A method and system for constructing a highway maintenance digital twin application system

Abstract

The application discloses a kind of highway maintenance digital twinborn application system construction method and system, it is related to digital twinborn technical field, the steps of this method include: real-time acquisition of the multi-source perception data of highway section, and introduce constraint engine, to highway three-dimensional entity construction, generate several three-dimensional sub-image model;Among them, multi-source perception data includes highway image, highway laser point cloud and component attribute;Based on three-dimensional sub-image model, abstractly constructed to form topological space, adopt interpolation algorithm to generate datum curve, and combine the grid vertex of three-dimensional sub-image model, identify envelope boundary, to execute cross section scanning and generate target rendering road;Based on target rendering road, identify disease area, and based on disease area introduce preset maintenance rule, through semantic mounting mapping back to topological space, form digital twinborn application system;The application improves the efficiency of highway maintenance management, improves the computing power and rendering performance of three-dimensional rendering.

Description

A method and system for constructing a digital twin application system for highway maintenance Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a method and system for constructing a digital twin application system for highway maintenance. Background Technology

[0002] Digital twin technology integrates information technologies such as 3D digital models, the Internet of Things, and big data analysis. By constructing a digital virtual model corresponding to a physical entity, and through software definition, it describes, diagnoses, predicts, and manages the state of physical space, thereby achieving interactive mapping between physical and virtual spaces. This technology has been applied in the engineering field, especially in highway maintenance.

[0003] While traditional digital twin construction and application technologies for highways have achieved three-dimensional scene reproduction to a certain extent, they still have the following problems in practical applications: On the one hand, highway defect detection mostly relies on two-dimensional images captured by a single camera or three-dimensional point clouds scanned by LiDAR. A single two-dimensional image is difficult to quantify the depth information of defects, such as pothole depth, while a single three-dimensional point cloud, although possessing geometric information, lacks texture details and is difficult to identify minute cracks or repair marks. This results in the constructed three-dimensional model being merely a geometric mesh, lacking semantic embedding of defect areas. On the other hand, in the process of building digital twin application systems, high-precision three-dimensional image models contain massive amounts of texture data and mesh vertices. When performing real-time rendering of long-distance road sections, the computational resources are extremely consumed. The lack of effective view clipping and computational scheduling constraints easily causes system lag and frame rate drops, failing to meet the needs of maintenance personnel for smooth interaction and real-time analysis. At the same time, maintenance management and defect identification are usually independent technologies, lacking an automated connection mechanism between the two, which reduces the efficiency of highway maintenance. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for constructing a digital twin application system for highway maintenance. The system materializes the highway into a three-dimensional entity and divides it into three-dimensional sub-image models for different highway segments, including slope sub-models, pavement sub-models, bridge and tunnel sub-models, and ancillary facility sub-models. A topological space is constructed, and target rendering channels are generated using circular and rectangular cross-sections. By adding time series data, geometric identifiers of defects, and maintenance identifiers for different road segments and mapping them back to the topological space, a digital twin application system is formed. The system uses a topological mapping index file to convert defect vector groups into maintenance work orders with station numbers, successfully distinguishing defects such as cracks and potholes and matching routine, preventative, or restorative maintenance plans. This facilitates subsequent highway operation and maintenance management and solves the problems mentioned in the background technology.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In a first aspect, this application provides a method for constructing a digital twin application system for highway maintenance, the method comprising:

[0009] Multi-source perception data of highway sections are collected in real time, and a constraint engine is introduced to construct a three-dimensional entity of the highway, generating several three-dimensional sub-image models; among them, the multi-source perception data includes highway images, highway laser point clouds, and component attributes;

[0010] Based on the 3D sub-image model, an abstract topological space is constructed, an interpolation algorithm is used to generate a baseline curve, and the envelope boundary is identified by combining the mesh vertices of the 3D sub-image model to perform cross-sectional scanning to generate the target rendering road;

[0011] Based on the target rendering road, the diseased areas are identified, and pre-set maintenance rules are introduced based on the diseased areas. The semantic mapping is then used to map back to the topological space to form a digital twin application system.

[0012] Furthermore, the constraint engine has built-in first constraints on the three-dimensional space and second constraints on computing resources; the first constraints include the field of view, depth range, and occlusion culling threshold, and the second constraints include the maximum number of polygon faces and the lower limit of the number of frames transmitted per second.

[0013] Furthermore, the 3D physical model of the highway is constructed, generating several 3D sub-image models, including:

[0014] The first rendering model is constructed by default. Multi-source perception data is imported into the first rendering model, and feature deconstruction is performed to extract multiple basic features. Key features are selected by combining the first and second constraints. An adaptive training layer is introduced into the first rendering model. The selected key features are combined with the original basic features to form a new training dataset. The adaptive training layer is trained and updated to obtain the second rendering model.

[0015] Meanwhile, the data is layered according to time series, and each time series is marked as a rendering process. The multi-source perception data under each rendering process is imported into the second rendering model to generate corresponding three-dimensional sub-image models, including slope sub-model, road surface sub-model, bridge and tunnel sub-model and ancillary facility sub-model.

[0016] Furthermore, the first rendering model is built on a hybrid neural network architecture.

[0017] Furthermore, construct the target rendering path, including:

[0018] The three-dimensional sub-image model is analyzed, and the road centerline and key component structural points are extracted as target construction nodes. A directed topological graph structure is established, and the target construction nodes are smoothly fitted using the B-spline interpolation algorithm to generate the target baseline curve.

[0019] Identify the mesh vertices of the slope sub-model and the ancillary facility sub-model, filter the maximum projected distance and maximum Euclidean distance of all mesh vertices under each road segment relative to the target reference curve to generate the envelope boundary, and perform a cross-section scan on the envelope boundary in combination with the target reference curve to generate the target rendered road.

[0020] Furthermore, a cross-sectional scan is performed on the envelope boundary, including:

[0021] Cross-sectional scanning includes circular scanning and rectangular scanning;

[0022] Identify the component attributes of the current target node. If it is identified as a tunnel, perform a circular scan: extract the maximum projected distance to determine the envelope radius, generate a circular cross-section, and sample M circular contour points according to a preset step size; otherwise, perform a rectangular scan: extract the maximum Euclidean distance to determine the envelope half-width and envelope height, generate a rectangular cross-section, and perform equal-interval interpolation on the four sides of the rectangular contour to generate a rectangular contour point set with a total number of M; where M is greater than 0.

[0023] Extract the cross-sectional contour point set corresponding to the two adjacent target construction nodes in sequence, perform triangulation, close the head and tail of the generated triangular facets to form the target rendering road.

[0024] Furthermore, a digital twin application system is formed, including:

[0025] Texture and geometric features are separated based on the target rendering road;

[0026] The first abnormal region is determined based on texture features, and the second abnormal region is determined based on geometric features. The first and second abnormal regions are then subjected to a union calculation to obtain the corresponding diseased region. The boundary coordinates of the diseased region are then extracted to generate a geometric identifier for the disease. The boundary coordinates are in three-dimensional spatial coordinate form.

[0027] Meanwhile, the crack width, damaged area, and pit depth of the diseased area are extracted based on the boundary coordinates as evaluation indicators.

[0028] Based on geometric identifiers, at least one disease vector group is generated, including the entered disease geometric identifiers and the maintenance identifiers to be acquired; the evaluation indicators are imported into the preset maintenance rule engine, and through rule matching, the corresponding disease type and maintenance level are output and defined as maintenance semantic data, and the corresponding maintenance identifiers are generated and entered into the disease vector group.

[0029] Based on the disease vector group, a topology mapping index file corresponding to the target rendering road is generated; based on the topology mapping index file, the maintenance semantic data corresponding to the maintenance semantic identifier is mapped back to the topology space, and the target rendering road is integrated with the mapped topology data to form a digital twin application system.

[0030] Furthermore, the preset maintenance rule engine includes disease type determination and maintenance level determination.

[0031] Secondly, this application provides a system for constructing a digital twin application system for highway maintenance, the system comprising:

[0032] The first generation module: collects multi-source perception data of highway sections in real time, and introduces a constraint engine to construct a three-dimensional entity of the highway, generating several three-dimensional sub-image models; among them, the multi-source perception data includes highway images, highway laser point clouds, and component attributes;

[0033] The second generation module: Based on the 3D sub-image model, it abstracts and constructs a topological space, uses an interpolation algorithm to generate a baseline curve, and combines the mesh vertices of the 3D sub-image model to identify the envelope boundary in order to perform cross-sectional scanning to generate the target rendering road;

[0034] Application system construction module: Based on the target rendering road, identify the diseased areas, and introduce the preset maintenance rules based on the diseased areas. Through semantic mounting, map back to the topology space to form a digital twin application system.

[0035] Thirdly, this application provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the method for constructing a digital twin application system for highway maintenance provided in the first aspect above.

[0036] (III) Beneficial Effects

[0037] This invention provides a method and system for constructing a digital twin application system for highway maintenance, which has the following beneficial effects:

[0038] 1. This invention integrates an area array camera and a lidar to capture high-definition highway images and high-precision highway lidar point clouds in real time, and synchronously associates component attributes, providing full-element data support for the subsequent construction of a digital twin;

[0039] 2. This invention uses a preset first rendering model and introduces an adaptive training layer to classify the collected highway images, highway laser point clouds, and component attributes by feature classification. By constraining the basic features deconstructed from the features, and using the first and second constraints to dynamically balance the field of view and computing resources in three-dimensional space, different key features in each rendering process are selected, which indirectly improves the rendering accuracy of the second classification model and achieves smooth operation of the system.

[0040] 3. This invention abstracts and constructs a topological space by geometric analysis of a three-dimensional sub-image model, and generates a reference curve using an interpolation algorithm, thus constructing a spatial skeleton with geometric continuity. Furthermore, it sets circular and rectangular scans for component attributes to generate target rendering roads, enabling the rendering range to accurately fit complex structures such as slopes and tunnels, thereby improving computing power and rendering performance to a certain extent.

[0041] 4. This invention accurately identifies diseased areas based on target rendering of roads. Under the condition of identifying diseased areas, it triggers the matching mechanism of the maintenance rule engine. Through disease type determination and maintenance level determination, it performs maintenance semantic embedding on the diseased areas, generates disease vector groups, and maps them back to the topology space through the topology mapping index file to form a digital twin application system. To a certain extent, this ensures the data consistency of the digital twin application system and improves the intelligence level and response efficiency of highway maintenance decision-making. Attached Figure Description

[0042] Figure 1 is a flowchart of the present invention;

[0043] Figure 2 is a schematic diagram of the framework of the three-dimensional sub-image model in this invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] The core of this invention lies in the three-dimensional solidification of highways and the division of them into three-dimensional sub-image models for different highway segments, including slope sub-models, pavement sub-models, bridge and tunnel models, and ancillary facility sub-models. Each three-dimensional sub-image model is composed of bottom and upper layers. By analyzing the three-dimensional sub-image models, a topological space is constructed, and a high-precision target rendering channel is built by adaptively selecting the cross-section scanning method based on component attributes. This can realistically restore the appearance and structural details of road segments, achieving accurate perception of the physical environment. Time series, defect geometric markers, and maintenance markers for different road segments are added and mapped back to the topological space to form a digital twin application system, realizing visual interaction and business closed loop.

[0046] Example 1:

[0047] This invention provides a method for constructing a digital twin application system for highway maintenance; Figure 1 is a flowchart of this invention; Figure 2 is a framework diagram of the three-dimensional sub-image model in this invention; referring to Figures 1 and 2, the method includes the following steps:

[0048] S1: Real-time acquisition of highway images, highway laser point clouds, and component attributes of highway sections, and introduction of a constraint engine to construct a 3D solidified highway and generate several 3D sub-image models.

[0049] In this embodiment, multi-source sensing devices are set up along the highway section, and multiple collection points are distributed to collect highway images and highway laser point clouds in real time. Specifically, the multi-source sensing devices include area array cameras and lidar: the area array cameras are equipped with CMOS sensors and polarization filters to eliminate specular reflections on the asphalt pavement. They adopt a distance-triggered mode, using encoders or GNSS signals to trigger the shutter once every fixed distance traveled, for example, at a distance of 5 meters, to collect highway images of the road surface, traffic markings, and roadside facilities; the lidar emits laser pulses and receives reflected echoes, identifies the three-dimensional spatial coordinates and reflection intensity of the reflection points, and acquires highway laser point clouds of the road surface, guardrails, slopes, and surrounding areas; all sensing devices are synchronized at the nanosecond level via the PTP protocol to ensure that data collected at the same time corresponds to the same physical location.

[0050] For the same physical location at a given moment, based on the GIS model, the corresponding highway object is retrieved from the database using the physical location as the index key. Then, according to the object classification in the BIM model, the highway is horizontally decomposed into a bottom layer representing slopes, a middle layer representing pavement, a special layer representing bridges and tunnels, and an upper layer representing ancillary facilities. At the same time, based on the GIS model, the highway mileage markers are obtained, and the highway is divided into multiple highway segments of fixed length along the driving direction, such as each highway segment being 1 kilometer, and these segments are numbered 1, 2, ..., n, where n is a positive integer greater than 0.

[0051] The significance of the above data collection lies in the fact that by integrating area array cameras and lidar, it is possible to capture high-definition highway images and high-precision highway laser point clouds in real time, and simultaneously associate them with component attributes, providing full-element data support for the subsequent construction of digital twins.

[0052] A constraint engine is introduced to construct a 3D physical model of the highway and generate several 3D sub-image models. The constraint engine has a first constraint on the 3D rendering space and a second constraint on computing resources. The first constraint includes the field of view and depth range of the 3D rendering space, and the second constraint includes the maximum number of polygon faces and the lower limit of the number of frames transmitted per second.

[0053] The following is an explanation of the terms used in the constraint engine:

[0054] Field of view: refers to the range of visible angles within the 3D rendering space of a virtual camera, with the viewpoint as the vertex. It includes the horizontal and vertical field of view. When constructing the first rendering model, the field of view determines the size of the view frustum. A larger field of view covers a wider road surface but increases the number of models per frame; a smaller field of view focuses on local details. Depth of field: refers to the effective depth distance between the near clipping plane and the far clipping plane along the virtual camera's line of sight in the 3D rendering space. Road entities within this distance range, such as road surfaces, guardrails, and signs, will be preserved by the graphics pipeline and subjected to projection calculations. Objects smaller than the near clipping distance or larger than the far clipping distance will be depth-clipped. Occlusion culling threshold: refers to the critical value used to determine whether a 3D rendering object is occluded by foreground objects and therefore does not require rendering calculations. It is usually represented by the proportion of visible pixels or the percentage of projected area.

[0055] Minimum frame rate per second: The minimum image refresh rate that must be maintained to ensure visual smoothness and real-time operation response during real-time rendering and interaction. If the current real-time rendering frame rate is detected to be lower than this value, it is determined that the current load is too high, and a degradation strategy is automatically triggered, such as switching to low precision, reducing the field of view, or reducing the depth of field. Maximum number of polygon faces: The upper limit of the total number of geometric triangle faces that can be loaded and processed in a single rendering frame. For example, when building the target rendering pipeline in the future, the system will count the total number of polygons of all road surface, slope and ancillary facility sub-models in the current field of view in real time. If the total number exceeds this value, for example, more than 2 million faces, the system will forcibly simplify non-critical models in the distance according to the preset priority strategy, such as replacing distant trees with bulletin boards to prevent memory overflow or rendering delay.

[0056] A first rendering model is pre-constructed, and the first rendering model is based on a hybrid neural network architecture, including a perceptual encoding layer and an adaptive training layer. The perceptual encoding layer includes a convolutional neural network branch for processing high-dimensional highway image data and highway laser point cloud data, and a fully connected neural network branch for processing low-dimensional component attribute data. The highway image, highway laser point cloud, and component attributes are imported into the first rendering model. The convolutional neural network branch performs feature deconstruction on the highway image and highway laser point cloud, extracts texture features and geometric features through convolution and pooling operations, and performs linear mapping on the component attributes through the fully connected neural network branch to extract semantic features. The texture features, geometric features, and semantic features are defined as basic features.

[0057] Key features were selected by combining the first and second constraints, including:

[0058] Based on the field of view and depth range in the first constraint, the rendering geometric frustum in 3D space is determined; the geometric features in the basic features are mapped to 3D space and bounding box detection is performed to obtain the corresponding geometric bounding boxes, and spatial intersection operation is performed between them and the rendering geometric frustum, directly culling the basic features that do not intersect with the rendering geometric frustum; the projected area of ​​the retained basic features on the corresponding visualization interface is calculated, and the projected area is compared with the occlusion culling threshold in the first constraint: if the projected area is less than the occlusion culling threshold, the feature is determined to be an invisible feature and is culled a second time; if the projected area is greater than or equal to the first constraint... The occlusion culling threshold is used to generate a feature set to be rendered. Based on the feature set, the total number of polygon faces of all basic features in the feature set is counted. The system monitors whether the total number of polygon faces is greater than the maximum number of polygon faces in the second constraint, or whether the current real-time transmission frame rate is less than the lower limit of transmission frames per second. If either condition is detected, multi-level detail degradation processing is performed on the semantic features in the basic features until the second constraint is met. If neither of the above conditions is detected, the accuracy of the original basic features is maintained. Feature filtering is performed on multiple basic features that meet the first and second constraints to obtain key features.

[0059] The significance of the above analysis lies in the fact that by using the first and second constraints to dynamically balance the field of view and computing resources, while ensuring the visual fidelity of the core road sections, rendering stuttering, frame rate instability and video memory overflow are reduced, thus achieving smooth system operation.

[0060] An adaptive training layer is introduced into the first rendering model. The adaptive training layer is set at the output of the perceptual coding layer and is composed of a multi-layer perceptual mechanism. The input of the adaptive training layer is configured with a vector concatenation function. The selected key features are combined with the original basic features to form a new training dataset. Random sample data is provided from the dataset composed of highway images, highway laser point clouds and component attributes to train and update the adaptive training layer. The backpropagation algorithm is used to iteratively update the neuron connection weights in the adaptive training layer to obtain the second rendering model.

[0061] Simultaneously, the data is layered according to time series, and any time series is marked as a rendering process. The highway images, highway laser point clouds, and component attributes under each rendering process are imported into the second rendering model to generate corresponding three-dimensional sub-image models, including roadside slope sub-models, road surface sub-models, bridge and tunnel sub-models, and ancillary facility sub-models. Among them, the three-dimensional sub-image models are divided into several hierarchical architectures, including the bottom layer representing the slope sub-model, the middle layer representing the road surface sub-model, the special layer representing the bridge and tunnel model, and the upper layer representing the ancillary facility sub-model.

[0062] The significance of the above analysis is that by introducing an adaptive training layer into the first rendering model, the collected highway images, highway laser point clouds, and component attributes are classified, and different key features in each rendering process are selected, which indirectly improves the rendering accuracy of the second classification model.

[0063] S2: Based on the 3D sub-image model, an abstract topological space is constructed, an interpolation algorithm is used to generate a baseline curve, and the mesh vertices of the 3D sub-image model are combined to identify the envelope boundary in order to perform cross-sectional scanning to generate the target rendering road;

[0064] Constructing the target rendering path includes:

[0065] The geometric analysis of the 3D sub-image model extracts the road centerline and key component structural points as target construction nodes, establishes the adjacency topology between nodes, and constructs a directed topology graph structure, represented by the form: G(V, E); where V represents the geometry of the target construction node, and each node uniquely corresponds to a multi-dimensional attribute vector, including: node number, node component attribute type, such as: road center point, and the coordinate position of the node in 3D space; E represents the set of edges, which consists of the adjacency topology between nodes. For example, if nodes V1 and V2 are continuous at the highway mileage markers, then a directed edge is established.

[0066] The B-spline interpolation algorithm is used to smoothly fit the target construction nodes and generate the target reference curve. The target construction nodes are traversed along the target reference curve, and a series of sampling points are extracted according to a preset step size, such as every 1 meter. At each sampling point, the tangent vector at that node is calculated, a local normal plane perpendicular to the tangent vector is constructed, and a local rectangular coordinate system with the target construction node as the origin is established in the local normal plane.

[0067] Simultaneously, the grid vertices of the slope sub-model and the ancillary facility sub-model are extracted, and the maximum projection distance and maximum Euclidean distance of all grid vertices under each road segment relative to the target reference curve are calculated: Based on the spatial indexing algorithm, the grid vertices of the slope sub-model and the ancillary facility sub-model within a preset distance range before and after each target construction node are retrieved to form the local candidate vertex set corresponding to that node; each vertex in the local candidate vertex set is orthogonally projected onto the local normal plane to obtain a two-dimensional projection point; in the local normal plane, with the intersection point of the target reference curve passing through the plane as the origin, the maximum Euclidean distance from all projection points to the origin is calculated to determine the envelope radius; the maximum projection distance of all projection points on the horizontal axis of the normal plane is calculated and marked as the envelope half-width, and the maximum projection distance on the vertical axis is marked as the envelope height; a gradient threshold of wc% is applied to the maximum Euclidean distance to determine the maximum envelope radius; where wc is greater than 0; the boundary points determined by the maximum envelope radius at each sampling point are connected to form an envelope boundary that varies in width with the route direction;

[0068] First, identify the component attributes of the current target node. If it is identified as a tunnel, perform a circular scan: extract the maximum projected distance to determine the envelope radius, generate a circular cross-section, and sample M circular contour points according to a preset step size. Otherwise, perform a rectangular scan: extract the maximum Euclidean distance to determine the envelope half-width and envelope height, and determine the coordinates of the four vertices of the rectangle: where M is greater than 0; assuming the road surface is located on the 0 plane and extends upwards, the coordinates of the four generated vertices are (-width, 0), bottom right (width, 0), top right (width, height), and top left (-width, height); where width h represents the envelope half-width, and height represents the envelope height. Simultaneously, equal-interval interpolation is performed on the four sides of the rectangular outline to generate a rectangular outline point set with a total quantity of M, and the starting point of this point set is consistent with the circular outline point set. Then, the cross-sectional outline point sets corresponding to two adjacent target construction nodes are extracted sequentially. Using a triangular mesh algorithm, triangulation is performed, and continuous triangular patches are generated by connecting corresponding points on adjacent cross-sectional outlines, resulting in closed beginning and end. For the transition area between the circular and rectangular cross-sections, linear interpolation is used to generate gradient transition surface patches, closing the beginning and end ports, ultimately forming a target rendering road that includes all visible elements of the highway segment.

[0069] The significance of the above analysis lies in the following: by performing geometric analysis on the 3D sub-image model, abstracting and constructing the topological space, and using interpolation algorithms to generate baseline curves, discrete noise in multi-source perception data is effectively filtered out, and a spatial skeleton with geometric continuity is constructed, which to a certain extent ensures the smoothness of the target rendering channel on long-distance highway sections; and by setting circular and rectangular scanning for component attributes to generate the target rendering road, the rendering range can accurately fit complex structures such as slopes and tunnels, which to a certain extent improves computing power and rendering performance.

[0070] S3: Based on the target rendering road, identify the diseased area, and introduce the preset maintenance rules based on the diseased area. Then, through semantic mounting, map back to the topology space to form a digital twin application system.

[0071] To form a digital twin application system, including:

[0072] Texture and geometric features are separated based on the target rendering road;

[0073] The first anomalous region is determined based on texture features: the texture mapping layer of the target rendering corridor is obtained, converted into a grayscale image, and Gaussian filtering is performed to eliminate high-frequency noise interference; the Sobel operator is used to perform convolution operation on the grayscale image, the grayscale gradient value of each pixel is calculated, and pixels with grayscale gradient values ​​greater than the texture contrast threshold are selected and marked as anomalous pixels; morphological closing operation is performed on the anomalous pixels: first dilation and then erosion, connecting the broken pixels to form a closed connected region, and this connected region is defined as the first anomalous region; where the first anomalous region is in the form of a two-dimensional texture coordinate system;

[0074] The second anomalous region is determined based on geometric features: the geometric mesh layer of the target rendering corridor is obtained, and the least squares method is used to perform plane fitting on the mesh vertices within a local range to construct an ideal reference plane; each vertex in the mesh is traversed, and the vertical Euclidean distance from the vertex to the reference plane is calculated; the average curvature of each vertex is calculated, and mesh vertices with an average curvature greater than the curvature threshold are selected and marked as anomalous vertices. The set of topological meshes formed by adjacent anomalous vertices is defined as the second anomalous region; the second anomalous region is in three-dimensional space.

[0075] The union calculation of the first and second abnormal regions includes: performing spatial alignment on the first and second abnormal regions, back-projecting the abnormal vertices in the second abnormal region onto the two-dimensional texture coordinate system to generate corresponding abnormal masks; wherein, the first abnormal region is represented by a texture abnormal mask, and the second abnormal region is represented by a geometric abnormal mask, both located in the same two-dimensional space; performing a logical OR operation on the texture abnormal mask and the geometric abnormal mask to obtain a union mask; based on the range covered by the union mask, extracting the corresponding three-dimensional spatial mesh set in the target rendering channel to determine the disease area, and generating a disease geometric identifier by extracting the boundary coordinates of the disease area; wherein, the boundary coordinates are in three-dimensional spatial coordinate form.

[0076] Simultaneously, the crack width, damaged area, and pit depth of the diseased area are extracted based on boundary coordinates as evaluation indicators. Specifically, this includes: Crack width: The boundary region is obtained based on boundary coordinates, and skeleton extraction is performed. The intercepts from the skeleton points to the boundary are calculated along the normal direction of the skeleton line, and the average value of the intercepts is taken and marked as the crack width; Damaged area: Triangular mesh faces within the boundary coordinates are extracted, the area of ​​each face is calculated by vector cross product, and the sum is performed and marked as the damaged area; Pit depth: Boundary coordinate points are obtained based on boundary coordinates, and a virtual reference plane is fitted. The maximum vertical distance from the vertex inside the boundary to the virtual reference plane is calculated and marked as the pit depth.

[0077] Based on the geometric identifiers of the disease, at least one disease vector group is generated, including the entered geometric identifiers of the disease and the maintenance identifiers to be acquired; the evaluation indicators are imported into the preset maintenance rule engine, and through rule matching, the corresponding disease type and maintenance level are output and defined as maintenance semantic data, and the corresponding maintenance identifiers are generated and entered into the disease vector group.

[0078] The preset maintenance rule engine includes disease type determination and maintenance level determination. Through rule matching, it includes: Disease type determination: Extracting the geometric description of the evaluation indicators and matching the geometric description with preset geometric shape templates; if the geometric description does not match the preset geometric shape template, for example, matching the circular feature of a pit to the shape template of a crack, then the disease area is determined not to belong to the current disease type; if the geometric description matches the preset geometric shape template, it is determined to be the current disease type, for example, determining whether the disease area is a crack or a pit. Simultaneously, the corresponding values ​​in the evaluation indicators are extracted and matched with the current... Comparison of the corresponding grading threshold range under the disease type: If the value is not within the grading threshold range, it means that it belongs to the sub-category of the disease type; if the value is within the grading threshold range, it means that it does not belong to the sub-category of the disease type; perform maintenance level determination on the determined sub-category; for example: taking the road sub-model as an example, the output disease types include crack type, loose type, deformation type, settlement type and other types. The sub-categories of crack type include transverse crack, longitudinal crack, alligator crack and block crack. The sub-categories of loose type include pothole and loose. The sub-categories of deformation type include rutting, settlement and wave bulge. Other types include bleeding and repair.

[0079] Maintenance level determination: First, the determined sub-category is matched with the maintenance rules under the first maintenance level. If no matching rule exists for the first maintenance level, the sub-category is matched with the maintenance rules under the second maintenance level. If no matching rule exists for the second maintenance level, the sub-category is matched with the maintenance rules under the third maintenance level. If a matching maintenance rule exists for the first, second, or third maintenance level, the corresponding maintenance level label is obtained, and maintenance semantic embedding is performed on the diseased area. It should be noted that the first maintenance level is routine maintenance, the second maintenance level is preventive maintenance, and the third maintenance level is restorative maintenance. The maintenance rules corresponding to each level are all predefined settings.

[0080] Based on the defect vector group, a topology mapping index file corresponding to the target rendered road is generated to establish the inverse transformation relationship between the three-dimensional spatial coordinates of the defect geometric identifier and the mileage station coordinates of the preset topology space. Based on the topology mapping index file, the maintenance semantic data corresponding to the maintenance semantic identifier is mapped back to the topology space. The topology space is divided into K maintenance management grids according to the maintenance level. Defect areas with maintenance semantics are stored in the K maintenance management grids according to the maintenance level. Data interaction is performed, that is, the defect areas with completed semantic attachment are assigned to the corresponding maintenance management grids to drive the subsequent maintenance work order generation. Here, K is a positive integer greater than 0. The target rendered road and the mapped topology data are integrated to form a digital twin application system.

[0081] The significance of the above analysis lies in the following: Based on the target rendering of the road, the defect area is accurately identified. Under the condition of identifying the defect area, the matching mechanism of the maintenance rule engine is triggered. Through the determination of defect type and maintenance level, maintenance semantic embedding is performed on the defect area to generate defect vector groups. Through the topology mapping index file, these vector groups are mapped back to the topology space, so that dynamically discovered defects can be accurately traced back to the highway mileage markers. This ensures the data consistency of the digital twin application system and improves the intelligence level and response efficiency of highway maintenance decision-making.

[0082] Example 2:

[0083] This invention provides a system for constructing a digital twin application system for highway maintenance. The system includes a first generation module, a second generation module, and an application system construction module, and the first generation module, the second generation module, and the application system construction module are communicatively connected.

[0084] The first generation module: collects multi-source perception data of highway sections in real time, and introduces a constraint engine to construct a three-dimensional entity of the highway, generating several three-dimensional sub-image models; among them, the multi-source perception data includes highway images, highway laser point clouds, and component attributes;

[0085] The second generation module: Based on the 3D sub-image model, it abstracts and constructs a topological space, uses an interpolation algorithm to generate a baseline curve, and combines the mesh vertices of the 3D sub-image model to identify the envelope boundary in order to perform cross-sectional scanning to generate the target rendering road;

[0086] Application system construction module: Based on the target rendering road, identify the diseased areas, and introduce the preset maintenance rules based on the diseased areas. Through semantic mounting, map back to the topology space to form a digital twin application system.

[0087] Example 3:

[0088] Embodiment 3 of the present invention provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, it implements a method for constructing a digital twin application system for highway maintenance provided in Embodiment 1.

[0089] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0090] In a possible implementation, the present invention can also be implemented as a program product comprising program code, which, when the program product is run on a terminal device, is used to cause the terminal device to execute the steps of implementing the method for constructing a digital twin application system for highway maintenance in Embodiment 1.

[0091] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for constructing a digital twin application system for highway maintenance, characterized in that, This method includes: real-time acquisition of multi-source perception data of highway sections, and the introduction of a constraint engine to construct a 3D solid model of the highway, generating several 3D sub-image models; wherein, the multi-source perception data includes highway images, highway laser point clouds, and component attributes; wherein, the introduction of a constraint engine to construct a 3D solid model of the highway and generate several 3D sub-image models includes: the constraint engine has a first constraint on the 3D space and a second constraint on computing resources; the first constraint includes the field of view, depth range, and occlusion culling threshold, and the second constraint includes the maximum number of polygon faces and the lower limit of the number of frames transmitted per second; a first rendering model is pre-constructed, the multi-source perception data is imported into the first rendering model, feature deconstruction is performed to extract multiple basic features, and key features are selected by combining the first and second constraints; an adaptive training layer is introduced into the first rendering model, and... The selected key features are combined with the original basic features to form a new training dataset. This dataset is then used to train and update the adaptive training layer, resulting in the second rendering model. Simultaneously, the data is layered according to time series, with each time series labeled as a rendering process. The multi-source perception data from each rendering process is imported into the second rendering model to generate corresponding 3D sub-image models, including slope sub-models, road surface sub-models, bridge and tunnel sub-models, and ancillary facility sub-models. Based on the 3D sub-image models, a topological space is abstractly constructed. An interpolation algorithm is used to generate a baseline curve, and the envelope boundary is identified by combining the mesh vertices of the 3D sub-image models to perform cross-sectional scanning and generate the target rendering road. Based on the target rendering road, diseased areas are identified, and preset maintenance rules are introduced based on these areas. These rules are then semantically mapped back to the topological space, forming a digital twin application system.

2. The method for constructing a digital twin application system for highway maintenance according to claim 1, characterized in that: The first rendering model is based on a hybrid neural network architecture.

3. The method for constructing a digital twin application system for highway maintenance according to claim 1, characterized in that: The process of constructing the target rendering road includes: parsing the 3D sub-image model, extracting the road surface centerline and key component structural points as target construction nodes, establishing a directed topological graph structure, using the B-spline interpolation algorithm to smoothly fit the target construction nodes, and generating the target baseline curve; identifying the mesh vertices of the slope sub-model and the ancillary facility sub-model, filtering the maximum projection distance and maximum Euclidean distance of all mesh vertices under each road segment relative to the target baseline curve to generate the envelope boundary, and combining the target baseline curve to perform a cross-sectional scan on the envelope boundary to generate the target rendering road.

4. The method for constructing a digital twin application system for highway maintenance according to claim 3, characterized in that: Perform a cross-sectional scan on the envelope boundary, including circular and rectangular scans; identify the component attributes to which the current target construction node belongs; if identified as a tunnel, perform a circular scan: extract the maximum projection distance to determine the envelope radius, generate a circular cross-section, and sample and obtain M circular contour point sets according to a preset step size; otherwise, perform a rectangular scan: extract the maximum Euclidean distance to determine the envelope half-width and envelope height, generate a rectangular cross-section, and perform equal-interval interpolation on the four sides of the rectangular contour to generate a total of M rectangular contour point sets; where M is greater than 0; sequentially extract the cross-sectional contour point sets corresponding to two adjacent target construction nodes, perform triangulation, close the head and tail of the generated triangular facets to form the target rendering road.

5. The method for constructing a digital twin application system for highway maintenance according to claim 1, characterized in that: The formation of a digital twin application system includes: separating texture features and geometric features from the target rendered road; determining a first abnormal region based on the texture features and a second abnormal region based on the geometric features; performing a union calculation on the first and second abnormal regions to obtain the corresponding disease regions, and extracting the boundary coordinates of the disease regions to generate disease geometric identifiers; wherein the boundary coordinates are in three-dimensional spatial coordinate form; simultaneously, extracting the crack width, damaged area, and pothole depth of the disease regions as evaluation indicators based on the boundary coordinates; generating at least one disease vector group based on the geometric identifiers, including the entered disease geometric identifiers and the maintenance identifiers to be acquired; importing the evaluation indicators into a preset maintenance rule engine, and outputting the corresponding disease type and maintenance level through rule matching, defining them as maintenance semantic data, and generating corresponding maintenance identifiers to be entered into the disease vector group; generating a topology mapping index file corresponding to the target rendered road based on the disease vector group; mapping the maintenance semantic data corresponding to the maintenance semantic identifiers back into the topology space based on the topology mapping index file, and integrating the target rendered road with the mapped topology data to form a digital twin application system.

6. The method for constructing a digital twin application system for highway maintenance according to claim 5, characterized in that, The preset maintenance rule engine includes disease type determination and maintenance level determination.

7. A system for constructing a digital twin application system for highway maintenance, characterized in that, The system includes: a first generation module: real-time acquisition of multi-source perception data of highway sections, and the introduction of a constraint engine to construct a 3D solid model of the highway, generating several 3D sub-image models; wherein, the multi-source perception data includes highway images, highway laser point clouds, and component attributes; wherein, the constraint engine, which constructs a 3D solid model of the highway and generates several 3D sub-image models, includes: the constraint engine has a first constraint on the 3D space and a second constraint on computing resources; the first constraint includes the field of view, depth range, and occlusion culling threshold, and the second constraint includes the maximum number of polygon faces and the lower limit of the number of frames transmitted per second; a first rendering model is pre-built, the multi-source perception data is imported into the first rendering model, feature deconstruction is performed to extract multiple basic features, and key features are selected by combining the first and second constraints; an adaptive training layer is introduced into the first rendering model to train the selected key features. The features are combined with the original basic features to form a new training dataset, which is used to train and update the adaptive training layer to obtain the second rendering model. At the same time, the data is layered according to time series, and each time series is marked as a rendering process. The multi-source perception data under each rendering process is imported into the second rendering model to generate corresponding 3D sub-image models, including slope sub-models, road surface sub-models, bridge and tunnel sub-models, and ancillary facility sub-models. The second generation module: Based on the 3D sub-image models, a topological space is abstractly constructed, a baseline curve is generated using an interpolation algorithm, and the envelope boundary is identified by combining the mesh vertices of the 3D sub-image models to perform cross-section scanning to generate the target rendering road. The application system construction module: Based on the target rendering road, the disease area is identified, and a preset maintenance rule is introduced based on the disease area. The semantic mapping is then used to map back to the topological space to form a digital twin application system.

8. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements a method for constructing a digital twin application system for highway maintenance as described in any one of claims 1-6.

Images