of structural connection changes, the parsed sleeper numbers and fastener numbers are compared one by one with the current connection status in the calibrated geometric representation according to the time period and spatial position interval corresponding to each sleeper number. If there is a sleeper number in the list but not in the calibrated geometric representation, it is determined to be a newly added sleeper connection; if there is a fastener number in the calibrated geometric representation but not in the list, it is determined to be a fastener removal, and a list of component connection relationship changes is generated;

[0024] According to the component connection relationship change list, a standardized instruction text format is constructed for all connection relationships marked as added or removed, and the instruction type, target number, operation node index and impact range identifier are supplemented in sequence to generate a connection change instruction set.

[0025] Preferably, the steps of obtaining the evolutionary topology network diagram are:

[0026] Based on the connection change instruction set, the target number, operation node index and impact range identifier in each connection change instruction are parsed in sequence, the rail node coordinates and sleeper component index in the calibration geometric representation are called, and it is determined whether the target number corresponds to an existing node set. If it is a new instruction, the number is added to the node set and an initial connection entry is constructed. If it is a removal instruction, all connection entries associated with the number are deleted to obtain a structural connection reconstruction node table;

[0027] According to the structural connection reconstruction node table, the connection status between each rail node coordinate and the corresponding sleeper component is scanned to extract all node pairs with physical connection relationships, and the connection orientation attributes are calculated according to the rotation matrix state of the sleeper component to form a constraint relationship reconstruction entry table;

[0028] Based on the constraint relationship, the entry table is reconstructed, and an adjacency list structure with node number as the main index is established one by one. The connection status, connection direction and connection label are embedded in the corresponding adjacency entry as edge attributes, and the edge index table is constructed simultaneously to generate an evolutionary topology network diagram.

[0029] Preferably, the steps of obtaining the hybrid-fidelity model configuration are:

[0030] Based on the node structure relationship in the evolutionary topological network graph, the connection edge set between all node pairs is extracted, and the node number and spatial distribution position associated with each connection edge are identified one by one. The structural fatigue monitoring parameter table provided in the sleeper rail maintenance cycle simulation analysis requirements is matched, and the regional edge set whose connection edges fall into the fatigue risk identification interval is screened to generate a stress area boundary index list;

[0031] Calculating a track element model adaptation factor for each stress region based on the stress region boundary index list;

[0032] Based on the adaptation factor of the track unit model corresponding to each stress area, the adaptation factor of the track unit model is compared with the adaptation value benchmark threshold item by item. If the adaptation factor of the track unit model corresponding to the area is greater than or equal to the adaptation value benchmark threshold, the area is designated as a track unit model with complete parameter configuration; otherwise, it is designated as a track unit model with compressed parameter configuration. The two types of track unit models are integrated into a hybrid fidelity model configuration of the overall track system.

[0033] Preferably, the steps of obtaining the adaptive simulation output result are:

[0034] Based on each area delineated by the hybrid fidelity model configuration, the track unit model with full parameter configuration or parameter compression configuration is called area by area, the current load conditions and material parameters of each track unit model in the area are loaded respectively, dynamic force analysis and calculation are performed, and the stress data series of each regional node are obtained in real time to generate a real-time regional node stress data set;

[0035] Based on the real-time stress data set of the regional nodes, the stress value change difference within consecutive time steps in each region is calculated and divided by the corresponding time interval to obtain the stress gradient change rate of each region. The calculated stress gradient change rate is compared with a preset stress gradient change rate threshold for each region, and abnormal regions whose change rate exceeds the preset limit are identified to form an abnormal stress region identification list;

[0036] Based on the abnormal stress area identification list, for each abnormal area in the list, the track unit model originally configured as a parameter compression configuration is upgraded to a complete parameter configuration, the track unit model originally configured as a complete parameter configuration is parameter refined, the stress data is re-updated, and the adaptive simulation output results are generated.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are:

[0038] In the present invention, by integrating laser scanner point cloud data and drone photogrammetry image sequences, the spatial coordinates of the rail profile and the geometric parameters of the sleepers are obtained, and a unified spatial coordinate system is constructed, so that the accuracy and stability of data matching are improved; and the displacement vector and posture difference between the measured data and the digital model are used as the adjustment basis, so that the adjustment operation of the rail node coordinates and the sleeper rotation matrix is more in line with the actual track facility status, the consistency of the simulated geometric structure and the actual track structure is achieved, and the credibility of the model simulation analysis is improved; then, based on the actual maintenance records of the maintenance cycle, the changes in component connections caused by sleeper replacement and addition and removal of fasteners are identified, and the structural constraint relationship between the rails and sleepers is reconstructed in time, so that the topological network reflects the actual status changes of the facilities in real time, and the dynamic evolution ability of the topological model is improved; at the same time, the model granularity and fidelity are adjusted according to the identified stress areas, the complexity and computational load of the simulation model are controlled, and the computational efficiency is improved while ensuring the simulation accuracy, and the simulation resource requirements are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] See also Figure 1 The present invention provides a technical solution: a digital twin simulation system for the sleeper rail maintenance cycle includes:

[0042] A multi-source data integration module integrates laser scanner point cloud data and drone photogrammetry image sequences to extract rail profile point coordinates and sleeper position geometric parameters, generating an integrated structural attribute set.

[0043] The geometric consistency calibration module calculates the spatial displacement vector between the reference point of the rail component in the digital twin model and the measured coordinates based on the preset geometric shape and integrated structural attribute set of the digital twin model, obtains the geometric deviation measurement element, and adjusts the rotation matrix of the corresponding rail node coordinates and sleeper components in the digital twin model based on the geometric deviation measurement element, maps the geometric shape changes, and generates a calibration geometric representation volume;

[0044] The topology evolution module retrieves the calibrated geometric representation and, combined with maintenance records during the sleeper-rail maintenance cycle, determines changes in component connection relationships caused by sleeper replacement or the addition or removal of fasteners. This module then generates a connection change instruction set, reconstructs the constraints between the rails and sleepers, and constructs an evolving topology network diagram based on the connection change instruction set.

[0045] The adaptive collaborative simulation module receives the evolutionary topology network diagram and the simulation analysis requirements for the sleeper rail maintenance cycle. Based on the identified stress areas, it selects matching track unit models and combines them into a mixed-fidelity model configuration. The module then performs simulation operations based on the mixed-fidelity model configuration to obtain adaptive simulation output results.

[0046] The steps to obtain the integrated structure attribute set are:

[0047] Based on the rail surface point cloud data collected by the laser scanner and the image sequence obtained by drone aerial photography, the three-dimensional coordinates of the rail edge contour points and the geometric position parameters of the center point of the sleeper cross section in each frame are extracted. After uniformly projecting them into the WGS-84 coordinate system, all acquisition timestamps are aligned. The corresponding frame sequences are indexed and matched one by one to form a set of joint feature point pairs, resulting in a list of aligned joint feature point pairs.

[0048] According to the aligned joint feature point pair list, the rail profile curve registration factor is calculated using the following formula:

[0049] ;

[0050] in, is the rail profile curve registration factor, is the number of joint feature points, Laser scanner The three-dimensional coordinates of the points, The image measurement sequence corresponds to The three-dimensional coordinates of a point;

[0051] According to the rail profile curve registration factor and the three-dimensional offset value of each pair of feature points, point pairs with offset values greater than the rail profile curve registration factor are screened, and the remapping vector is recalculated and the coordinates are replaced in the global coordinate system. At the same time, the geometric parameters of the center point of the sleeper cross section are corrected by coordinate fitting. The remapped point coordinates and the corrected geometric parameters are combined to generate an integrated structural attribute set.

[0052] Specifically, based on the track surface point cloud data collected by the laser scanner and the image sequence obtained by drone aerial photography, the laser point cloud data is first preprocessed, including removing noise points and outliers generated by non-track objects (such as vegetation and debris) in the track environment, and eliminating points whose distance from neighboring points exceeds three times the standard deviation. Then, the drone image sequence is subjected to distortion correction and feature extraction. The camera's intrinsic and extrinsic parameters are used to perform geometric correction on each frame of the image to eliminate lens distortion and projection deformation, and a scale-invariant feature transformation algorithm is used to extract stable feature points in the image. Then, for the extraction of the three-dimensional coordinates of the rail edge contour points, in the laser point cloud data, Identify the rail area, then perform edge detection and fitting on the rail cross-section point cloud, obtain the edge points of the rail profile, and record their 3D coordinates. For drone images, use a deep learning target detection model (such as YOLOv5) to identify the rail area, and then use an edge detection operator (such as the Canny operator) to extract the rail edge. Combined with photogrammetry principles and ground control points, restore its 3D coordinates. Extract the geometric position parameters of the center point of the sleeper cross section. In the laser point cloud, identify the sleeper point cloud cluster and calculate its geometric center. In the drone image, use template matching or target detection to identify the sleeper, calculate its image center point, and then use triangulation to find the center point of the sleeper. The three-dimensional position is restored by measurement or structured light projection. Subsequently, the coordinates extracted from the laser point cloud (usually based on the scanner's own coordinate system or the local engineering coordinate system) and the coordinates extracted from the drone image (usually based on the image coordinate system or the geographic coordinate system of the initial GPS positioning) are uniformly projected into the WGS-84 coordinate system. This process requires the use of the GPS / IMU data recorded by each data acquisition device and the preset ground control points, and the conversion is performed through the seven-parameter coordinate conversion model (Bursa model). Then, the timestamps of all collected data are aligned to ensure that the laser scanning data frame and the drone image frame are strictly corresponding in time. If the acquisition device clock is not accurately synchronized, In the first step, synchronization events that can be clearly identified in both data sources (such as the moment when a specific train passes) are identified or clock signals are used for post-synchronization calibration. Finally, according to the aligned timestamps, the laser scanning point cloud frames and drone image frames at the same moment or within a very short time interval (such as 0.05 seconds) are associated by index number. In these associated frames, the corresponding feature points from the two data sources are matched according to spatial proximity and feature similarity (such as extracted rail edge points or sleeper corner points) to form joint feature point pairs containing laser point coordinates and corresponding image measurement point coordinates. These pairs are compiled into a list to obtain an aligned joint feature point pair list.

[0053] formula: The usefulness of the formula is that it provides a quantitative evaluation index for the spatial registration quality of multi-source data by calculating the average three-dimensional Euclidean distance between the corresponding feature points in the laser scanning data and the UAV image measurement data, namely the rail profile curve registration factor This can determine the degree of geometric consistency of the fusion of the two data sources, provide a reliable basis for subsequent data screening and model calibration, and ensure the accuracy of the geometric form of the digital twin model. Specifically, The average distance of the joint feature points is used to measure the overall registration effect, avoiding the extreme deviation of a single or a few point pairs from having too great an impact on the overall evaluation;

[0054] is the number of joint feature points, which refers to the total number of pairs of feature points with the same name that are successfully matched in the laser scanning data and the UAV image measurement data. This parameter is directly derived from the statistical count of the previous step "obtaining a list of aligned joint feature point pairs". For example, if the list contains 1500 pairs of successfully matched feature points, then .

[0055] Laser scanner The three-dimensional coordinates of the points are represented in the "aligned joint feature point pair list". For the joint feature points, the easting coordinates, northing coordinates and elevation values in the WGS-84 coordinate system, which are collected by the laser scanner, these coordinate values are directly measured by the laser scanning device and obtained through coordinate conversion. For example, for the feature points, whose laser scanning coordinates are .

[0056] The image measurement sequence corresponds to The three-dimensional coordinates of the points are represented in the "aligned joint feature point pair list". The easting coordinates, northing coordinates and elevation values of the same-name points in the joint feature points are obtained from the UAV image sequence and solved by photogrammetry methods (such as SfM or binocular vision) in the WGS-84 coordinate system. These coordinates are the three-dimensional point coordinates obtained after the UAV aerial image is processed by aerial triangulation and bundle adjustment. For example, for the feature points, whose image measurement coordinates are .

[0057] Calculation process:

[0058] It is assumed that 3 pairs of joint feature points are selected from the "aligned joint feature point pair list" for calculation, namely .

[0059] The first pair of feature points ( ):

[0060] Laser scanning coordinates ;

[0061] Image measurement coordinates ;

[0062] The spatial distance between the first pair of points

[0063] ;

[0064] ;

[0065] ;

[0066] The second pair of feature points ( ):

[0067] Laser scanning coordinates ;

[0068] Image measurement coordinates ;

[0069] The spatial distance between the second pair of points ;

[0070] ;

[0071] ;

[0072] ;

[0073] The third pair of feature points ( ):

[0074] Laser scanning coordinates ;

[0075] Image measurement coordinates ;

[0076] Spatial distance of the third pair of points ;

[0077] ;

[0078] ;

[0079] ;

[0080] Calculate the rail profile curve registration factor :

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] The results show that in this example, the average three-dimensional spatial deviation between the laser scanning data and the UAV image measurement data at the three selected pairs of joint feature points is approximately 0.0306 meters (or 30.6 millimeters). This value reflects the overall geometric accuracy of the current multi-source data fusion.

[0086] According to the rail profile curve registration factor calculated in the previous step As well as the three-dimensional coordinates of each pair of feature points in the "aligned joint feature point pair list", first calculate the three-dimensional Euclidean distance between each pair of feature points, that is, the three-dimensional offset value, which is calculated as follows Then, the 3D offset value calculated for each point pair is compared with the preset screening threshold, which is set as the rail profile curve registration factor A specific multiple of, for example, 1.5 times, which is the screening threshold , if in the example ,but , if the 3D offset value of a point pair is greater than this , then the point pair is determined to be a gross error data point and removed from the list, or marked as a low weight to reduce its impact on subsequent processing. Next, for the valid feature point pairs retained after screening, the remapping vector is recalculated in the global WGS-84 coordinate system and the coordinates are replaced. This step aims to unify and optimize the coordinates of the feature points. A data source with higher accuracy (such as laser scanning data) is selected as the benchmark. If the laser point coordinates are And the image point coordinates are , and the point pair is not filtered out, the laser point coordinates can be used as the final fusion coordinates of the feature point, that is, the remapped coordinates are , or calculate the weighted average of the two, where the weights are determined based on their prior accuracy. For example, the accuracy of lidar point clouds is usually better than that of drone photogrammetry, so lidar data can be given a higher weight, such as 0.7, and drone data has a weight of 0.3. The remapped coordinates are At the same time, the geometric parameters of the center point of the sleeper cross section (mainly its three-dimensional coordinates) are corrected by coordinate fitting, and a series of sleeper center point coordinates distributed along the track are projected onto the track center line fitted by these points, or the sleeper point cloud in the local area is plane fitted to correct its elevation and plane position to eliminate isolated measurement errors. For example, the least squares method can be used to fit the mathematical expression of a section of the track center line, and then each sleeper center point is projected onto this fitting line to obtain the corrected coordinates of the sleeper center point. Finally, the rail profile point coordinates after remapping are integrated with the geometric parameters of the sleeper cross section center point after coordinate fitting correction to generate an integrated structural attribute set.

[0087] The steps to obtain the geometric deviation metric are:

[0088] Based on the standard geometric forms of rails and sleepers preset in the digital twin model, the rail edge contour point coordinates and the geometric parameters of the sleeper cross-section center point in the integrated structural attribute set are called to match the corresponding rail component reference points and sleeper component center points in the digital twin model, forming a one-to-one correspondence between the model and the measured points in space point pairs;

[0089] Based on the coordinates of each pair of rail component reference points and the measured rail edge contour points in the spatial point pair, the coordinate differences of the point pair along the X, Y, and Z axes in three-dimensional space are calculated one by one, and a complete three-dimensional spatial displacement vector is synthesized and the vector length is recorded. At the same time, the average spatial displacement vector length of the point pair set is calculated to generate a rail spatial displacement vector set;

[0090] Based on the geometric parameters of the center points of each pair of sleeper components in the spatial point pair and the center point of the measured sleeper cross section, the Euler angle difference of the rotational attitude change between the two points is extracted, and the attitude angle differences in pitch angle, yaw angle and roll angle are recorded respectively to form a sleeper attitude angle difference set. The spatial displacement vector set of the rail and the sleeper attitude angle difference set are combined to obtain the geometric deviation metric.

[0091] Specifically, based on the standard geometric form of rails and sleepers preset in the digital twin model, the acquired integrated structural attribute set is first called. This attribute set contains the three-dimensional coordinates of the rail edge contour points and the geometric position parameters of the center point of the sleeper cross section after multi-source data fusion processing. Then, a matching operation is performed to match the corresponding rail component reference points predefined in the digital twin model (these reference points are arranged on the theoretical center line or rail top surface of the rail according to the design drawings during the construction phase of the digital twin model, for example, a three-dimensional coordinate is defined along the center line of the rail every 0.25 meters). The punctuation mark is used as a reference point) and the coordinates of the measured rail edge contour points extracted from the integrated structural attribute set are associated. At the same time, the center point of the sleeper component in the digital twin model (that is, the preset three-dimensional coordinates of the sleeper geometric center and the standard installation posture parameters, such as the Euler angle under ideal conditions) is matched with the geometric parameters of the center point of the measured sleeper cross section in the integrated structural attribute set (including the measured three-dimensional coordinates and the actual posture parameters obtained by local point cloud analysis). For each measured rail edge point, the system searches for the spatial distance on the corresponding rail theoretical contour line of the digital twin model. The search is limited to a dynamically adjusted search window, with an initial search radius of, for example, 0.1 m. This radius is based on the 95% deviation distribution range of historical track geometry detection data of the same type. If the deviation of the matching point pair is greater than this radius, the search window may be expanded or marked as a potential gross error point. For sleepers, a preliminary match is performed based on the three-dimensional Euclidean distance between the measured sleeper center point and the preset sleeper center point in the model. The search radius is, for example, set to 0.15 m. This value is based on the allowable deviation of the sleeper position (for example, transverse The upper limit of the matching is determined by taking into account the horizontal deviation of ±10 mm, the vertical deviation of ±5 mm, and the plane distortion, etc.) and a certain safety margin (for example, an increase of 50%). To ensure the uniqueness and accuracy of the match, the system will also use the local geometric feature information of the point cloud, such as the normal vector and curvature, or the sequence relationship of the point positions (such as the continuity along the direction of the rail) to assist in the judgment. When the distance between a measured point and multiple model reference points is within the threshold, the one with the most similar geometric features or the most consistent sequence is selected as the final matching pair, forming a spatial point pair with a one-to-one correspondence between the model and the measured point.

[0092] Based on the one-to-one correspondence of the rail component reference points in the digital twin model and the measured rail edge contour point coordinates obtained in the previous step, the system processes each pair of such spatial points. Specifically, if the first The coordinates of the reference points of the rail components are , and the corresponding measured rail edge contour point coordinates are , then first calculate the coordinate difference of the pair of points along the three orthogonal coordinate axes X, Y, and Z in three-dimensional space one by one, that is , ,as well as , these three difference components together constitute the three-dimensional spatial displacement vector from the model reference point to the measured point , which intuitively represents the spatial deviation of the measured point relative to its ideal design position. Subsequently, the length of the spatial displacement vector (i.e., the modulus) is calculated and recorded. , this length The absolute spatial deviation between the point pairs is quantified. After completing the above calculations for all spatial point pairs, the lengths of all calculated spatial displacement vectors will also be calculated. Perform statistics and calculate the average value of these length values, that is, the average spatial displacement vector length of the point pair set ,in is the total number of spatial point pairs. This average length value can be used as a macro-index for evaluating the overall geometric conformity of the rail. Finally, all the calculated individual spatial displacement vectors and its corresponding length , together with the calculated average spatial displacement vector length , organized together to generate a set of rail space displacement vectors.

[0093] Based on the spatial point pairs generated by the model in the above steps and the measured points, especially for the part of the sleeper component, each pair consists of the center point of the sleeper component in the digital twin model (including its ideal three-dimensional coordinates and the preset standard rotation posture, which can be expressed by a standard rotation matrix Or a set of base Euler angles such as The measured cross-sectional center geometric parameters obtained from the integrated structural attribute set (including the measured three-dimensional coordinates and the actual rotation posture obtained by analyzing the sleeper point cloud data or directly measuring the sensor, which can be expressed as the actual rotation matrix Or a set of measured Euler angles The system extracts and quantifies the rotational attitude change between the two. First, the relative rotation transformation from the model attitude to the measured attitude is calculated. If the attitude is represented by a rotation matrix, the difference rotation matrix can be calculated as (or , since for a rotation matrix, its inverse is equal to its transpose), from this difference rotation matrix In the equation, the Euler angle representing the attitude difference can be analyzed, that is, the pitch angle difference ( ), yaw angle difference ( ) and the roll angle difference ( ), this parsing process uses the standard conversion formula from rotation matrix to Euler angle, and needs to pay attention to handling special cases such as universal locks to ensure the uniqueness and correctness of the angles. The calculated attitude angle difference values ​​of pitch, yaw and roll angles are recorded one by one to form a set of sleeper attitude angle differences. Finally, this set of sleeper attitude angle differences is combined with the set of rail spatial displacement vectors generated in the previous step (including the displacement vectors of each reference point of the rail and their length information). This combination integrates two different types of deviation information (linear displacement deviation of the rail and angular attitude deviation of the sleeper) into a unified data structure, for example, forming a list containing the geometric deviation of each sleeper and its associated rail segment to obtain a geometric deviation metric.

[0094] The steps to obtain the calibration geometric representation are:

[0095] Based on the geometric deviation metric, each spatial displacement vector in the rail spatial displacement vector set is extracted, and the displacement components of the spatial displacement vector on the three coordinate axes of X, Y, and Z are analyzed respectively. For each corresponding rail node coordinate in the digital twin model, the analyzed displacement component values of each axis are superimposed on the original coordinate value of the node to form the adjusted rail node coordinate sequence;

[0096] Based on the adjusted rail node coordinate sequence, the sleeper attitude angle difference set in the geometric deviation metric is called, the pitch angle difference, yaw angle difference and roll angle difference of each sleeper in the set are analyzed one by one, the rotation matrix parameters of each sleeper component are recalculated, and the corrected sleeper rotation matrix sequence is generated;

[0097] Based on the adjusted rail node coordinate sequence and the corrected sleeper rotation matrix sequence, the rail node coordinate sequence and the sleeper rotation matrix sequence are applied one by one to the original geometric configuration of rails and sleepers in the digital twin model, and the three-dimensional morphological mapping is re-updated to form a calibrated geometric representation.

[0098] Specifically, based on the geometric deviation metric element obtained in the previous step, a rail spatial displacement vector set is first extracted from it. For each spatial displacement vector in the set, for example, Space displacement vector , the system will analyze the displacement components on the three coordinate axes X, Y, and Z in the global coordinate system, namely 、 and , these components directly represent the deviations in all directions of the measured rail point relative to the corresponding reference point in the digital twin model. Then, all preset rail nodes in the digital twin model are traversed. For each rail node, for example, its original coordinates are , if the node is aligned with a certain spatial displacement vector in the rail spatial displacement vector set (Its weight is ) has a clear corresponding relationship (this corresponding relationship is established when calculating the geometric deviation measurement element, that is, the matching of the model reference point and the measured point), then the resolved displacement component values of each axis are directly superimposed on the original coordinate value of the node, that is, the adjusted node coordinate If there are multiple measured point displacement vectors near a model node, the comprehensive displacement correction of the node may be calculated by weighted averaging (for example, weighting based on the inverse of the distance), or the coordinates of all relevant rail nodes may be updated according to predefined mapping rules (for example, interpolating the displacement vectors of all measured points on a section of rail to the discrete model nodes of this section of rail). By performing this coordinate superposition operation on all rail nodes in the digital twin model that have corresponding relationships with the measured data, an adjusted rail node coordinate sequence is formed.

[0099] Based on the adjusted rail node coordinate sequence obtained in the previous step, the system then calls the sleeper attitude angle difference set in the geometric deviation measurement element. This set records the pitch angle difference, yaw angle difference, and roll angle difference of each measured sleeper relative to its preset standard attitude in the digital twin model. The system will process the attitude angle difference data of each sleeper in the set one by one. For example, for the first sleepers, the recorded pitch angle difference is , the yaw angle difference is , the roll angle difference is At the same time, the original rotation matrix of the sleeper in the digital twin model is obtained and recorded as (or its original Euler angles ), then use these angle difference values to recalculate the rotation matrix parameters of the sleeper component. If the original posture and the difference are given in Euler angles, the difference Euler angle can be converted into a difference rotation matrix , and then update the original rotation matrix via matrix multiplication: (Note the convention of the rotation order. Usually, the intrinsic rotation is applied first, then the deviation rotation, or the difference angle is directly added to the original angle and then converted to a new rotation matrix. However, directly adding Euler angles may introduce problems when there are large angle rotations or when approaching universal lock, so converting to a rotation matrix is more secure.) For example, you can first convert the original Euler angles Convert to original rotation matrix , then the difference Euler angle Convert to difference rotation matrix , and finally calculate the adjusted rotation matrix , the rotation matrix recalculation process is performed on all sleeper components with attitude angle difference records to generate a corrected sleeper rotation matrix sequence.

[0100] Based on the adjusted rail node coordinate sequence and the corrected sleeper rotation matrix sequence obtained in the first two steps, the system applies these two sets of updated geometric parameters one by one to the original geometric configuration of rails and sleepers in the digital twin model. In the specific operation, for the rail part, the coordinates of each node of the rail centerline or contour line originally defined by a series of nodes are replaced by the corresponding values in the adjusted rail node coordinate sequence, which directly leads to the update of the rail geometry in the digital twin model, making it closer to the measured state. For the sleeper part, the spatial posture of each sleeper component in the model is determined by its rotation matrix. Now its original rotation matrix is replaced by the new rotation matrix corresponding to the sleeper in the corrected sleeper rotation matrix sequence. The center position of the sleeper may also need to be adjusted in conjunction with the displacement of its associated rail node, or the coordinates of the sleeper center point corrected by the integrated structural attributes can be directly used. After updating the coordinates of all relevant rail nodes and the rotation matrices of all relevant sleeper components, the system will trigger a re-rendering or geometric reconstruction process of the 3D model. In this process, the model rendering engine or geometric modeling kernel will regenerate the 3D solid or surface of the rail according to the new node coordinates, and reposition and orient the 3D solid or surface of each sleeper according to the new rotation matrix and position parameters. This re-performed 3D morphological mapping update ensures that the digital twin model is consistent with the latest measurement status of the actual track structure at the geometric level, forming a calibrated geometric representation.

[0101] The steps to obtain the connection change instruction set are:

[0102] Retrieve the rail node coordinates and sleeper component rotation matrix from the calibration geometry representation, and identify the adjacency information and connection type identifiers in the original connection relationship between each node and each sleeper. Combined with the maintenance record text content during the sleeper-rail maintenance cycle, extract all record sentences with the keywords "replacement", "removal" and "additional", and parse the associated sleeper numbers and fastener numbers to form an initial list of structural connection changes;

[0103] Based on the initial list of structural connection changes, the resolved sleeper numbers and fastener numbers are compared one by one with the current connection status in the calibrated geometric representation according to the time period and spatial position interval corresponding to each sleeper number. If there is a sleeper number in the list but not in the calibrated geometric representation, it is determined to be a new sleeper connection. If there is a fastener number in the calibrated geometric representation but not in the list, it is determined to be a fastener removal, and a list of component connection relationship changes is generated;

[0104] According to the list of component connection relationship changes, a standardized instruction text format is constructed for all connection relationships marked as added or removed, and the instruction type, target number, operation node index and impact range identifier are supplemented in sequence to generate a connection change instruction set.

[0105] Specifically, the calibration geometric representation generated in the previous step is retrieved, which contains the updated rail node coordinates and sleeper component rotation matrix. On this basis, the original connection relationship between each rail node and each sleeper in the model is first parsed, and the adjacency information of each connection is identified item by item. For example, it is clearly stated which fastener position of which sleeper a certain rail node is directly connected to, as well as the connection type identification, such as "elastic fastener connection", "rigid connection" or other specific types of constraints. These original connection information are usually preset according to the design specifications when the digital twin model is initialized. Next, the system combines the maintenance record text content within the sleeper maintenance cycle imported from an external database or file. The text content records historical maintenance activities, such as what operations were performed on a certain section on a certain day of a certain month of a certain year, using keyword matching and named entity recognition methods. , all sentences containing preset keywords are extracted from these maintenance record texts. These keywords are set as "replacement", "removal" and "addition". For example, when a record such as "On May 10, 2023, 3 sleepers (numbered S001, S002, S003) were replaced at K10+500, and 2 sets of fasteners (numbered F005, F006 to sleeper S001)" is detected, the keywords "replacement" and "addition" will be identified, and the specific sleeper numbers (such as S001, S002, S003) and fastener numbers (such as F005, F006) associated with these operations will be further parsed, as well as the location (K10+500) and time of the operations. By performing this kind of automated extraction and parsing on all maintenance record texts, all identified structural change information is summarized to form an initial list of structural connection changes.

[0106] Based on the initial list of structural connection changes generated in the previous step, the system will carefully compare the timestamp and spatial location information (such as kilometer mark, line name, etc.) corresponding to the sleeper number associated with each record in the list with the component connection status currently recorded in the calibration geometric representation. This comparison process involves aligning the time information in the maintenance record with the specific time snapshot represented by the calibration geometric representation. For example, if the calibration geometric representation represents the status on June 1, 2023, then in the initial list of structural connection changes, all relevant maintenance records before this date should be considered. For spatial location, the system will locate the corresponding sleeper or rail section in the calibration geometric representation according to the sleeper number or section information provided in the maintenance record. Then, for each sleeper number mentioned in the maintenance record, check whether the number exists in the current sleeper list of the calibration geometric representation. If a sleeper number appears in the initial list of structural connection changes, If a sleeper is not found in the list (usually due to a "replacement" or "addition" record), but the corresponding sleeper entity cannot be found in the calibration geometric representation through its unique identifier (such as the sleeper number), then the sleeper is determined to have been added to the track structure after the maintenance record time and before the current representation time, and it is recorded as a new sleeper connection. Conversely, for fasteners, the fastener connection relationship currently existing in the calibration geometric representation will be checked. If a fastener number (for example, uniquely determined by its installation position and type on the sleeper) exists in the connection record of the calibration geometric representation, but the initial list of structural connection changes indicates that the fastener has been "removed" or the sleeper to which it belongs has been "replaced" (replacing a sleeper usually means that the old fastener on it is removed together), then the fastener is determined to have been removed and recorded as a fastener removal. Through this two-way comparison, the changes in component additions and subtractions caused by maintenance activities can be identified, and these changes can be sorted and recorded to generate a list of component connection relationship changes.

[0107] According to the component connection relationship change list generated in the previous step, the list lists in detail all the sleeper and fastener connection relationships that are determined to be newly added or removed. The system then processes each change record in the list. The goal is to build a standardized instruction text format so that the subsequent topology evolution module can accurately understand and execute these changes. For each connection relationship marked as "new" (for example, adding a sleeper or a set of fasteners), the system will generate an instruction. The instruction first contains the instruction type, which is clearly identified as "ADD_CONNECTION", and then supplements the target number, which is the unique identifier of the newly added component (such as the number S101 of the new sleeper, or the position number C01-L of the new fastener on the sleeper to which it belongs), followed by the operation node index, which refers to the index of the rail node or sleeper component associated with the newly added connection in the digital twin model. Quotation marks or ID, and finally the impact range identifier, which is used to describe the local area or adjacent components affected by this new connection. For example, a new sleeper will affect the constraints of the rail segments on both sides of it, and a new fastener will change the connection characteristics between a specific rail and a specific sleeper. For each connection relationship marked as "remove" (such as removing an old sleeper or a set of damaged fasteners), the system also generates an instruction with the instruction type identified as "REMOVE_CONNECTION", the target number is the identifier of the removed component, the operation node index indicates the model node affected by the removal operation, and the impact range identifier indicates which connections no longer exist or have changed characteristics after removal. By executing this standardized instruction text construction process for all items in the component connection relationship change list, the complete instruction elements are supplemented in turn, and finally a connection change instruction set with a clear structure and clear content is formed.

[0108] The steps to obtain the evolutionary topology network diagram are:

[0109] Based on the connection change instruction set, the target number, operation node index and impact range identifier in each connection change instruction are parsed in turn. The rail node coordinates and sleeper component index in the calibration geometry representation are called to determine whether the target number corresponds to an existing node set. If it is a new instruction, the number is added to the node set and the initial connection entry is constructed. If it is a removal instruction, all connection entries associated with the number are deleted to obtain the structural connection reconstruction node table.

[0110] Based on the structural connection reconstruction node table, the connection status between each rail node coordinate and the corresponding sleeper component is scanned, all node pairs with physical connection relationships are extracted, and the connection orientation attributes are calculated according to the rotation matrix state of the sleeper component to form a constraint relationship reconstruction entry table;

[0111] The entry table is reconstructed based on the constraint relationship, and an adjacency list structure with node number as the main index is established one by one. The connection status, connection direction and connection label are embedded in the corresponding adjacency entry as edge attributes, and the edge index table is constructed simultaneously to generate an evolutionary topology network graph.

[0112] Specifically, based on the connection change instruction set obtained in the previous step, the system will read and parse each connection change instruction in turn. For each instruction, the target number contained therein (for example, the unique ID of the sleeper or fastener to be added or removed), the operation node index (indicating the internal index or ID of the rail node or sleeper component directly associated with the change in the digital twin model) and the impact range identifier (describing the local topological structure affected by the change, such as the ID list of adjacent components or the description of the spatial area) are extracted. While parsing the instruction, the system will call the rail node coordinate data and sleeper component index information in the calibration geometric representation. This information constitutes the geometry and identification basis of all structural components in the current digital twin model. Then, according to the target number in the instruction, the system needs to determine whether the component represented by the number already exists in the existing node set of the calibration geometric representation (that is, the list of all rail nodes and sleeper components in the current model). If the instruction type is a new instruction (for example, the instruction text starts with "ADD_CONNECTION") and the target If the number is confirmed to be not in the existing node set after query, the system will add this new target number to the node set and construct one or more initial connection entries for it based on the operation node index and impact range identifier in the instruction. These initial connection entries define the basic connection relationship between the new component and its surrounding components. For example, if a sleeper S101 is added, its operation node index may point to the rail nodes G50 and G51 on both sides of it, and the impact range identifier may indicate that it is connected through a standard fastener type. If the instruction type is a removal instruction (for example, the instruction text starts with "REMOVE_CONNECTION") and the target number is confirmed to exist in the existing node set, the system will delete the target number from the node set and all connection entries associated with the number at the same time. This means that all connections pointing to the removed component and connections issued from the component will be cleared. By performing the above judgment and operation on all instructions in the connection change instruction set, the nodes in the model and their basic connection information are dynamically updated to obtain a structural connection reconstruction node table.

[0113] According to the structural connection reconstruction node table updated and obtained in the previous step, which reflects the current rail nodes and sleeper components in the model and their preliminary connection intentions after the increase or decrease of components due to maintenance activities, the system will then conduct a detailed scan and analysis of the coordinate information of each rail node in the model and the connection status between the sleeper components to which it may be connected. The purpose of this scan is to identify and extract all actual physical connection relationships to form specific node pairs. For example, the system will check whether there is a sleeper component (also defined by its index in the table and the associated rotation matrix) near each rail node (defined by its coordinates in the structural connection reconstruction node table) and judge it based on the preset connection judgment criteria (for example, the spatial distance between the rail node and the fastener installation position on the sleeper is less than a specific connection tolerance threshold, which is set to 5 mm. This value is based on the precision requirements of the fastener installation and the actual measurement error. For example, if the fastener design allows an installation gap of 3 mm and the measurement error is 2 mm, the tolerance threshold is 5 mm) to determine whether there is a physical connection between the two. Once it is determined that a pair of rail nodes and sleeper components (or more specifically, a fastener position on the sleeper) are physically connected, the system will record the pair of nodes and further calculate the orientation attributes of the pair of connections based on the rotation matrix state of the sleeper component (the rotation matrix has been updated in the calibration geometric representation and reflects the actual spatial posture of the sleeper). The orientation attributes describe the direction and relative posture of the connection. For example, the vector pointing from the rail node to the fastener contact point can be calculated in the local coordinate system of the sleeper, or the contact angle of the rail relative to the sleeper can be described. By performing such connection status scanning and orientation attribute calculation on all rail nodes and sleeper components, the system summarizes all identified valid physical connections and their orientation attributes to form a constraint relationship reconstruction entry table.

[0114] Based on the constraint relationship reconstruction entry table generated in the previous step, the table contains all the actual physical connection information that has been verified and attribute calculated. The system now begins to build the final evolutionary topology network graph. The specific operation is, first, for each node in the network graph (representing a rail node or a sleeper component), establish an adjacency list structure with its unique node number (derived from the structural connection reconstruction node table) as the main index. The adjacency list is a commonly used graph representation method, in which each node is associated with a list that stores the numbers of all other nodes directly connected to the node. In the process of building the adjacency list, the system will extract the connection status (for example, "normal connection", "looseness", "damage warning", etc.) extracted from the constraint relationship reconstruction entry table. These status information may come from further analysis of maintenance records or sensor data) and the previously calculated connection direction (i.e. Part of the connection orientation attribute, such as the unit vector or specific angle representing the connection direction) and the connection label (for example, "the left rail is connected to the fastener No. 1 of the sleeper A", "elastic fastener type II") are embedded in the corresponding adjacency entry as the attributes of the edge. This means that when recording node B in the adjacency list of node A, these detailed attribute information of the connecting edge between A and B will also be attached. In order to improve query efficiency and facilitate subsequent analysis, the system will simultaneously build an edge index table while building the adjacency list. The table stores the information of all edges in the network. Each edge is uniquely identified by the number pair of the two nodes it connects, and all its attributes are associated. All entries in the entry table are reconstructed by processing the constraint relationships one by one, and the adjacency list and edge index table are filled in according to the above method. Finally, an evolutionary topological network diagram reflecting the latest topological connection relationship of the current sleeper-rail system is completely constructed.

[0115] The steps to obtain the mixed-fidelity model configuration are:

[0116] Based on the node structure relationship in the evolutionary topological network graph, the connection edge set between all node pairs is extracted, and the node number and spatial distribution position associated with each connection edge are identified one by one. The structural fatigue monitoring parameter table provided in the sleeper maintenance cycle simulation analysis requirements is matched, and the regional edge set whose connection edges fall into the fatigue risk identification interval is screened to generate a stress area boundary index list;

[0117] Based on the stress region boundary index list, the track element model adaptation factor of each stress region is calculated using the following formula:

[0118] ;

[0119] in, is the stress region track element model adaptation factor, is the total number of nodes associated with the connecting edges in the stress region, For the The vertical dynamic load value measured at each node during the current monitoring period, For the The cumulative stress value of each node in the current maintenance cycle, For the The theoretical stress reference value of each node defined in the current structural standard, is the equivalent effective area of the structural contact region, For the The characteristic length of the rail micro-segment represented by the node is defined as , Indicates the The physical length of the rail segment corresponding to each node in the topology map is actually mapped. Indicates the minimum characteristic length limit value set in the simulation configuration;

[0120] Based on the adaptation factor of the track unit model corresponding to each stress area, the adaptation factor of the track unit model is compared with the adaptation value benchmark threshold item by item. If the adaptation factor of the track unit model corresponding to the area is greater than or equal to the adaptation value benchmark threshold, the area is designated as a track unit model with complete parameter configuration; otherwise, it is designated as a track unit model with compressed parameter configuration. The two types of track unit models are integrated into a hybrid fidelity model configuration of the overall track system.

[0121] Specifically, based on the nodes (representing rail nodes or sleeper components) and their connection relationships contained in the evolutionary topological network diagram obtained in the previous step, the network diagram is first traversed to extract all edges representing direct physical connections between nodes to form a set of connection edges. For each connection edge in this set, the system will identify the unique numbers of the two associated endpoint nodes and the precise distribution positions of these nodes in three-dimensional space one by one. These spatial position information are directly derived from the rail node coordinates and sleeper component geometric parameters in the calibration geometric representation. Subsequently, the system will match the structural fatigue monitoring parameter table pre-defined in the sleeper maintenance cycle simulation analysis requirements. This parameter table is compiled based on common disease sites summarized in historical maintenance data (for example, welded joints, wear areas of rails outside small radius curves), and lists the fatigue sensitivity indicators of different structural components under different working conditions and their corresponding risk level thresholds. For example, the upper limit of the allowable crack growth rate in the rail welded joint area is 0.01 mm per day, or the warning value of the rolling contact fatigue index of the rail tread contact fatigue is 0.8. These parameters The entries in the table define specific fatigue risk identification criteria. The current state parameters of each edge and its associated nodes (such as cumulative stress, number of passes, and known defects, obtained from the digital twin model's historical database or real-time monitoring data) are matched against the risk identification criteria in the structural fatigue monitoring parameter table. When the relevant parameter values for the region in which a particular edge resides reach a preset fatigue risk identification interval—for example, if the cumulative total weight of a section of rail exceeds 80% of its design life (e.g., the design total weight is 600 million ton-kilometers, and the current total weight is 480 million ton-kilometers), or if the stress concentration factor of its associated nodes exceeds 1.5 (this factor is pre-calculated through finite element analysis or estimated based on geometric irregularities)—then the edge is determined to fall within the fatigue risk identification interval. All edges thus identified are collected to form a regional edge set. Finally, based on these selected regional edge sets, the system aggregates and removes duplicates from the endpoint nodes of these edges and generates a list of stress region boundary indexes based on their spatial order or logical grouping along the track.

[0122] formula: The formula is useful in that it calculates a track element model adaptation factor for each potential stress area by comprehensively considering the instantaneous dynamic load and cumulative stress deviation of the node and combining its geometric characteristic length. , this factor can quantify the necessity and urgency of high-fidelity simulation in this area. In the formula, dynamic load and the cumulative stress deviation term The square root of the sum of the squares is similar to calculating the modulus of a two-dimensional vector, which combines the contribution of short-term impact and long-term damage, multiplied by the equivalent action area Convert the stress deviation into an equivalent force so that it is equal to Dimensionally comparable, divided by the characteristic length The range of stress influence or the element size effect is taken into account, making the factor sensitive to both local high stress and widely distributed secondary high stress. The final average value makes It can represent the comprehensive state of the entire stress area and provide a basis for the subsequent adaptive selection of simulation models with different fidelity.

[0123] is the total number of nodes associated with the connecting edges in the stress region. This parameter is obtained by counting the number of unique nodes contained in the "stress region boundary index list" generated in the previous step. These nodes are the structural key points that constitute a specific stress region. For example, a stress region contains 3 connecting edges, which together connect 4 different rail nodes or sleeper components. .

[0124] For the The vertical dynamic load value measured at each node during the current monitoring period is collected in real time by sensors deployed on the track structure (such as strain gauges attached to the rail foot or rail waist, fiber Bragg grating sensors, or accelerometers installed on the sleepers). When a train passes, the sensor records the dynamic response data and extracts the maximum vertical dynamic load or an equivalent dynamic force amplitude that each monitoring node withstands within a specific time window (for example, during the passage of a train). For example, by analyzing the data of the strain sensor at a certain node when a train passed the most recently, the peak strain is extracted and multiplied by the corresponding force-strain conversion coefficient to obtain .

[0125] For the The cumulative stress value of a node in the current maintenance cycle is calculated by combining the historical load spectrum, the material's SN curve (stress-life curve), and linear cumulative damage theory (such as Miner's law). The number of various load cycles and their corresponding stress amplitudes experienced by the node since the last major maintenance or component replacement are recorded, and the current cumulative damage degree or equivalent cumulative stress is estimated based on this. The calculation formula example is: ,in Is the stress level The actual number of cycles under is the number of allowable cycles at this stress level (obtained from the SN curve), It is a reference stress value, or it can directly track the equivalent damage degree. For example, if a node has been subjected to a load equivalent to 60% of the design fatigue life during the current maintenance cycle and its design fatigue limit stress is 250MPa, its current cumulative stress state can be estimated. .

[0126] For the The theoretical stress reference value of each node defined in the current structural standard represents the expected stress level of this type of node under ideal conditions and when subjected to standard design loads, or serves as the stress threshold for fatigue assessment (stress below this value is not included in fatigue accumulation). For example, for a 60kg / m rail, under the action of a train with a standard axle load of 25 tons, the theoretical calculated stress reference value at the rail head jaw is set to .

[0127] is the equivalent effective area of the structural contact area. This parameter characterizes the effective contact range when the load is transferred between components. For wheel-rail contact, the area of the wheel-rail contact patch can be calculated using the Hertz contact theory. For the contact between the fastener and the rail or sleeper, the effective bearing area under the rated preload is estimated based on the fastener design drawings and the material elastic modulus. This value is usually determined according to the specific contact pair type and geometry. For example, for a typical elastic cushion between the rail and the sleeper, its equivalent effective area under the standard buckling pressure is (i.e. 100cm ).

[0128] For the The characteristic length of the rail micro-segment represented by the node is defined as ,in Indicates the The physical length of the rail segment that each node actually corresponds to in the evolutionary topology network diagram. This length is determined according to the discretization degree of the rail in the digital twin model. For example, if a node is set every 0.5 meters along the length of the rail, then , It is a minimum characteristic length limit value set in the simulation configuration to avoid numerical calculation problems caused by excessive discretization of the model. For example, according to experience or simulation stability requirements, , then if , .

[0129] Calculation process:

[0130] With a The stress area of a node is calculated as an example.

[0131] Node 1( ):

[0132] ;

[0133] ;

[0134] ;

[0135] (e.g. the same for all nodes);

[0136] ;

[0137] Node 2( ):

[0138] ;

[0139] ;

[0140] (The benchmark value may be different at different locations);

[0141] ;

[0142] ;

[0143] First calculate the contribution of each node :

[0144] For ease of calculation, convert MPa to kN / m (1MPa=1000kN / m ).

[0145] .

[0146] For node 1:

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] ;

[0152] For node 2:

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] calculate :

[0159] ;

[0160] ;

[0161] ;

[0162] ;

[0163] The results show that the track element model adaptation factor of this stress area Approximately , this value comprehensively reflects the severity of the dynamic load effect and cumulative stress deviation of the nodes in the area.

[0164] Based on the track element model adaptation factor corresponding to each stress region calculated in the previous step , the system will compare these items one by one The adaptive value and a preset benchmark threshold are set based on historical simulation data, maintenance experience, and the trade-off between simulation accuracy and computational cost. For example, the setting process is to collect historical data of 100 different track sections. The calculated value and its subsequent actual structural status (determined as "high risk requiring detailed analysis" or "low risk with simplified analysis" through inspection or monitoring) are recorded, and the calculation time and prediction accuracy of the simulation analysis using the complete parameter configuration model and the parameter compression configuration model are recorded, and an evaluation function is set. ,in is the candidate adaptation value reference threshold, express The average accuracy improvement of the high-fidelity model compared to the low-fidelity model in the region is The average computing time ratio thus increased is represented by the weight and Reflecting a preference for accuracy and efficiency, e.g. , by testing a series of Values (e.g. from historical The 50th to 95th percentile of the distribution, with a step size of 5th percentile), are selected so that Maximized The value is used as the final adaptation value reference threshold. For example, after the above analysis, the adaptation value reference threshold is determined to be , in the comparison process, if the track unit model adaptation factor calculated in a certain stress area is Greater than or equal to (For example, the ,but ), the system determines that the area needs high-precision simulation and specifies it as a track unit model with complete parameter configuration. The model will contain detailed geometric structure, accurate material constitutive relations, and complex contact and boundary conditions. On the contrary, if the area corresponds to Less than , it is considered that the structural risk of this area is relatively low, or its behavior has little impact on the overall system. A track unit model with parameter compression configuration can be adopted. This model may use simplified geometry, linear elastic materials or equivalent macroscopic mechanical parameters to reduce computational complexity. After performing this judgment and specification operation on all stress areas, the system organically combines these two types of track unit models with different fidelity (full parameter configuration model and parameter compression configuration model) according to their spatial position and connection relationship in the actual track, and integrates them into a hybrid fidelity model configuration of the overall track system.

[0165] The steps to obtain the adaptive simulation output results are:

[0166] Based on the areas defined by the hybrid-fidelity model configuration, the track unit models with full or compressed parameter configurations are called area by area. The current load conditions and material parameters of each track unit model in the area are loaded separately to perform dynamic force analysis and calculation. The stress data series of each regional node are obtained in real time to generate a real-time regional node stress data set.

[0167] Based on the real-time stress data set of regional nodes, the stress value change difference within consecutive time steps in each region is calculated and divided by the corresponding time interval to obtain the stress gradient change rate of each region. The calculated stress gradient change rate is compared with the preset stress gradient change rate threshold for each region, and abnormal regions with change rates exceeding the preset limit are identified to form an abnormal stress region identification list;

[0168] Based on the abnormal stress area identification list, for each abnormal area in the list, the track unit model originally configured with parameter compression configuration is upgraded to a complete parameter configuration, the track unit model originally configured with complete parameter configuration is parameter refined, the stress data is updated again, and the adaptive simulation output results are generated.

[0169] Specifically, based on the mixed-fidelity model configuration obtained in the previous step, which specifies what fidelity of track unit model (full parameter configuration or parameter compression configuration) should be adopted for each spatial region in the track system, the system will call the corresponding track unit model region by region according to this configuration. For the region designated as the full parameter configuration, the system will instantiate a finite element model or multi-body dynamics model containing detailed geometry (such as accurate rail and sleeper cross-sections, detailed fastener connections), nonlinear material constitutive relations (such as considering the elastic-plastic and fatigue properties of steel), complex contact algorithms (such as wheel-rail Hertzian contact or more sophisticated non-Hertzian contact models) and precise boundary conditions. For the region designated as the parameter compression configuration, a simplified beam unit model, an equivalent spring-damper system or a macroscopic mechanical model based on empirical formulas may be called. After calling the model, the system will load the current load conditions for the track unit model in each region. These load conditions include the dynamic wheel load time series generated by the train operation (obtained from the vehicle dynamics simulation module or actual monitoring data, for example, providing the three-dimensional force on the wheel at each time step). ), additional stress caused by temperature changes (calculated according to ambient temperature and rail constraints), and additional excitation caused by track irregularities, etc. At the same time, the latest material parameters are loaded. These parameters may be adjusted according to the service life of the components, accumulated damage or environmental factors (for example, the elastic modulus of the material changes with temperature, or fatigue damage causes the material strength to be reduced). After the loading is completed, the system performs independent or coupled dynamic force analysis and calculation on the model of each area, using numerical integration methods (such as Newmark- The dynamic response equation of the structure is solved using the time integration method or implicit / explicit time integration, and the stress (such as Mises equivalent stress and principal stress) data series of each key node in the area (such as specific monitoring points on rails, sleepers, and fasteners) in the simulation time history are obtained in real time. These sequences record stress values at a fixed time step (for example, 0.001 seconds). Finally, the node stress data series of all regions are integrated to generate a real-time regional node stress data set.

[0170] Based on the regional node stress real-time dataset generated in the previous step, which contains the stress values of key nodes in each region at continuous time steps during the simulation process, the system then processes the stress data in each region to calculate its stress gradient change rate. Specifically, for a node in a specific region, select its stress gradient change rate at two consecutive time steps (for example, time point and , where the time interval ) are and , calculate the difference between these two stress values , and then divide this difference by the corresponding time interval , and obtain the stress gradient change rate of the node in this time period, that is, This calculation will be performed on all representative nodes in the area, and the stress gradient change rates of these nodes may be averaged or maximized to represent the stress gradient change status of the entire area. Subsequently, the system will compare the stress gradient change rate calculated for each area with a preset stress gradient change rate threshold. The setting of this threshold is based on the statistical analysis of the stress change rate of the track structure under normal operating conditions and the recognition of the characteristics of drastic stress changes under abnormal working conditions (such as sudden breakage of track components, abnormal impact of trains, etc.). For example, by analyzing historical monitoring data and simulation cases, it is found that under normal circumstances, a key part of the rail is The stress change rate of a point generally does not exceed 500 MPa / s. However, in the event of an emergency such as weld cracking or fastener failure, the stress change rate may soar to more than 2000 MPa / s in a short period of time. Therefore, a stress gradient change rate threshold of, for example, 1000 MPa / s can be set as a judgment criterion. When the stress gradient change rate calculated in a certain area (for example, the maximum stress gradient change rate of the nodes in the area reaches 1500 MPa / s) exceeds this preset limit of 1000 MPa / s, the system will identify the area as an abnormal area, and summarize these identified abnormal areas to form an abnormal stress area identification list.

[0171] Based on the abnormal stress area identification list generated in the previous step, the list indicates which areas have stress change rates beyond the normal range during the current simulation process, which may indicate a sudden change in structural behavior or a rapid development of potential risks. The system will perform adaptive model adjustment and re-simulation strategies for each identified abnormal area in this list. Specifically, if an abnormal area was originally configured as a track unit model with parameter compression configuration (i.e., a low-fidelity model), the system will automatically upgrade it to a track unit model with full parameter configuration (i.e., a high-fidelity model), and will re-instantiate a model for the area that contains more detailed geometry, more accurate material constitutive and more complex contact conditions. If an abnormal area has originally been configured as a track unit model with full parameter configuration, but its stress gradient change rate is still However, if the preset limit is exceeded, this may indicate that even the high-fidelity model cannot fully capture the complex mechanical behavior of the area, or there is local stress concentration that cannot be accurately analyzed. At this time, the system will refine the parameters of the complete parameter configuration model of the area. Parameter refinement includes further encrypting the finite element mesh of the area (for example, reducing the unit size by 50%), adopting a higher-order unit type, and introducing a more sophisticated material nonlinear model. After completing the model upgrade or parameter refinement of all abnormal areas, the latest load conditions and material parameters will be reloaded based on the adjusted hybrid-fidelity model configuration, and dynamic force analysis and calculation will be performed again to obtain updated stress data. The final adaptively adjusted and recalculated stress data and related analysis results will generate the adaptive simulation output results.

Claims

1. The digital twin simulation system for sleeper rail maintenance cycle is characterized by: The system comprises: A multi-source data integration module integrates laser scanner point cloud data and drone photogrammetry image sequences to extract rail profile point coordinates and sleeper position geometric parameters, generating an integrated structural attribute set. A geometric consistency calibration module calculates the spatial displacement vector between the reference point of the rail component corresponding to the digital twin model and the measured coordinates based on the preset geometric form of the digital twin model and the integrated structural attribute set, obtains a geometric deviation metric element, and adjusts the rotation matrix of the corresponding rail node coordinates and the sleeper component in the digital twin model based on the geometric deviation metric element, maps the geometric form changes, and generates a calibration geometric representation volume; A topology evolution module retrieves the calibrated geometric representation and, in combination with maintenance records during the sleeper-rail maintenance cycle, determines changes in component connection relationships caused by sleeper replacement or addition or removal of fasteners, forms a connection change instruction set, and, based on the connection change instruction set, reconstructs the constraints between the rails and sleepers to construct an evolutionary topology network diagram. The adaptive collaborative simulation module receives the evolutionary topology network diagram and the simulation analysis requirements of the sleeper rail maintenance cycle, selects matching track unit models based on the identified stress areas, combines them into a mixed-fidelity model configuration, performs simulation operations based on the mixed-fidelity model configuration, and obtains adaptive simulation output results.

2. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the integrated structure attribute set are: Based on the rail surface point cloud data collected by the laser scanner and the image sequence obtained by drone aerial photography, the three-dimensional coordinates of the rail edge contour points and the geometric position parameters of the center point of the sleeper cross section in each frame are extracted. After uniformly projecting them into the WGS-84 coordinate system, all acquisition timestamps are aligned. The corresponding frame sequences are indexed and matched one by one to form a set of joint feature point pairs, resulting in a list of aligned joint feature point pairs. Calculating a rail profile curve registration factor according to the aligned joint feature point pair list; According to the rail profile curve registration factor and the three-dimensional offset value of each pair of feature points, point pairs with offset values greater than the rail profile curve registration factor are screened, and the remapping vector is recalculated and the coordinates are replaced in the global coordinate system. At the same time, the geometric parameters of the center point of the sleeper cross section are corrected by coordinate fitting. The remapped point coordinates and the corrected geometric parameters are combined to generate an integrated structural attribute set.

3. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the geometric deviation metric are as follows: Based on the standard geometric forms of rails and sleepers preset in the digital twin model, the coordinates of the rail edge contour points and the geometric parameters of the center point of the sleeper cross section in the integrated structural attribute set are called, and the corresponding rail component reference points and sleeper component center points in the digital twin model are matched to form a spatial point pair with a one-to-one correspondence between the model and the measured points; Based on the coordinates of each pair of rail component reference points and the measured rail edge contour points in the spatial point pairs, the coordinate differences of the point pairs along the X, Y, and Z axes in three-dimensional space are calculated one by one, a complete three-dimensional spatial displacement vector is synthesized and the vector length is recorded, and the average spatial displacement vector length of the point pair set is calculated at the same time to generate a rail spatial displacement vector set; Based on the geometric parameters of the center points of each pair of sleeper components in the spatial point pair and the center point of the measured sleeper cross section, the Euler angle difference value of the rotational posture change between the two points is extracted, and the posture angle differences in pitch angle, yaw angle and roll angle are recorded respectively to form a sleeper posture angle difference set, and the rail spatial displacement vector set and the sleeper posture angle difference set are combined to obtain a geometric deviation measurement element.

4. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the calibration geometric representation are: Based on the geometric deviation metric, each spatial displacement vector in the rail spatial displacement vector set is extracted, and the displacement components of the spatial displacement vector on the three coordinate axes of X, Y, and Z are analyzed respectively. For each corresponding rail node coordinate in the digital twin model, the analyzed displacement component values of each axis are superimposed on the original coordinate value of the node to form an adjusted rail node coordinate sequence; Based on the adjusted rail node coordinate sequence, calling the sleeper attitude angle difference set in the geometric deviation measurement unit, analyzing the pitch angle difference, yaw angle difference and roll angle difference of each sleeper in the set one by one, recalculating the rotation matrix parameters of each sleeper component, and generating a revised sleeper rotation matrix sequence; Based on the adjusted rail node coordinate sequence and the corrected sleeper rotation matrix sequence, the rail node coordinate sequence and the sleeper rotation matrix sequence are applied one by one to the original geometric configuration of the rails and sleepers in the digital twin model, and the three-dimensional morphological mapping is re-performed to update and form a calibrated geometric representation.

5. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the connection change instruction set are: Retrieving the rail node coordinates and sleeper component rotation matrix in the calibration geometric representation, and identifying the adjacency information and connection type identification in the original connection relationship between each node and each sleeper item by item, extracting all record sentences with the keywords "replacement", "removal" and "addition" based on the maintenance record text content during the sleeper-rail maintenance cycle, and parsing the associated sleeper numbers and fastener numbers to form an initial list of structural connection changes; Based on the initial list of structural connection changes, the parsed sleeper numbers and fastener numbers are compared one by one with the current connection status in the calibrated geometric representation according to the time period and spatial position interval corresponding to each sleeper number. If there is a sleeper number in the list but not in the calibrated geometric representation, it is determined to be a newly added sleeper connection; if there is a fastener number in the calibrated geometric representation but not in the list, it is determined to be a fastener removal, and a list of component connection relationship changes is generated; According to the component connection relationship change list, a standardized instruction text format is constructed for all connection relationships marked as added or removed, and the instruction type, target number, operation node index and impact range identifier are supplemented in sequence to generate a connection change instruction set.

6. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the evolutionary topology network diagram are: Based on the connection change instruction set, the target number, operation node index and impact range identifier in each connection change instruction are parsed in sequence, the rail node coordinates and sleeper component index in the calibration geometric representation are called, and it is determined whether the target number corresponds to an existing node set. If it is a new instruction, the number is added to the node set and an initial connection entry is constructed. If it is a removal instruction, all connection entries associated with the number are deleted to obtain a structural connection reconstruction node table; According to the structural connection reconstruction node table, the connection status between each rail node coordinate and the corresponding sleeper component is scanned to extract all node pairs with physical connection relationships, and the connection orientation attributes are calculated according to the rotation matrix state of the sleeper component to form a constraint relationship reconstruction entry table; Based on the constraint relationship, the entry table is reconstructed, and an adjacency list structure with node number as the main index is established one by one. The connection status, connection direction and connection label are embedded in the corresponding adjacency entry as edge attributes, and the edge index table is constructed simultaneously to generate an evolutionary topology network diagram.

7. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the mixed-fidelity model configuration are: Based on the node structure relationship in the evolutionary topological network graph, the connection edge set between all node pairs is extracted, and the node number and spatial distribution position associated with each connection edge are identified one by one. The structural fatigue monitoring parameter table provided in the sleeper rail maintenance cycle simulation analysis requirements is matched, and the regional edge set whose connection edges fall into the fatigue risk identification interval is screened to generate a stress area boundary index list; Calculating a track element model adaptation factor for each stress region based on the stress region boundary index list; Based on the adaptation factor of the track unit model corresponding to each stress area, the adaptation factor of the track unit model is compared with the adaptation value benchmark threshold item by item. If the adaptation factor of the track unit model corresponding to the area is greater than or equal to the adaptation value benchmark threshold, the area is designated as a track unit model with complete parameter configuration; otherwise, it is designated as a track unit model with compressed parameter configuration. The two types of track unit models are integrated into a hybrid fidelity model configuration of the overall track system.

8. The digital twin simulation system for sleeper rail maintenance cycle according to claim 1 is characterized in that: The steps for obtaining the adaptive simulation output result are: Based on each area delineated by the hybrid fidelity model configuration, the track unit model with full parameter configuration or parameter compression configuration is called area by area, the current load conditions and material parameters of each track unit model in the area are loaded respectively, dynamic force analysis and calculation are performed, and the stress data series of each regional node are obtained in real time to generate a real-time regional node stress data set; Based on the real-time stress data set of the regional nodes, the stress value change difference within consecutive time steps in each region is calculated and divided by the corresponding time interval to obtain the stress gradient change rate of each region. The calculated stress gradient change rate is compared with a preset stress gradient change rate threshold for each region, and abnormal regions whose change rate exceeds the preset limit are identified to form an abnormal stress region identification list; Based on the abnormal stress area identification list, for each abnormal area in the list, the track unit model originally configured as a parameter compression configuration is upgraded to a complete parameter configuration, the track unit model originally configured as a complete parameter configuration is parameter refined, the stress data is re-updated, and the adaptive simulation output results are generated.

Images