Intelligent metro tunnel cavity disease disposal method and device based on BIM and machine learning

Abstract

The application discloses a subway tunnel cavity disease intelligent treatment method and device based on BIM and machine learning. The method builds a BIM model containing tunnel three-dimensional structure information according to subway tunnel design and completion data, and builds a global unified coordinate system matched with a field measurement coordinate system. Three-dimensional center coordinates, contour point clouds and disease severity indexes of cavity diseases are collected based on the coordinate system, and a structured disease data package is generated and returned to a server. The server generates a three-dimensional disease mark at a corresponding position of the BIM model according to disease coordinates, binds geometric, hydrological and detection time attributes to form an integrated disease database. Multidimensional feature vectors of the diseases are extracted and input into a machine learning model to determine disease risks and recommend repair parameters. In combination with skylight length and team quantity constraints, maintenance task clusters are divided and optimal operation routes are generated through clustering and path planning, and a visual maintenance plan is output. The application solves the problems of non-uniform coordinate reference and data island in the prior art.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance management technology for urban rail transit infrastructure, specifically to an intelligent treatment method and device for cavity defects in subway tunnels based on BIM and machine learning. Background Technology

[0002] As the core underground structure of urban rail transit, subway tunnels are subjected to the dynamic load of trains, creep of surrounding rock, and erosion of groundwater over a long period of time. The back of the lining is prone to forming loose filling, voids and cavities, which can gradually induce water seepage, cracking, spalling and even local collapse of the lining, directly threatening the safety of train operation. Routine detection and maintenance of cavity defects has become a core business of rail operation and maintenance. Currently, the industry has formed a multi-faceted parallel application technology system: at the on-site inspection level, ground-penetrating radar, infrared thermal imagers, and 3D laser scanners are commonly used in combination for non-destructive testing of linings, relying on manual signal interpretation to identify cavity defects, and simultaneously using total stations and GNSS equipment to complete simple location of defect points; at the data management level, some lines have introduced 2D as-built drawings, simple Excel ledgers, or lightweight BIM models to separately store textual and image records of defects; at the intelligent analysis level, a small number of studies are attempting to use basic machine learning algorithms to identify defects in radar images; at the operation and maintenance scheduling level, maintenance personnel rely on nighttime shutdown windows to carry out defect repairs, manually dividing maintenance sections and scheduling work orders based on line mileage and human experience. Meanwhile, digital technologies such as digital twins, BIM, IoT sensing, and artificial intelligence are gradually penetrating the tunnel operation and maintenance field. Existing patents and engineering solutions have optimized detection equipment, defect ledger management, and single image recognition, respectively, initially achieving digital upgrades in some aspects and providing a technical foundation for intelligent management of tunnel defects.

[0003] However, existing technical solutions mostly focus on optimizing individual aspects of detection, storage, and analysis, failing to form an integrated technical system that connects the entire process. This results in several core pain points in engineering implementation: First, inconsistent spatial benchmarks create information silos. The positioning coordinates of detection equipment, local coordinates of the BIM model, and the independent coordinate system of the on-site project lack standardized three-layer transformation mathematical relationships. Defect locations rely solely on manual comparison with drawings and manual input, making it impossible to achieve automatic and accurate mapping between detection data and the 3D model. This makes defect location tracing and multi-dimensional statistical analysis difficult. Second, defect analysis relies on subjective human experience. Existing solutions simply classify defect levels based on fixed thresholds, without integrating multi-dimensional features such as defect geometry, lining structure, seepage, and operational accessibility to train a grading model. Risk assessment, repair processes, and construction time and materials are also problematic. The first problem is the lack of quantitative data to support the forecast, which hinders the scientific basis of maintenance decisions. The second problem is the inefficient and inefficient scheduling of maintenance resources during maintenance windows. The limited daily available maintenance window time for the subway means that the current scheduling methods do not use maintenance window hours as a rigid constraint. They also fail to utilize clustering and path optimization algorithms to coordinate the spatial distribution of defects, risk priorities, and team capacity. This results in repetitive back-and-forth work paths and the common practice of splitting maintenance work for defects within the same section, leading to low utilization of effective maintenance hours within maintenance windows. The third problem is the fragmented data flow, which prevents automated closure. The detection and acquisition, model mounting, intelligent assessment, and work order output processes are independent of each other. Offline export of detection data and manual secondary transcription are prone to errors, making it difficult to achieve fully automated workflow for defect perception, analysis, decision-making, and work assignment. Overall, there is significant room for improvement in the efficiency of defect management and the level of refined management.

[0004] Therefore, there is an urgent need for an intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning to solve the problems of inconsistent coordinate benchmarks, data silos, reliance on manual experience for defect assessment, extensive maintenance scheduling during track maintenance windows, and fragmented data flow in various processes that exist in existing technologies. Summary of the Invention

[0005] To address these issues, this invention provides a method and device for intelligent handling of cavity defects in subway tunnels based on BIM and machine learning. This method aims to solve problems such as inconsistent coordinate benchmarks, data silos, reliance on manual experience for defect assessment, crude maintenance scheduling during track maintenance windows, fragmented data flows across different processes, and a lack of an integrated intelligent handling system for the entire process in the current operation and maintenance of cavity defects in subway tunnels.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning, comprising:

[0007] Based on the subway tunnel design and as-built data, a building information model containing three-dimensional information about the tunnel structure is constructed; a globally unified coordinate system aligned with the coordinate system of the actual engineering measurements is constructed in the building information model;

[0008] Based on the global unified coordinate system, the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index are captured and recorded, a structured disease database is generated and sent back to the server.

[0009] Based on the three-dimensional coordinates embedded in the disease database, a three-dimensional disease identifier is generated at the corresponding spatial location in the building information model. The geometric, hydrological, and detection time attributes of the disease are bound and entered into the structured disease database to generate an integrated disease database.

[0010] Extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, use clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and output a visualized complete set of maintenance operation plans.

[0011] As a preferred solution for the intelligent treatment method of cavity defects in subway tunnels based on BIM and machine learning, in the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the coordinate origin configuration of the BIM software, and the horseshoe-shaped tunnel section segment width calculation function is implanted to realize the binding of tunnel section geometry with global three-dimensional coordinates.

[0012] The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension;

[0013] The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix;

[0014] The expression for the Gauss-Kruger projection forward calculation model is as follows:

[0015]

[0016]

[0017]

[0018] In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoidal coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gaussian projection, respectively.

[0019] The expression for the local independent coordinate system translation scale correction formula is as follows:

[0020]

[0021] In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , denoted as the translation constant of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system.

[0022] The expression for the seven-parameter coordinate transformation matrix is:

[0023] =

[0024] In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software.

[0025] The expression for the elevation datum transformation model is:

[0026]

[0027] In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software.

[0028] The function for calculating the segment width of the horseshoe-shaped tunnel section is:

[0029] Arch section:

[0030]

[0031]

[0032] Straight wall section:

[0033]

[0034]

[0035] Inverted arch section:

[0036]

[0037]

[0038] In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

[0039] As a preferred solution for the intelligent treatment method of cavity defects in subway tunnels based on BIM and machine learning, in the process of capturing and recording the three-dimensional coordinates of the center point, the contour point cloud data and the severity index of the cavity defect based on the global unified coordinate system, when the cavity defect is identified, the three-dimensional coordinates of the center point of the defect are recorded; the contour point cloud of the cavity defect on the tunnel surface boundary is obtained by scanning with infrared equipment; the maximum cavitation depth is inverted according to the radar signal; and the contour point cloud inside the cavity is obtained by laser signal.

[0040] As a preferred solution for the intelligent treatment method of cavity defects in subway tunnels based on BIM and machine learning, the multi-dimensional feature vectors of each defect extracted from the integrated defect database include: lining defect type encoding, seepage state characteristics, effective lining thickness, longitudinal extension length of defect, circumferential expansion width of defect, quantitative value of tunnel surrounding rock grade, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of defect from the nearest maintenance entrance / exit.

[0041] As a preferred solution for intelligent treatment of cavity defects in subway tunnels based on BIM and machine learning, in the process of determining the risk level of defects and recommending repair scheme parameters through the pre-trained machine learning model, the pre-trained machine learning model is a multi-class GBDT gradient boosting tree model; an accumulation tree function is constructed for the five types of defect levels:

[0042]

[0043] In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category;

[0044] The multi-class GBDT gradient boosting tree model outputs five risk levels corresponding to predicted probability vectors, and the category with the highest probability is taken as the disease determination level.

[0045] The criteria for determining the five risk levels are as follows:

[0046] Grade I: Loose filling with no damp stains or standing water in the lining;

[0047] Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm;

[0048] Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm;

[0049] Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm;

[0050] Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm;

[0051] The recommended prediction model for the repair scheme parameters is a multi-output regression model; the expression for the multi-output regression model is:

[0052]

[0053] In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials.

[0054] As a preferred embodiment of the intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning, in the process of grouping maintenance task clusters and generating optimal operation sequences for single task clusters through the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is:

[0055]

[0056]

[0057] In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. Space-time balance weights;

[0058] The path planning algorithm is a genetic algorithm used to solve the NP-Hard tunnel maintenance problem with risk priority.

[0059] The objective function of the genetic algorithm is:

[0060]

[0061] In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

[0062] This invention also provides an intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, employing the above-mentioned intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning, including:

[0063] The model and coordinate management module is used to construct a building information model containing three-dimensional information about the tunnel structure based on the subway tunnel design and as-built data; and to construct a globally unified coordinate system in the building information model that is aligned with the coordinate system of the actual engineering measurement.

[0064] The intelligent acquisition and transmission module is used to capture and record the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index based on the global unified coordinate system, generate a structured disease database and transmit it back to the server.

[0065] The data mapping and integration module is used to generate a three-dimensional disease identifier in the corresponding spatial location of the building information model based on the three-dimensional coordinates embedded in the disease database, bind the geometric, hydrological and detection time attributes of the disease and enter them into the structured disease database to generate an integrated disease database.

[0066] The intelligent decision support module is used to extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, the module uses clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and outputs a visualized complete set of maintenance operation plans.

[0067] As a preferred solution for a smart treatment device for cavity defects in subway tunnels based on BIM and machine learning, in the model and coordinate management module, during the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the coordinate origin configuration of the BIM software, and the horseshoe-shaped tunnel section segment width calculation function is implanted to realize the binding of tunnel section geometry with global three-dimensional coordinates.

[0068] The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension;

[0069] The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix;

[0070] The expression for the Gauss-Kruger projection forward calculation model is as follows:

[0071]

[0072]

[0073]

[0074] In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoidal coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gaussian projection, respectively.

[0075] The expression for the local independent coordinate system translation scale correction formula is as follows:

[0076]

[0077] In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , denoted as the translation constant of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system.

[0078] The expression for the seven-parameter coordinate transformation matrix is:

[0079] =

[0080] In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software.

[0081] The expression for the elevation datum transformation model is:

[0082]

[0083] In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software.

[0084] The function for calculating the segment width of the horseshoe-shaped tunnel section is:

[0085] Arch section:

[0086]

[0087]

[0088] Straight wall section:

[0089]

[0090]

[0091] Inverted arch section:

[0092]

[0093]

[0094] In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

[0095] As a preferred solution for an intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, the intelligent acquisition and transmission module, in the process of capturing and recording the three-dimensional coordinates of the center point, the contour point cloud data, and the severity index of the cavity defect based on the global unified coordinate system, records the three-dimensional coordinates of the center point of the defect when the cavity defect is identified; obtains the contour point cloud of the cavity defect on the tunnel surface boundary through infrared scanning; retrieves the maximum cavitation depth based on radar signals; and obtains the contour point cloud inside the cavity through laser signals.

[0096] As a preferred solution for an intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, the intelligent decision support module extracts the multi-dimensional feature vector of each defect from the integrated defect database, including: lining defect type encoding, seepage state characteristics, effective lining thickness, longitudinal extension length of the defect, circumferential expansion width of the defect, quantitative value of tunnel surrounding rock grade, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of the defect from the nearest maintenance entrance / exit.

[0097] As a preferred solution for an intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, the intelligent decision support module, in the process of determining the defect risk level and recommending repair scheme parameters through the pre-trained machine learning model, uses a multi-class GBDT gradient boosting tree model; a cumulative tree function is constructed for the five defect levels:

[0098]

[0099] In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category;

[0100] The multi-class GBDT gradient boosting tree model outputs five risk levels corresponding to predicted probability vectors, and the category with the highest probability is taken as the disease determination level.

[0101] The criteria for determining the five risk levels are as follows:

[0102] Grade I: Loose filling with no damp stains or standing water in the lining;

[0103] Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm;

[0104] Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm;

[0105] Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm;

[0106] Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm;

[0107] The recommended prediction model for the repair scheme parameters is a multi-output regression model; the expression for the multi-output regression model is:

[0108]

[0109] In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials.

[0110] As a preferred embodiment of the intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, in the intelligent decision support module, during the process of grouping maintenance task clusters and generating optimal operation sequences for single task clusters through the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is:

[0111]

[0112]

[0113] In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. Space-time balance weights;

[0114] The path planning algorithm is a genetic algorithm used to solve the NP-Hard tunnel maintenance problem with risk priority.

[0115] The objective function of the genetic algorithm is:

[0116]

[0117] In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

[0118] The present invention has the following advantages:

[0119] First, it connects data flow and achieves full-process automation: Through a unified spatiotemporal benchmark, it seamlessly connects all aspects of design, testing, management and decision-making, eliminates errors from manual transcription, and improves data flow efficiency by more than 70%.

[0120] Second, constructing a digital twin enables panoramic visualization of management: integrating the spatiotemporal evolution information of all defects into the 3D BIM model provides managers with an intuitive "one-map" decision-making dashboard, achieving transparent and refined management of tunnel health status.

[0121] Third, empower AI algorithms to achieve scientific decision-making: transform the experience of domain experts into iteratively optimized machine learning models to quantitatively assess and intelligently recommend disease risks and repair solutions, shifting maintenance decisions from subjective experience-based judgments to objective data-driven approaches.

[0122] Fourth, focusing on operational constraints to maximize maintenance efficiency: The optimization model uniquely incorporates the duration of maintenance windows as a hard constraint. Through intelligent clustering and path planning, it automatically generates refined maintenance work orders that match "time-space-task", which is expected to improve the utilization rate of effective working time during maintenance windows by more than 30% and significantly reduce operation and maintenance costs. Attached Figure Description

[0123] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0124] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0125] Figure 1 This is a flowchart illustrating the intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning provided in Embodiment 1 of the present invention.

[0126] Figure 2 This is a schematic diagram illustrating the specific implementation process of the intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning provided in Embodiment 1 of the present invention.

[0127] Figure 3 This is a detailed flowchart illustrating the intelligent analysis and maintenance planning stage of the intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning provided in Embodiment 1 of the present invention.

[0128] Figure 4 This is a schematic diagram of the architecture of the intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning provided in Embodiment 2 of the present invention.

[0129] Figure 5 This is a schematic diagram of the specific composition architecture and data flow of the intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning provided in Embodiment 2 of the present invention. Detailed Implementation

[0130] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0131] Example 1

[0132] See Figure 1 and Figure 2 Embodiment 1 of the present invention provides an intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning, including the following steps:

[0133] S1. Construct a building information model containing three-dimensional information about the tunnel structure based on the subway tunnel design and as-built data; construct a globally unified coordinate system in the building information model that is aligned with the coordinate system of the actual engineering measurement.

[0134] S2. Based on the global unified coordinate system, capture and record the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index, generate a structured disease database and send it back to the server.

[0135] S3. Based on the three-dimensional coordinates embedded in the disease database, generate a three-dimensional disease identifier at the corresponding spatial location in the building information model, bind the geometric, hydrological and detection time attributes of the disease and enter them into the structured disease database to generate an integrated disease database.

[0136] S4. Extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, use clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and output a visualized complete set of maintenance operation plans.

[0137] In this embodiment, in step S1, a building information model containing three-dimensional information of the tunnel structure is constructed based on the subway tunnel design and as-built data; a globally unified coordinate system aligned with the engineering field measurement coordinate system is constructed in the building information model.

[0138] Specifically, based on the original design drawings and as-built data of the subway tunnel, a high-precision three-dimensional digital model of the tunnel is constructed using BIM modeling tools to fully reflect key structures such as the initial support and secondary lining. From the initial creation of this model, a globally unified coordinate system strictly consistent with the measurement coordinate system used on the construction site (such as a local independent coordinate system) is incorporated as the benchmark for all spatial data association.

[0139] Subway projects generally adopt local independent plane rectangular coordinate system: a custom central meridian, origin, and scale ratio. In the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the coordinate origin configuration of the BIM software, and the horseshoe-shaped tunnel cross-section segment width calculation function is implanted to realize the binding of tunnel cross-section geometry with global three-dimensional coordinates.

[0140] The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension;

[0141] The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix;

[0142] The expression for the Gauss-Kruger projection forward calculation model is as follows:

[0143]

[0144]

[0145]

[0146] In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoid coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gauss projection, respectively.

[0147] The expression for the local independent coordinate system translation scale correction formula is as follows:

[0148]

[0149] In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , , where are the translation constants of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system.

[0150] In this embodiment, all tunnel initial support, secondary lining, axis, and monitoring points are automatically mapped to the measured global coordinates on site using a seven-parameter coordinate transformation matrix, achieving coordinate unification between drawings, as-built surveys, and the BIM model. The expression for the seven-parameter coordinate transformation matrix is:

[0151] =

[0152] In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software.

[0153] In this embodiment, the subway adopts the 1985 National Elevation Datum. The tunnel structure includes multiple elevation layers such as initial support, secondary lining, rail surface, and bottom slab. The expression for the elevation datum conversion model is as follows:

[0154]

[0155] In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software.

[0156] In this embodiment, the function for calculating the segment width of the horseshoe-shaped tunnel section is:

[0157] Arch section:

[0158]

[0159]

[0160] Straight wall section:

[0161]

[0162]

[0163] Inverted arch section:

[0164]

[0165]

[0166] In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

[0167] In this embodiment, in step S2, based on the global unified coordinate system, the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index are captured and recorded, a structured disease database is generated and sent back to the server.

[0168] Specifically, a unified coordinate system is integrated into the positioning and attitude determination system and data acquisition software of the mobile detection platform. During the detection process, the detection equipment uses the aforementioned unified coordinate system for real-time positioning. When a cavity defect is identified, the three-dimensional coordinates of the defect's center point are automatically recorded; the point cloud of the cavity defect's boundary contour on the tunnel surface is obtained by infrared scanning; the maximum cavitation depth is inverted from radar signals; and the internal contour point cloud of the cavity is obtained from laser signals. These are packaged into a standard-format structured defect database and transmitted back to the server.

[0169] In this embodiment, in step S3, a three-dimensional disease identifier is generated at the corresponding spatial location of the building information model based on the three-dimensional coordinates embedded in the disease database. The geometric, hydrological, and detection time attributes of the disease are bound and entered into the structured disease database to generate an integrated disease database.

[0170] Specifically, after receiving the defect data packet, the central server drives the BIM engine to automatically generate a 3D defect body at the corresponding location in the model based on its coordinates. The internal color is defined according to water content characteristics, and attributes such as volume, depth, and detection time are bound to it. All defects form a dynamically updatable "disease panorama" in the BIM model, supporting multi-dimensional queries and statistical chart generation by interval, type, time, risk level, and other dimensions.

[0171] In this embodiment, in step S4, a multi-dimensional feature vector for each disease is extracted from the integrated disease database; the multi-dimensional feature vector is input into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, the maintenance task clusters are grouped and the optimal operation sequence for a single task cluster is generated through clustering algorithms and path planning algorithms, and a visualized complete set of maintenance operation plans is output.

[0172] Specifically, such as Figure 3 As shown, the multi-dimensional feature vectors extracted from the integrated disease database for each disease include: lining defect type encoding, seepage status characteristics, effective lining thickness, longitudinal extension length of the defect, circumferential expansion width of the defect, quantified value of the surrounding rock grade of the tunnel, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of the disease from the nearest maintenance entrance / exit.

[0173] In this embodiment, during the process of determining the disease risk level and recommending remediation parameters using the pre-trained machine learning model, the pre-trained machine learning model is a multi-class GBDT gradient boosting tree model; an accumulation tree function is constructed for the five disease levels:

[0174]

[0175] In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category.

[0176] The multi-class GBDT gradient boosting tree model outputs predicted probability vectors corresponding to five risk levels, and the category with the highest probability is taken as the disease determination level.

[0177] The criteria for determining the five risk levels are as follows:

[0178] Grade I: Loose filling with no damp stains or standing water in the lining;

[0179] Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm;

[0180] Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm;

[0181] Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm;

[0182] Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm.

[0183] In this embodiment, the prediction model recommended for the repair scheme parameters is a multi-output regression model; the expression of the multi-output regression model is:

[0184]

[0185] In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials.

[0186] In this embodiment, during the process of grouping maintenance task clusters and generating optimal work sequences for single task clusters using the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is:

[0187]

[0188]

[0189] In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. The space-time balance weight.

[0190] In this embodiment, the path planning algorithm is a genetic algorithm used to solve the NP-Hard problem of tunnel maintenance with risk priority; the specific steps of the solution are as follows:

[0191] T1, Population Initialization

[0192] An initial population is generated according to sequential coding rules, and the overall population size is predetermined. Each chromosome corresponds to a set of maintenance path schemes. The genes within a chromosome represent the order in which diseases are visited during maintenance. All disease numbers within a chromosome are unique to ensure that the path arrangement is legal and compliant. A specified number of independent arrangement schemes are generated in batches to form the first generation initial population.

[0193] T2, Individual Fitness Calculation

[0194] Substitute each chromosome in the population into the path synthesis objective function, calculate the path loss value, and then convert it into fitness for evaluating the quality of individuals.

[0195] The objective function of the genetic algorithm is:

[0196]

[0197] In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

[0198] The fitness function is:

[0199]

[0200] In the formula, For individual fitness.

[0201] T3, Roulette Selection Operation

[0202] Selection probabilities are assigned based on individual fitness, with individuals of higher fitness having a higher probability of inheritance. The initial selection of the offspring population is completed by traversing the entire population. The probabilities of all individuals are summed to divide the population into roulette intervals, and values ​​are randomly selected to determine the selected individuals. This process is repeated M times to obtain the intermediate population after selection.

[0203] Individual choice probability:

[0204]

[0205] In the formula, Let be the probability that the j-th individual is selected to participate in the inheritance; M is the total population size.

[0206] T4. Pair the selected individuals into pairs and use sequential crossover to generate new offspring individuals, ensuring that the chromosomes are always valid, non-repetitive sequences.

[0207] Specifically, a pair of parent chromosomes are randomly selected, two unequal gene cut points are chosen, and gene fragments between the parent 1 cut points are directly fixed to the corresponding positions in the offspring; according to the gene sequence of parent 2, the fixed genes are removed and the offspring gaps are filled in sequence to complete the generation of a set of offspring. Crossover is performed on all paired populations to obtain a crossover population.

[0208] T5, Exchange Mutation Operation

[0209] Set a fixed mutation probability, perform mutation on each individual in the population after crossover, and slightly shuffle the sequence to escape local optima.

[0210] Specifically, each chromosome is traversed, and a random number between 0 and 1 is generated. If the random number is less than the preset fixed mutation probability, two gene sites are randomly selected on the chromosome, and the positions of the two disease numbers are swapped. If the probability condition is not met, the chromosome remains unchanged, and finally a new generation of complete population is obtained.

[0211] T6, Iteration Termination Determination

[0212] Determine if the number of iterations has reached the preset maximum number of iterations:

[0213] If the upper limit is not reached: use the new generation population as the initial population and return to step T2 to iterate again;

[0214] Reaching the upper limit: Select the chromosome corresponding to the maximum fitness in the final population and analyze it to obtain the optimal disease repair operation sequence.

[0215] In one possible implementation, a verification example is provided as follows:

[0216] This embodiment selects a subway tunnel section maintenance project in a certain city to fully implement the present invention, and deploys the intelligent handling system of the present invention. The system is built with a B / S architecture. The backend relies on microservices to carry coordinate transformation, feature extraction, machine learning inference, clustering and path planning algorithms, while the frontend uses WebGL technology to complete the visualization rendering of BIM 3D model, defect information and maintenance route.

[0217] I. Specific Implementation Steps:

[0218] M1. Construct a tunnel BIM digital base equipped with a unified global coordinate system.

[0219] The operation and maintenance unit retrieved the complete set of as-built drawings, structural design data, and line control network measurement results reports for the target tunnel section. Within the BIM modeling platform, a 3D BIM model was constructed, including the tunnel excavation outline, initial support, and secondary lining geometric dimensions and material parameters. During the modeling phase, the mathematical relationships of three layers of transformation—geocentric latitude and longitude, local independent plane coordinates, and BIM local coordinates—were embedded, along with the 1985 National Elevation Transformation Model. The global reference coordinate system of the model was verified and locked to be uniformly adopted with the Beijing 54 local independent coordinate system for the on-site control network. The segmented geometric functions of the horseshoe-shaped tunnel section were implanted, and the complete set of coordinate transformation parameters was exported as a standardized configuration file, serving as the sole spatial reference for subsequent equipment positioning and spatial mapping of defects.

[0220] After M2 coordinate synchronization, it is equipped with multi-level collaborative detection equipment to complete on-site disease collection and data packet transmission.

[0221] The coordinate system configuration file output from M1 is imported into the intelligent acquisition and transmission module built into the inspection vehicle to synchronize the coordinate parameters of the positioning unit and data acquisition unit of the inspection equipment. The inspection vehicle moves at a constant speed along the tunnel track, and the onboard infrared thermal imager completes a rapid full-area scanning and screening of the lining surface. The ground-penetrating radar conducts coarse detection of cavities on the back of the lining in abnormal areas identified by infrared imaging, and the laser scanner collects millimeter-level inner and outer contour point clouds of the cavity areas confirmed by radar. The equipment automatically calculates the three-dimensional coordinates of the cavity center point, the maximum void depth, the water seepage and water content, etc. The system detects indicators such as confidence level and automatically encapsulates and generates a unique structured disease data packet, using the disease ID as the primary key identifier. In this embodiment, the typical data packet content is {Disease ID: T-20260701-001, Center 3D Coordinates: (4056123.156, 504321.789, -15.234), Cavity Contour Point Cloud: External Storage File Link, Maximum Hollowing Depth: 0.12m, Detection Confidence Level: 0.95}. The data packet is transmitted back to the central server in real time through the tunnel 5G industrial private network.

[0222] M3 and the server automatically map the BIM model and build an integrated disease database.

[0223] The server's data mapping and integration module receives the defect data package, reads the embedded global 3D coordinates within the data package, and completes coordinate matching based on the M1 unified coordinate benchmark. It then automatically generates an ellipsoidal 3D defect marker at the corresponding spatial location in the BIM model. The system sets differentiated rendering colors for the defect marker based on the water content and seepage status parameters within the data package. Simultaneously, it writes all attributes, such as defect ID, 3D coordinates, point cloud file, void depth, detection time, and lining location, into the database, forming an integrated defect database that can be searched in multiple dimensions and dynamically updated, thus achieving a panoramic 3D visualization of tunnel defects.

[0224] M4. Based on the integrated disease database, machine learning intelligent analysis and global optimization of maintenance tasks during maintenance windows are carried out.

[0225] M41. Automatic extraction of multi-dimensional features of diseases:

[0226] The system backend automatically extracts the complete feature vector of the target disease from the integrated disease database, and automatically calculates the geometric features for data package T-20260701-001: disease volume 0.5m³, lining projection area 1.8m²; identifies structural spatial features: the disease is located in the secondary lining area of ​​the arch; extracts operational reachability features: the straight-line distance of the disease from the maintenance entrance of ventilation shaft No. 2 is 520m, and completes the unified standardization processing of 8 standardized features and operational distance features.

[0227] M42, Machine Learning Model Risk Classification and Intelligent Recommendation of Remediation Solutions:

[0228] The GBDT multi-class risk model and repair scheme prediction model used in this embodiment were pre-trained using thousands of cavity defect treatment samples from the line's history. Hyperparameters were optimized using cross-validation grid search, and incremental learning was performed by adding new defect samples during daily operation and maintenance to continuously stabilize the model's inference accuracy. The standardized feature vector was input into the pre-trained GBDT gradient boosting tree model, which outputs the predicted probability corresponding to the five risk levels. The defect was determined to be a Level I risk: loose filling of the lining without wet stains or standing water. The defect feature vector and the distance from the entrance / exit were input into the repair scheme prediction model, which intelligently outputs a triplet recommendation result: the repair process is low-pressure epoxy resin grouting, the estimated standard working hours for a single defect is 1.8 person-hours, and the estimated amount of grouting material is 0.25 m³.

[0229] M43, Improved K-means clustering for dividing maintenance task clusters.

[0230] The maintenance personnel input the following constraints for this maintenance window: a total available time of 3.5 hours per window and 2 maintenance teams. The system retrieves the coordinates, estimated working hours, and risk level data of a total of 50 cavity defects awaiting treatment along the entire line. With the maintenance window working hours as a hard constraint and minimizing the spatial dispersion within each group and balancing the working hours of each group as optimization objectives, the system uses an improved K-means algorithm to cluster the defects, resulting in 4 independent maintenance task clusters. Among them, task cluster C1, assigned to team A, contains 6 defects (including T-20260701-001), with an estimated total working time of 3.2 hours for all defects within the cluster. The defects have a longitudinal spatial span of 150 meters along the tunnel, which meets the maintenance window working hour constraint.

[0231] M44. Genetic algorithm to solve the optimal maintenance path for a single task cluster.

[0232] For the six defect locations within task cluster C1, a tunnel topology network is constructed using ventilation shaft No. 2 as the starting and ending point for maintenance. A genetic algorithm is used to solve for the optimal work sequence that balances the shortest travel distance and prioritizes the treatment of high-risk defects. The algorithm performs multiple iterations, including sequential encoding population initialization, fitness calculation, roulette wheel selection, sequential crossover, and exchange mutation, and outputs the optimal path after reaching the preset maximum number of iterations: entrance of ventilation shaft No. 2 → high-risk defect location 1 → defect location 2 → defect location 3 → defect location 4 → defect location 5 → target defect T-20260701-001 → return to ventilation shaft No. 2.

[0233] M45, generating visual maintenance work orders, implementing them, and iteratively updating the model.

[0234] The system integrates a task cluster of defects, BIM defect location views, optimal maintenance path navigation, standardized process cards for single defects, and a list of time and materials to generate a complete set of visualized electronic maintenance work orders and push them to the on-site team leader's smart terminal. The team completes all defect repair work within the maintenance window according to the work order. After the repair is completed, the team uploads measured data such as actual construction time, main material consumption, and lining treatment effect through the terminal. The measured data is automatically written back into the integrated defect database and used as new samples to participate in the incremental training of the machine learning model, completing the data closed loop of detection-analysis-repair-model iteration.

[0235] II. Implementation Results:

[0236] This embodiment eliminates data silos across multiple stages, including detection, BIM, and operation and maintenance, by using a unified coordinate system. It leverages machine learning to quantitatively assess defects and repair solutions, and combines track maintenance window constraints to intelligently group and optimize maintenance tasks. Compared to traditional manual ledgers and experience-based scheduling, this approach improves data flow efficiency by over 70% and increases the utilization rate of effective working hours during track maintenance windows by over 30%, significantly reducing the manpower and time costs of maintaining defects in subway tunnel cavities.

[0237] The application scenarios of this invention are as follows:

[0238] In the scenario of acceptance testing of newly built subway tunnels, this invention relies on a unified coordinate BIM model to synchronously collect data on the cavity behind the lining, automatically collect defect information, and output a complete set of maintenance and treatment plans.

[0239] In the scenario of routine nighttime maintenance of subway tunnels during operation, this invention combines the duration of the maintenance window and the resources of the work team to complete the clustering of defects and the arrangement of work paths, and generates electronic maintenance work orders that can be directly issued.

[0240] In the scenario of long-term dynamic monitoring and health management of defects in subway tunnels, this invention continuously accumulates cavity detection data and updates the machine learning model through incremental learning to achieve continuous assessment of defect risks.

[0241] In the scenario of concentrated treatment of defects in long-distance tunnels in urban areas, this invention adapts to local independent coordinate systems to complete the unified BIM spatial mapping of the entire tunnel, analyzes a large number of cavity defects in batches, and coordinates the allocation of maintenance resources.

[0242] In scenarios involving the reconstruction and expansion of subway tunnels and the verification of the effectiveness of defect repair, this invention mounts the cavity detection data before and after repair onto the same BIM model, allowing for a direct comparison of the changes in defect treatment and supplementing the model training samples.

[0243] In the scenario of digital upgrade of rail transit operation and maintenance management platform, this invention adopts B / S microservice architecture to connect with operation and maintenance system, and relies on WebGL to realize three-dimensional visualization of tunnel defects and online intelligent decision-making.

[0244] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0245] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0246] Example 2

[0247] See Figure 4 and Figure 5 Embodiment 2 of the present invention also provides an intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning, comprising:

[0248] The model and coordinate management module 001 is used to construct a building information model containing three-dimensional information of the tunnel structure based on the subway tunnel design and as-built data; and to construct a globally unified coordinate system in the building information model that is aligned with the coordinate system of the actual engineering measurement.

[0249] The intelligent acquisition and transmission module 002 is used to capture and record the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index based on the global unified coordinate system, generate a structured disease database and transmit it back to the server.

[0250] The data mapping and integration module 003 is used to generate a three-dimensional disease identifier in the corresponding spatial location of the building information model based on the three-dimensional coordinates embedded in the disease database, bind the geometric, hydrological and detection time attributes of the disease and enter them into the structured disease database to generate an integrated disease database.

[0251] The intelligent decision support module 004 is used to extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, the module uses clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and outputs a visualized complete set of maintenance operation plans.

[0252] In this embodiment, in the model and coordinate management module 001, during the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the BIM software coordinate origin configuration, and the horseshoe-shaped tunnel cross-section segment width calculation function is implanted to realize the binding of tunnel cross-section geometry with global three-dimensional coordinates.

[0253] The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension;

[0254] The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix;

[0255] The expression for the Gauss-Kruger projection forward calculation model is as follows:

[0256]

[0257]

[0258]

[0259] In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoidal coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gaussian projection, respectively.

[0260] The expression for the local independent coordinate system translation scale correction formula is as follows:

[0261]

[0262] In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , denoted as the translation constant of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system.

[0263] The expression for the seven-parameter coordinate transformation matrix is:

[0264] =

[0265] In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software.

[0266] The expression for the elevation datum transformation model is:

[0267]

[0268] In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software.

[0269] The function for calculating the segment width of the horseshoe-shaped tunnel section is:

[0270] Arch section:

[0271]

[0272]

[0273] Straight wall section:

[0274]

[0275]

[0276] Inverted arch section:

[0277]

[0278]

[0279] In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

[0280] In this embodiment, the intelligent acquisition and transmission module 002, during the process of capturing and recording the three-dimensional coordinates of the center point of the cavity defect, the contour point cloud data, and the severity index based on the global unified coordinate system, records the three-dimensional coordinates of the center point of the defect when the cavity defect is identified; obtains the contour point cloud of the cavity defect on the tunnel surface boundary by scanning with an infrared device; retrieves the maximum cavitation depth based on the radar signal; and obtains the contour point cloud inside the cavity by using a laser signal.

[0281] In this embodiment, the intelligent decision support module 003 extracts the multi-dimensional feature vector of each defect from the integrated defect database, including: lining defect type code, seepage state characteristics, effective lining thickness, longitudinal extension length of defect, circumferential expansion width of defect, quantitative value of tunnel surrounding rock grade, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of defect from the nearest maintenance entrance / exit.

[0282] In this embodiment, in the intelligent decision support module 003, during the process of determining the disease risk level and recommending repair scheme parameters through the pre-trained machine learning model, the pre-trained machine learning model is a multi-class GBDT gradient boosting tree model; an accumulation tree function is constructed for the five disease levels:

[0283]

[0284] In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category;

[0285] The multi-class GBDT gradient boosting tree model outputs five risk levels corresponding to predicted probability vectors, and the category with the highest probability is taken as the disease determination level.

[0286] The criteria for determining the five risk levels are as follows:

[0287] Grade I: Loose filling with no damp stains or standing water in the lining;

[0288] Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm;

[0289] Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm;

[0290] Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm;

[0291] Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm;

[0292] The recommended prediction model for the repair scheme parameters is a multi-output regression model; the expression for the multi-output regression model is:

[0293]

[0294] In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials.

[0295] In this embodiment, in the intelligent decision support module 003, during the process of grouping maintenance task clusters and generating optimal operation sequences for single task clusters using the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is:

[0296]

[0297]

[0298] In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. Space-time balance weights;

[0299] The path planning algorithm is a genetic algorithm used to solve the tunnel maintenance (TSP) problem with risk priority.

[0300] The objective function of the genetic algorithm is:

[0301]

[0302] In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

[0303] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0304] Example 3

[0305] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a method for intelligent treatment of cavity defects in subway tunnels based on BIM and machine learning. The program code includes instructions for executing the method for intelligent treatment of cavity defects in subway tunnels based on BIM and machine learning as described in Embodiment 1 or any possible implementation thereof.

[0306] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0307] Example 4

[0308] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0309] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute the intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning, as described in Embodiment 1 or any possible implementation thereof.

[0310] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0311] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0312] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0313] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for intelligent treatment of cavity defects in subway tunnels based on BIM and machine learning, characterized in that, include: A building information model containing three-dimensional information about the tunnel structure was constructed based on the subway tunnel design and as-built data. A globally unified coordinate system aligned with the coordinate system of on-site engineering measurements is constructed in the building information model; Based on the global unified coordinate system, the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index are captured and recorded, a structured disease database is generated and sent back to the server. Based on the three-dimensional coordinates embedded in the disease database, a three-dimensional disease identifier is generated at the corresponding spatial location in the building information model. The geometric, hydrological, and detection time attributes of the disease are bound and entered into the structured disease database to generate an integrated disease database. Extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, use clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and output a visualized complete set of maintenance operation plans.

2. The intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning according to claim 1, characterized in that, In the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the BIM software coordinate origin configuration, and the horseshoe-shaped tunnel cross-section segment width calculation function is implanted to realize the binding of tunnel cross-section geometry with global three-dimensional coordinates; The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension; The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix; The expression for the Gauss-Kruger projection forward calculation model is as follows: In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoidal coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gaussian projection, respectively. The expression for the local independent coordinate system translation scale correction formula is as follows: In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , denoted as the translation constant of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system. The expression for the seven-parameter coordinate transformation matrix is: = In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software. The expression for the elevation datum transformation model is: In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software. The function for calculating the segment width of the horseshoe-shaped tunnel section is: Arch section: Straight wall section: Inverted arch section: In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

3. The intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning according to claim 2, characterized in that, In the process of capturing and recording the three-dimensional coordinates of the center point, the contour point cloud data, and the severity index of the cavity defect based on the global unified coordinate system, when the cavity defect is identified, the three-dimensional coordinates of the center point of the defect are recorded; the contour point cloud of the cavity defect on the tunnel surface boundary is obtained by scanning with an infrared device; the maximum delamination depth is inverted based on the radar signal; and the contour point cloud inside the cavity is obtained by laser signal.

4. The intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning according to claim 3, characterized in that, The multi-dimensional feature vectors extracted from the integrated disease database for each disease include: lining defect type code, seepage status characteristics, effective lining thickness, longitudinal extension length of the defect, circumferential expansion width of the defect, quantitative value of tunnel surrounding rock grade, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of the disease from the nearest maintenance entrance / exit.

5. The intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning according to claim 4, characterized in that, In the process of determining the disease risk level and recommending remediation parameters using the pre-trained machine learning model, the pre-trained machine learning model is a multi-class GBDT gradient boosting tree model; an accumulation tree function is constructed for the five disease levels: In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category; The multi-class GBDT gradient boosting tree model outputs five risk levels corresponding to predicted probability vectors, and the category with the highest probability is taken as the disease determination level. The criteria for determining the five risk levels are as follows: Grade I: Loose filling with no damp stains or standing water in the lining; Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm; Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm; Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm; Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm; The recommended prediction model for the repair scheme parameters is a multi-output regression model; the expression for the multi-output regression model is: In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials.

6. The intelligent treatment method for cavity defects in subway tunnels based on BIM and machine learning according to claim 5, characterized in that, In the process of grouping maintenance task clusters and generating optimal work sequences for single task clusters using the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is: In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. Space-time balance weights; The path planning algorithm is a genetic algorithm used to solve the NP-Hard tunnel maintenance problem with risk priority. The objective function of the genetic algorithm is: In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

7. A BIM- and machine learning-based intelligent treatment device for cavity defects in subway tunnels, employing the BIM- and machine learning-based intelligent treatment method for cavity defects in subway tunnels as described in any one of claims 1-6, characterized in that, include: The model and coordinate management module is used to construct a building information model containing three-dimensional information about the tunnel structure based on the subway tunnel design and as-built data. A globally unified coordinate system aligned with the coordinate system of on-site engineering measurements is constructed in the building information model; The intelligent acquisition and transmission module is used to capture and record the three-dimensional coordinates of the center point of the cavity disease, the contour point cloud data and the severity index based on the global unified coordinate system, generate a structured disease database and transmit it back to the server. The data mapping and integration module is used to generate a three-dimensional disease identifier in the corresponding spatial location of the building information model based on the three-dimensional coordinates embedded in the disease database, bind the geometric, hydrological and detection time attributes of the disease and enter them into the structured disease database to generate an integrated disease database. The intelligent decision support module is used to extract multi-dimensional feature vectors for each disease from the integrated disease database; input the multi-dimensional feature vectors into a pre-trained machine learning model to determine the disease risk level and recommend repair scheme parameters; based on the disease spatial coordinates, estimated repair time for a single disease, and risk level, and combined with the available time of the maintenance window and the number of maintenance teams as constraints, the module uses clustering algorithms and path planning algorithms to group maintenance task clusters and generate the optimal operation sequence for a single task cluster, and outputs a visualized complete set of maintenance operation plans.

8. The intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning according to claim 7, characterized in that, In the model and coordinate management module, during the process of constructing the global unified coordinate system in the building information model, the three-dimensional coordinate unified transformation system is embedded into the underlying logic of the BIM software coordinate origin configuration, and the horseshoe-shaped tunnel cross-section segment width calculation function is implanted to realize the binding of tunnel cross-section geometry with global three-dimensional coordinates. The unified three-dimensional coordinate transformation system includes the three-layer transformation mathematical relationship of geodetic latitude and longitude, local independent plane coordinates, and BIM model local coordinates in the planar dimension, and the elevation datum transformation model in the vertical dimension; The three-layer transformation mathematical relationship includes: Gauss-Kruger projection forward calculation model, local independent coordinate system translation scale correction formula, and seven-parameter coordinate transformation matrix; The expression for the Gauss-Kruger projection forward calculation model is as follows: In the formula, L and B are the latitude and longitude of the geodetic coordinates, respectively; a is the semi-major axis of the CGCS2000 reference ellipsoid; N is the radius of curvature of the zonal circle; and l is the difference in longitude between the measuring point and the central meridian. is the longitude of the central meridian of the local independent coordinate system; e is the first eccentricity of the ellipsoid. For projection auxiliary parameters; t is the latitude tangent; X is the meridian arc length corresponding to dimension B; , , , All are fixed ellipsoidal coefficients from the meridian arc length series expansion; x and y are the X and Y coordinates of the natural plane of the Gaussian projection, respectively. The expression for the local independent coordinate system translation scale correction formula is as follows: In the formula, , These are the X and Y coordinates of the local independent coordinate system plane at the engineering site; , denoted as the translation constant of the origin of the local coordinate system; k is the scale ratio of the local independent coordinate system. The expression for the seven-parameter coordinate transformation matrix is: = In the formula, X, Y, and H are the global three-dimensional coordinates of the subway project; X and Y are the local independent plane coordinates; and H is the 85 elevation. , , The three parameters are: 3D translation; m is the scale parameter. It is a three-dimensional infinitesimal rotational orthogonal matrix; These represent minute rotations around the X, Y, and Z axes, respectively; u, v, and w are local right-handed Cartesian 3D coordinates within the BIM software. The expression for the elevation datum transformation model is: In the formula, Store elevations for BIM model components; The actual elevation measured on-site; This is the elevation datum offset constant for BIM software. The function for calculating the segment width of the horseshoe-shaped tunnel section is: Arch section: Straight wall section: Inverted arch section: In the formula, x is the width at any cross-sectional height Z; R is the radius of the arch arc. B is the radius of the inverted arch; B is the standard clear width of the tunnel. , These are the elevations of the center of the arch and the inverted arch, respectively. These are the elevation boundary values ​​for each section segment.

9. The intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning according to claim 8, characterized in that, In the intelligent acquisition and transmission module, during the process of capturing and recording the three-dimensional coordinates of the center point, the contour point cloud data, and the severity index of the cavity defect based on the global unified coordinate system, when the cavity defect is identified, the three-dimensional coordinates of the center point of the defect are recorded; the contour point cloud of the cavity defect on the tunnel surface boundary is obtained by scanning with an infrared device; the maximum delamination depth is inverted based on the radar signal; and the contour point cloud inside the cavity is obtained by laser signal.

10. The intelligent treatment device for cavity defects in subway tunnels based on BIM and machine learning according to claim 9, characterized in that, In the intelligent decision support module, the multi-dimensional feature vector of each disease extracted from the integrated disease database includes: lining defect type code, seepage state characteristics, effective lining thickness, longitudinal extension length of defect, circumferential expansion width of defect, quantitative value of tunnel surrounding rock grade, tunnel burial depth, cross-sectional convergence deformation monitoring value, and linear distance of disease from the nearest maintenance entrance / exit. In the intelligent decision support module, during the process of determining the disease risk level and recommending repair scheme parameters through the pre-trained machine learning model, the pre-trained machine learning model is a multi-class GBDT gradient boosting tree model; an accumulation tree function is constructed for the five disease levels: In the formula, The predicted score for the k-th disease category; M is the total number of base decision trees; Output the value for the m-th weak decision tree under the k-th category; The multi-class GBDT gradient boosting tree model outputs five risk levels corresponding to predicted probability vectors, and the category with the highest probability is taken as the disease determination level. The criteria for determining the five risk levels are as follows: Grade I: Loose filling with no damp stains or standing water in the lining; Grade II: Water seepage and loose filling exist, and the effective lining thickness is greater than 15cm; Grade III: Water seepage and loose filling exist, and the effective lining thickness is no more than 15cm; Grade IV: There are defects such as water seepage and voids, and the effective thickness of the lining is greater than 15cm; Grade V: There are defects such as water seepage and voids, and the effective thickness of the lining is not greater than 15cm; The recommended prediction model for the repair scheme parameters is a multi-output regression model; the expression for the multi-output regression model is: In the formula, x is the disease feature vector; L is the disease operation distance; and p is the recommended repair process. M represents the estimated standard working hours for a single defect; M represents the estimated usage of main materials. In the intelligent decision support module, during the process of grouping maintenance task clusters and generating optimal operation sequences for single task clusters using the clustering algorithm and the path planning algorithm respectively, the clustering algorithm is an improved K-means clustering algorithm; the objective function of the improved K-means clustering algorithm is: In the formula, The spatial coordinates of the disease P; For clusters The center; Estimate working hours for disease prevention; The maximum available man-hours for a single skylight opening; This represents the average working hours within the cluster. Space-time balance weights; The path planning algorithm is a genetic algorithm used to solve the NP-Hard tunnel maintenance problem with risk priority. The objective function of the genetic algorithm is: In the formula, The overall objective function for the maintenance path; This represents the sequence of maintenance tasks for defects; n represents the total number of defects within a single maintenance task cluster. This refers to the distance weighting coefficient. This is the risk penalty weighting coefficient; Points in the tunnel topology network To the next point is The actual travel distance; For point Risk level of cavity defects; This is the risk weight mapping function.

Images