A floor pit excavation method based on a graph neural network and interface fusion

Abstract

The application provides a floor pit excavation method based on a graph neural network and interface fusion; in view of the problems that the corresponding relationship between abnormal segments detected by nondestructive detection and existing anti-floating anchor rods is not clear, the floor structure boundary has a separating effect on the continuous path, and the anti-floating anchor rods are easily cut off by the boundary of the proposed excavation, under the condition of only obtaining the nondestructive detection results, the existing anti-floating anchor rod data, the floor structure boundary and the proposed excavation boundary, the multi-source objects are constructed into a heterogeneous relationship graph network, and the source consistency, path continuation and boundary separation of the candidate connection are jointly judged through the anchor rod through hidden variable layer, the same rod through parameters and the crossing parameters are calculated, the through relationship graph and the non-cutting zone are generated, and the proposed excavation contour is adjusted and the floor cutting excavation is implemented; the method can improve the reliability of the anti-floating anchor rod avoidance judgment and the accuracy of the excavation implementation.

Description

Technical Field

[0001] This invention relates to the field of foundation pit excavation technology for building engineering, and in particular to a foundation pit excavation method based on graph neural network and interface fusion. Background Technology

[0002] During the renovation of existing buildings or underground projects, it is often necessary to add equipment pits, manholes, sump pits, or localized sinking spaces on the basement floor slab. Since anti-buoyancy anchors are usually installed beneath the floor slab to resist groundwater buoyancy and maintain overall structural stability, accidental cutting or damage to these anchors during excavation can disrupt the original load-bearing system, increasing the risk of leakage, cracking, local instability, and subsequent reinforcement and rework. Therefore, before cutting and excavating the floor slab, it is usually necessary to identify the possible distribution of existing anti-buoyancy anchors within the proposed excavation area and its adjacent range, as well as their relationship to the excavation boundary. In actual projects, existing anti-buoyancy anchor data often suffers from missing drawings, misaligned construction, incomplete records, or changes in physical condition, making it difficult to directly determine the actual anchor path based solely on the original data. While non-destructive testing can obtain some response information, the results often appear as localized anomalies, and are affected by variations in burial depth, structural boundaries, material differences, and on-site noise; therefore, the abnormal response may not correspond one-to-one with a single anchor. Therefore, in the specific construction scenario of avoiding anti-buoyancy anchors in equipment foundation pits, the key issue in the construction of foundation plate modification is how to identify continuous paths that may be related to the same anchor, and constrain the excavation outline accordingly, under the realistic conditions of only obtaining non-destructive testing results, existing anchor data, bottom plate structural boundaries and proposed excavation boundaries.

[0003] In existing technologies, avoiding existing anti-buoyancy anchors during foundation excavation typically involves a combination of drawing verification, on-site surveying, and non-destructive testing. A common practice is to first retrieve existing anti-buoyancy anchor as-built data or design data to determine the approximate starting point, direction, length, and spacing of the anchors. Then, using non-destructive testing results such as ground-penetrating radar, electromagnetic scanning, or impact echo, abnormal responses within the proposed excavation area are identified. For identified abnormal areas, engineers generally perform manual comparisons based on geometric characteristics such as planar location, orientation, depth, and length, or use methods such as threshold screening, linear extrapolation, adjacent splicing, and boundary projection to infer the possible extension path of the anchors, and determine whether the proposed excavation boundary intersects with suspected anchors. If there is a risk of intersection, avoidance construction is carried out by locally modifying the excavation boundary, adjusting the pit location, or retaining a portion of the foundation slab.

[0004] While the aforementioned technical solutions can assist in construction judgment to some extent, their determination of whether "detected anomalies belong to the same existing anti-buoyancy anchor continuous path" relies heavily on geometric proximity and human experience, lacking a joint expression mechanism for consistency of origin, path continuity, and the separating effect of the base slab structure boundary. When underground anomaly responses are fragmented and discontinuous, existing methods struggle to reliably distinguish between continuous segments and segments not belonging to the same anchor. Consequently, it becomes difficult to reliably transform "anchor paths that may cross the excavation boundary" into specific restricted sections that can be directly used for construction contour adjustments, making it difficult to form a closed loop between excavation avoidance judgment and subsequent implementation.

[0005] Therefore, a method for excavating foundation pits that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method for excavating foundation pits based on graph neural networks and interface fusion. The core technical problem to be solved by this application is: in the scenario where equipment foundation pit excavation avoids existing anti-buoyancy anchors, how to reliably determine the continuity path related to the same anti-buoyancy anchor and its uncuttable section on the proposed excavation boundary under the condition of only non-destructive testing results, existing anti-buoyancy anchor data, foundation structure boundary, and proposed excavation boundary, so as to support feasible avoidance adjustments to the excavation profile, given that the relationship between abnormal segments detected by non-destructive testing and the actual connection between existing anchors is unclear and the boundary of the foundation plate structure interferes with the determination of continuous paths.

[0007] The foundation pit excavation method based on graph neural network and interface fusion according to an embodiment of the present invention includes:

[0008] S1. Obtain the non-destructive testing results of the area to be excavated, and extract the detection anomaly segments corresponding to the area to be excavated from the non-destructive testing results;

[0009] S2. Obtain existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary, construct the detected anomaly segments, existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary into different types of nodes in the heterogeneous relationship graph network, and establish candidate connections according to spatial continuity to form a candidate connection graph;

[0010] S3. Embed the anchor bolt penetration hidden variable layer into the candidate connection in the candidate connection diagram. Calculate the same bolt penetration parameter based on the continuous pointing relationship between the two end nodes of the candidate connection, the existing anti-buoyancy anchor bolt data and the bottom plate structure boundary. When the same bolt penetration parameter meets the penetration condition, retain the corresponding candidate connection and form a penetration chain segment. When the same bolt penetration parameter does not meet the penetration condition, remove the corresponding candidate connection to obtain the penetration relationship diagram.

[0011] S4. According to the connection relationship diagram, trace the connection path along the connection chain segment connected to the corresponding node of the proposed excavation boundary, calculate the crossing parameters of each connection path relative to the proposed excavation boundary, and when the crossing parameters meet the crossing conditions, project the corresponding connection path as an uncut strip, and divide the proposed excavation boundary into segments by the uncut strip to form the contour constraint result.

[0012] S5. Based on the contour constraint results and the proposed excavation boundary, adjust the proposed excavation boundary under the constraint of the inseverable zone to form an adjusted excavation contour that does not intersect with the inseverable zone.

[0013] S6. Cut the bottom plate according to the adjusted excavation outline to form an equipment foundation pit that avoids the anti-buoyancy anchor.

[0014] Optionally, S1 is as follows:

[0015] Obtain the non-destructive testing results corresponding to the area to be excavated, and determine the target detection range in the non-destructive testing results based on the boundary of the area to be excavated;

[0016] The detection responses that are continuously distributed along the same extension direction within the target detection range are defined as the continuous response region, and the boundary of the continuous response region is located to obtain the start and end positions of the continuous response region.

[0017] The segment's center plane position, burial depth, length, orientation, response width, response intensity, and attenuation difference at both ends are calculated based on the start and end positions of the continuous response region to form the segment characteristics corresponding to the continuous response region.

[0018] Anomaly detection is performed on the segment features. When the response intensity meets the anomaly threshold, the length meets the length threshold, and the continuity of the direction meets the connectivity threshold, the corresponding continuous response region is identified as the detected anomaly segment, and the segment features are used as the node input features for the detected anomaly segment.

[0019] Optionally, S2 is as follows:

[0020] Obtain existing anti-buoyancy anchor data, and extract the anchor starting point plane position, design direction, design length, design burial depth and spacing from the existing anti-buoyancy anchor data to form a six-dimensional node input feature;

[0021] Obtain the boundary of the base slab structure and the boundary to be excavated. Extract the boundary center plane position, boundary orientation, thickness variation and interface category from the boundary of the base slab structure to form five-dimensional node input features. Extract the boundary center plane position, edge direction, edge length and excavation depth from the boundary to be excavated to form five-dimensional node input features.

[0022] The center plane position, burial depth, length, orientation, response width, response intensity, and attenuation difference between the two ends of the detected anomaly segment are determined as the input features of the eight-dimensional node. The detected anomaly segment, existing anti-buoyancy anchor data, bottom plate structure boundary, and proposed excavation boundary are respectively constructed as different types of nodes in the heterogeneous relation graph network.

[0023] A node encoding layer is used to encode the eight-dimensional node input features, six-dimensional node input features, and five-dimensional node input features respectively. The node encoding layer includes a first layer of thirty-two fully connected neurons and a second layer of sixteen fully connected neurons to obtain sixteen-dimensional node features corresponding to different types of nodes.

[0024] Based on the sixteen-dimensional node features, node pairs that meet the distance threshold, direction threshold, and burial depth threshold are selected. The node pairs are then calculated for the node pairs for the distance, direction threshold, burial depth threshold, projection overlap, boundary separation value, anchor bolt direction consistency, and boundary intersection value, forming a seven-dimensional candidate connection feature.

[0025] A candidate connection coding layer is used to encode the seven-dimensional candidate connection features. The candidate connection coding layer consists of a first layer of twenty-four fully connected neurons and a second layer of twelve fully connected neurons, resulting in twelve-dimensional connection features.

[0026] Based on the sixteen-dimensional node features and twelve-dimensional connection features, when the anchor direction consistency meets the consistency threshold, the projection overlap meets the overlap threshold, the boundary separation value meets the separation threshold, and the boundary intersection value meets the intersection threshold, the spatial continuity relationship is determined and candidate connections are established to form a candidate connection map.

[0027] Optionally, S3 specifically refers to:

[0028] Each candidate connection in the candidate connection diagram is used as a processing unit of the anchor bolt penetration hidden variable layer. The sixteen-dimensional node features corresponding to the two ends of the candidate connection, the twelve-dimensional connection features corresponding to the candidate connection, the sixteen-dimensional node features corresponding to the existing anti-buoyancy anchor bolt data, and the sixteen-dimensional node features corresponding to the bottom plate structure boundary are extracted to form a sixty-dimensional penetration judgment input.

[0029] The sixty-dimensional through-determination input anchor is connected to the hidden variable layer. The hidden variable layer includes source branches, continuation branches, separation branches and fusion units. The source branches include twenty-four fully connected neurons and eight fully connected neurons, the continuation branches include thirty-two fully connected neurons and eight fully connected neurons, and the separation branches include sixteen fully connected neurons and eight fully connected neurons.

[0030] The source branch is calculated to determine the source consistency relationship between the candidate connection end and the existing anti-buoyancy anchor data, and the source branch output is generated.

[0031] The continuation branch is calculated, and the continuation relationship of the detected anomaly segment along the continuous pointing relationship is determined based on the direction, burial depth and length of the nodes at both ends of the candidate connection, and the continuation branch output is formed;

[0032] The separation branches are calculated, and the separation relationship of the bottom plate structure boundary to the continuous pointing relationship is determined according to the relative position of the bottom plate structure boundary and the candidate connection, and the separation branch output is formed.

[0033] The source branch output, the continuation branch output, and the split branch output are input into the fusion unit. The fusion calculation is performed using twelve fully connected neurons and one output neuron to obtain the one-dimensional same-rod connection parameter. When the same-rod connection parameter meets the connection condition, the corresponding candidate connection is retained and a connection chain segment is formed. When the same-rod connection parameter does not meet the connection condition, the corresponding candidate connection is eliminated.

[0034] The candidate connection input is retained into the connection relationship update layer. The connection relationship update layer includes two layers of heterogeneous relationship propagation units. The first layer of heterogeneous relationship propagation units performs local connection updates on the connection chain segments, and the second layer of heterogeneous relationship propagation units performs path continuous updates on the connection chain segments to form a connection relationship graph.

[0035] Optionally, S4 specifically refers to:

[0036] Extract the connecting chain segments that are directly connected to the nodes corresponding to the proposed excavation boundary from the connecting relationship diagram, and trace along the connecting chain segments to the termination node connected to the existing anti-buoyancy anchor data according to the connection direction and node type of the candidate connections retained in the connecting relationship diagram, thus forming each connecting path.

[0037] Each connection path is input into the connection relationship update layer. The first layer of heterogeneous relationship propagation unit is used to update the local connectivity relationship between the detected abnormal segments, existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary to form a local connection representation. Then, the second layer of heterogeneous relationship propagation unit is used to continuously update the local connection representation to obtain the sixteen-dimensional connection path features corresponding to each connection path.

[0038] Extract the 16-dimensional node features corresponding to the nodes of the proposed excavation boundary for each through path, and concatenate the 16-dimensional through path features with the 16-dimensional node features to form a 32-dimensional input;

[0039] The 32-dimensional input is fed into the traversal decision layer, which consists of 24 fully connected neurons and 3 output neurons. The 32-dimensional input is calculated to obtain the traversal value, the start point of the uncut zone, and the end point of the uncut zone. The three outputs together constitute the traversal parameters.

[0040] The crossing parameters are determined. If the crossing value meets the crossing threshold and the starting point and ending point of the incisional zone fall on the corresponding edge segment of the proposed excavation boundary, the corresponding through path is determined to cross the proposed excavation boundary, and the boundary segment between the starting point and the ending point of the incisional zone is determined as the effective projection segment.

[0041] The effective projection section is projected as an uncut zone. When the crossing value does not meet the crossing threshold, the corresponding through path is excluded from the uncut zone generation process. When multiple uncut zones fall into the same proposed excavation boundary section, the multiple uncut zones are merged.

[0042] The proposed excavation boundary is segmented according to the starting point and ending point of the inseverable zone, and the segmented proposed excavation boundary is correlated with the crossing parameters of the corresponding through path to form the contour constraint result.

[0043] Optional, S5 specifically includes:

[0044] Based on the contour constraint results, extract the corresponding boundary segments of the incisional zone start point, incisional zone end point and proposed excavation boundary, and determine the restricted and adjustable boundary segments in the proposed excavation boundary.

[0045] Using the restricted edge segment as the boundary constraint, the position of the adjustable edge segment is adjusted so that the adjustable edge segment is located outside the indivisible strip and remains connected to the restricted edge segment from beginning to end, thus forming a candidate adjustable edge segment.

[0046] The candidate adjustment edge segment is spliced ​​with the proposed excavation boundary that is not covered by the uncut zone, and the adjacent edge segments after splicing are connected. When the connection angle meets the angle threshold and the connection distance meets the distance threshold, a closed boundary is formed.

[0047] Intersection determination is performed between the closed boundary and the uncuttable zone. If there is no intersecting section, the closed boundary is determined as the adjusted excavation profile. If there is an intersecting section, the adjustable edge segment corresponding to the intersecting section is further adjusted according to the profile constraint result to form an adjusted excavation profile that does not intersect with the uncuttable zone.

[0048] Optionally, step S6 specifically includes:

[0049] The closed cutting path is determined by adjusting the excavation profile, and the closed cutting path is decomposed into profile cutting segments that are connected end to end.

[0050] Align the contour cutting segments sequentially, determine the cutting start point, cutting end point and cutting direction of each contour cutting segment, and form a continuous cutting trajectory;

[0051] The continuous cutting trajectory is closed. When the distance between the cutting start point and the cutting end point meets the closure threshold, the bottom plate is contour cut according to the continuous cutting trajectory, so that the bottom plate area enclosed by the continuous cutting trajectory is separated from the surrounding bottom plate.

[0052] The separated bottom plate area is excavated to form an equipment foundation pit corresponding to the adjusted excavation outline.

[0053] The beneficial effects of this invention are:

[0054] (1) This proposal proposes an improved method for determining the continuity of anti-buoyancy anchors based on heterogeneous relational graph networks. In the specific scenario where equipment pits avoid existing anti-buoyancy anchors, the method no longer directly extrapolates non-destructive detection anomalies to anchor paths based solely on planar proximity, similar direction, or manual splicing rules. Instead, it integrates detection anomaly segments, existing anti-buoyancy anchor data, bottom plate structural boundaries, and proposed excavation boundaries into a heterogeneous relational graph network, and introduces an anchor continuity latent variable layer at the candidate connection level. This latent variable layer establishes a joint determination mechanism for candidate connections based on three interpretable mechanisms: consistent source, path continuity, and boundary separation. After forming the continuity parameters of the same anchor, it then decides whether to retain the connection. Compared to existing solutions that mainly rely on geometric thresholds and empirical judgments, this proposal transforms the key intermediate quantity of "whether abnormal segments belong to the continuous path of the same anchor bolt" from a static geometric matching problem into a problem of determining the continuity relationship under multi-source constraints. Under the conditions of fragmented detection response, local discontinuity, and structural interface interference, it can more stably distinguish between continuous chain segments and segments not from the same bolt, thereby improving the reliability of the identification of the true continuous path of existing anti-buoyancy anchor bolts.

[0055] (2) This proposal puts forward a novel method for determining the crossing path and generating the uncuttable zone. Under the constraint of only obtaining the connection relationship diagram, the proposed excavation boundary, and the characteristics of the boundary nodes, instead of using the existing technology to directly project, extend, or simply intersect suspected anchor lines, the method first traces the connection path from the existing anti-buoyancy anchor data to the proposed excavation boundary along the connection relationship diagram. Then, combining the continuous update results of the path and the characteristics of the proposed excavation boundary nodes, the crossing value and the start and end positions of the uncuttable zone are calculated, so that "which boundary segment cannot be cut" becomes a boundary constraint result that can be directly determined. Since this method incorporates the end burial depth relationship, the connection direction relationship, and the effective projection of the boundary segment into the crossing parameters, it can further transform the path identification results, which were originally at the level of risk warning, into a description of the restricted section that can be applied to the excavation boundary. This reduces the semantic gap between the identification results and the construction outline, and enables the avoidance determination to directly serve the subsequent outline adjustment.

[0056] (3) This proposal puts forward a method for excavating foundation pits with anti-buoyancy anchors to facilitate construction. It links anomaly detection extraction, candidate connection construction, same-anchor connection screening, crossing judgment, uncuttable zone constraint, excavation contour adjustment, and foundation slab cutting path generation into a single technical chain. Unlike existing technologies that often require manual secondary trimming and repeated verification after identifying suspected anchors, this proposal uses contour constraint results to divide the proposed excavation boundary into restricted and adjustable segments, and generates an adjustable excavation contour without intersecting with the uncuttable zone, ultimately outputting an executable continuous cutting trajectory. Thus, this proposal not only focuses on identification at the detection level, but also further implements the identification results into deliverable construction boundaries and cutting paths. Under the condition that existing data is incomplete and only non-destructive detection results and structural boundary information can be obtained on site, it helps to reduce the risk of accidentally cutting existing anti-buoyancy anchors during foundation slab excavation and improves the consistency between the excavation avoidance plan and the on-site implementation. Attached Figure Description

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention; Figure 2 This invention presents a flowchart illustrating the process of obtaining non-destructive testing results and extracting abnormal fragments from the proposed excavation area using a method for excavating foundation pits based on graph neural networks and interface fusion. Figure 3 This is a flowchart illustrating the construction of a candidate connection graph for a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention. Figure 4 This is a flowchart of the anchor bolt penetration hidden variable layer and penetration relationship diagram formation process for a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention. Figure 5 This is a flowchart illustrating the process of determining the through-path crossing and the contour constraint results in a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention. Figure 6 This is a flowchart illustrating the process of adjusting the excavation profile in a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention. Figure 7 This is a flowchart illustrating the process of cutting the foundation slab and forming the equipment foundation pit in a foundation pit excavation method based on graph neural network and interface fusion proposed in this invention. Detailed Implementation

[0058] In Example 1, reference Figures 1 to 7A method for excavating foundation pits based on graph neural networks and interface fusion, comprising:

[0059] S1. Obtain the non-destructive testing results of the area to be excavated, and extract the detection anomaly segments corresponding to the area to be excavated from the non-destructive testing results.

[0060] S2. Obtain existing anti-buoyancy anchor data, bottom slab structural boundaries, and planned excavation boundaries. Construct the detected anomaly segments, existing anti-buoyancy anchor data, bottom slab structural boundaries, and planned excavation boundaries into different types of nodes in a heterogeneous relational graph network, and establish candidate connections according to spatial continuity to form a candidate connection graph.

[0061] S3. Embed the anchor bolt penetration hidden variable layer into the candidate connection diagram. Calculate the same-bolt penetration parameters based on the continuous pointing relationship between the two ends of the candidate connection, existing anti-buoyancy anchor bolt data, and the boundary of the bottom plate structure. When the same-bolt penetration parameters meet the penetration conditions, retain the corresponding candidate connection and form a penetration chain segment. When the same-bolt penetration parameters do not meet the penetration conditions, remove the corresponding candidate connection to obtain the penetration relationship diagram.

[0062] S4. Based on the connection diagram, trace the connection path along the connection chain segments connected to the corresponding nodes of the proposed excavation boundary. Calculate the crossing parameters of each connection path relative to the proposed excavation boundary. When the crossing parameters meet the crossing conditions, project the corresponding connection path as an indispensable zone. Then, segment the proposed excavation boundary using the indispensable zone to form the contour constraint result.

[0063] S5. Based on the contour constraint results and the proposed excavation boundary, adjust the proposed excavation boundary under the constraint of the incisional zone to form an adjusted excavation contour that does not intersect with the incisional zone.

[0064] S6. Cut the bottom plate according to the adjusted excavation outline to form an equipment foundation pit that avoids the anti-buoyancy anchor bolts.

[0065] In this embodiment, step S1 specifically includes:

[0066] The non-destructive testing results corresponding to the proposed excavation area are uniformly mapped into the base construction plane coordinate system and depth coordinate system. The non-destructive testing results include the sampling plane coordinates, sampling depth, and response value. The proposed excavation boundary is represented as a closed boundary, and extension distances are set along the inner and outer sides of the proposed excavation boundary, respectively. The target detection range is determined by the area enclosed by the closed boundary, the adjacent area inside the closed boundary, and the extended area outside the closed boundary, with the extended distance. The larger value between the design spacing and the design burial depth of the existing anti-buoyancy anchor data is used to cover the continuous response path that may cross the proposed excavation boundary. After the target detection range is determined, only the sampling plane coordinates, sampling burial depth and response value within the target detection range are retained as the actual processing data.

[0067] All sampling locations within the target detection range are discretized into response sampling units. If the response value of a response sampling unit is greater than the response values ​​of eight adjacent sampling units at the same burial depth, and also greater than the average response value of the corresponding projected neighborhood of the upper and lower adjacent burial depths, then the corresponding response sampling unit is determined as a local peak sampling unit. Using the local peak sampling unit as the seed unit, an adjacency set is established within the planar and burial depth neighborhoods of the seed unit. The local principal extension direction is calculated based on the response value distribution within the adjacency set. The local principal extension direction is taken as the principal axis direction of the weighted planar discrete point set of response values. When searching for adjacent sampling units along the local principal extension direction, the direction difference, unit spacing, and [other factors] are simultaneously checked. The burial depth difference is determined by three conditions: directional difference (the angle between the line connecting adjacent sampling units and the local main extension direction), unit spacing (the planar distance between the center points of adjacent sampling units), and burial depth difference (the absolute value of the difference in burial depth between adjacent sampling units). When these three conditions are not greater than the preset directional constraint value, preset continuous spacing value, and preset burial depth constraint value, respectively, adjacent sampling units are merged into the same continuous response region. When any condition is not met, the extension of the current continuous response region ends. When locating the boundary of each continuous response region, all sampling units within the continuous response region are projected onto the main axis direction, and the end with the smallest projected position is determined as the starting position. The end with the largest projected position is determined as the termination position. ;

[0068] Based on the starting position and termination position Extract the segment features corresponding to the continuous response region, and denote the position of the segment center plane as . , Includes planar coordinate components and , and Use starting position and termination position The midpoint of the plane projection is determined, and the burial depth is denoted as . , The burial depth of each sampling unit within the continuous response region is determined by the weighted average of the response values, and the length is denoted as . , Use starting position and the end position The projected distance along the principal axis is determined, and its direction is denoted as . , Use starting position Point to the end position The planar orientation angle is determined, and the response width is denoted as... , Adoption and Direction The response intensity is defined as the average width of a continuous high-response band on multiple vertical cutoff lines, where the response value is not less than 60% of the peak value of the continuous response region. , The attenuation difference between the two ends is denoted as the peak response value of the continuous response region minus the average response value of the background window at the same burial depth. , Use starting position The difference between the peak response value within the end window and the average response value at the end edge, and the termination position. The absolute difference between the peak response value within the end window and the average response value at the end edge is determined by... , , , , , , and This constitutes an eight-dimensional segment feature;

[0069] Inputting eight-dimensional segment features into the anomaly detection process, response intensity With abnormal threshold Comparison, length With length threshold The continuous response region is compared and divided into continuous segments of equal length along the principal axis. The local orientation of each continuous segment is calculated, and then the local orientation of each continuous segment is compared with the global orientation. The deviation value is calculated, and the discontinuous distance between adjacent continuous segments that are not filled by sampling units is also calculated. All deviation values ​​and all discontinuous distances are then converted into a continuity index. When the response intensity Not lower than the abnormal threshold ,length Not less than the length threshold Towards continuous indicators Not lower than the connectivity threshold At that time, the corresponding continuous response region is determined as the detection anomaly segment. Detecting abnormal fragments Defined as a fragment-level abnormal response object obtained by anomaly detection from a continuous response region within the target detection range; detect abnormal fragments. Used to characterize local detection clues that may be related to the continuous path of the anti-buoyancy anchor, detecting anomalous segments. The abnormal segments will not be directly used as the determination result of the anti-buoyancy anchor body. corresponding , , , , , , and The node input features are written into the node encoding layer.

[0070] In this embodiment, step S2 specifically includes:

[0071] The existing anti-buoyancy anchor data is uniformly mapped into the coordinate system of the base plate construction plane, and nodes of the existing anti-buoyancy anchor data are established for each individual anchor. ,in This refers to the anchor bolt node number, which contains existing anti-buoyancy anchor bolt data nodes. The six-dimensional node input features are derived from the plane coordinates of the anchor bolt starting point. and Design direction Design length Design burial depth and spacing The structure is constructed by dividing the boundary of the base plate into continuous interface segments, and establishing boundary nodes for the base plate structure based on a single continuous interface segment. ,in These are the boundary node numbers, representing the boundary nodes of the base plate structure. The five-dimensional node input features are derived from the coordinates of the boundary center plane. and Boundary Direction Thickness variation and interface categories The proposed excavation boundary is divided into segments, and nodes for each segment are established to represent the proposed excavation boundary. ,in The boundary segment number represents the boundary node to be excavated. The five-dimensional node input features are derived from the coordinates of the boundary center plane. and , Side segment direction Length of side segment and excavation depth Composition, will detect abnormal fragments Access to a heterogeneous relational graph network, where To detect the sequence number of the abnormal segment, detect the abnormal segment. The eight-dimensional node input features are composed of the fragment center plane position components. and burial depth ,length , towards Response width Response strength and the attenuation difference at both ends Composition, will detect abnormal fragments Existing anti-buoyancy anchor data nodes Boundary nodes of the base plate structure and the proposed excavation boundary node Different types of nodes are constructed in a heterogeneous relational graph network, where abnormal fragments are detected. Characterizing the local detection response, there are existing anti-buoyancy anchor data nodes. Characterizing the source of anchor bolts, boundary nodes of the base plate structure. Characterizing the structural separation location, proposed excavation boundary nodes Characterizes the cutting location;

[0072] A node coding layer is used to encode four different types of nodes. The node coding layer includes four independent coding branches, each containing 32 fully connected neurons in the first layer and 16 fully connected neurons in the second layer, which will detect abnormal segments. The eight-dimensional node input features are input to the response node encoding branch, resulting in sixteen-dimensional node features. The existing anti-buoyancy anchor data nodes The six-dimensional node input features are derived from the source node encoding branches to obtain sixteen-dimensional node features. The boundary nodes of the base plate structure The five-dimensional node input features are used to separate the node encoding branches, resulting in sixteen-dimensional node features. The proposed excavation boundary node The five-dimensional node input features are used to encode the boundary node branches, resulting in sixteen-dimensional node features. ,in, , , and These represent the sixteen encoded components obtained by mapping the corresponding input features through two fully connected layers;

[0073] Node pairs to be inspected are generated in a directional manner, and only three types of node pairs are retained: existing anti-buoyancy anchor data nodes. Pointing to detect abnormal fragments Node pairs, detecting abnormal fragments Pointing to another detected anomaly segment Node pairs, detecting abnormal fragments Pointing to the proposed excavation boundary node Node pairs, where This is the sequence number of another detected anomaly segment, a boundary node of the base plate structure. Not considered as candidate connection endpoints, but only involved in boundary separation value calculation; for each node pair, the distance between nodes is calculated. , direction difference and burial depth difference Distance between nodes The direction difference is determined by the Euclidean distance between the center plane coordinates of the starting node and the center plane coordinates of the ending node. The angle between the direction of the connection between nodes and the extension direction of the starting node is used to determine the location, which includes data on anti-buoyancy anchors. The extension direction is taken from the design direction. Detecting abnormal fragments The direction of extension is taken as the direction Difference in burial depth The depth is determined by the absolute value of the difference between the starting node depth and the ending node depth, which includes data on anti-buoyancy anchor bolt nodes. Take the design burial depth Detecting abnormal fragments and taking burial depth Proposed excavation of boundary nodes Determine the excavation depth Only the distance between nodes is retained. Not greater than the distance threshold , direction difference Not greater than the direction threshold Difference in burial depth Not greater than the burial depth threshold Node pairs;

[0074] Calculate the projection overlap for the retained node pairs. Boundary separation value Anchor bolt direction consistency Intersection value with boundary Projection overlap The boundary separation value is determined by the ratio of the overlapping length of the projection intervals of the start and end nodes along the extension direction of the start node to the shorter projection length. Connecting nodes to the boundary nodes of the base plate structure The number of intersections, thickness variation, interface type, and intersection angle of the corresponding boundary segments are calculated together:

[0075] ;

[0076] In the formula, For boundary separation values, This represents the number of intersections between the node pair connection and the boundary of the base plate structure. The intersection boundary segment number, , , These are the weighting coefficients for the thickness variation term, interface category term, and intersection angle term, and , For the first The thickness variation of each intersecting boundary segment, This represents the upper limit of the normalized thickness variation. For the first The interface category encoding value of each intersecting boundary segment. Encode the maximum value for the interface category. For the node pair, the connection direction is the same as the first The included angle between the directions of the intersecting boundary segments, and the consistency of the anchor bolt direction. Using node-to-line direction and design direction The angle between them is calculated for detecting abnormal segments. Pointing to another detected anomaly segment Node pairs and detection of anomaly fragments Pointing to the proposed excavation boundary node Node pairs, design direction Take the distance from the starting point to detect abnormal segments The center plane position does not exceed the distance threshold Recent existing anti-buoyancy anchor data nodes The design direction is such that, when there is no existing anti-buoyancy anchor data node that meets the conditions, At that time, the consistency of the anchor bolt direction Take zero, boundary intersection value The node-to-node connection is extended along the direction of the starting node and then connected to the boundary node of the planned excavation. The intersection of corresponding edge segments is determined when the intersection point falls within the length of the edge segment. When the corresponding edge segment range is within the boundary intersection value Select one; otherwise, select the boundary intersection value. Take zero, based on the distance between nodes. , direction difference Difference in burial depth Projection overlap Boundary separation value Anchor bolt direction consistency Intersection value with boundary Together they constitute the seven-dimensional candidate connectivity features;

[0077] A candidate connection coding layer is used to encode the seven-dimensional candidate connection features. The candidate connection coding layer consists of a first layer of twenty-four fully connected neurons and a second layer of twelve fully connected neurons, resulting in twelve-dimensional connection features. ,in The system consists of twelve connection coding components formed by the seven-dimensional candidate connection feature mapping. The sixteen-dimensional node features and the twelve-dimensional connection features are used together to determine the spatial continuity relationship. The spatial continuity relationship is defined as: a node pair simultaneously satisfies the distance between nodes. , direction difference Difference in burial depth The screening criteria must be met, and the anchor bolt direction consistency must be satisfied. Not lower than the consistency threshold Projection overlap Not lower than the overlap threshold Boundary separation value Not higher than the separation threshold Boundary intersection value Not lower than the intersection threshold The node pairs that are in a spatial continuity relationship state at the time will be established as candidate connections. Candidate connections Candidate connections are defined as unverified connection edges generated by oriented node pairs in a heterogeneous relational graph network and specifically used for calculating the same-rod continuity parameters. This does not mean that they have been confirmed to belong to the same anti-buoyancy anchor bolt; all different types of nodes are connected to all candidate connections. Together they form a candidate connectivity graph ,in This represents a candidate connection graph.

[0078] In this embodiment, step S3 specifically includes:

[0079] Candidate connection graph Each candidate connection in The anchor bolt penetration hidden variable layer is directly embedded as an independent processing unit. The embedding method is: candidate connections are not included. Instead of pre-treating the edges as fixed graph edges, candidate connections are first considered. Perform a joint continuity test to determine candidate connections. Whether a candidate connection participates in the propagation of the connection as a valid connection. Maintain the directional relationship, which is limited to the existing anti-buoyancy anchor data nodes. Pointing to detect abnormal fragments Detecting abnormal fragments Pointing to another detected anomaly segment Detecting abnormal fragments Pointing to the proposed excavation boundary node ,in The node number is the existing anti-buoyancy anchor data. and To detect the sequence number of abnormal segments, For each candidate connection, the node number of the proposed excavation boundary node is used. Extract the 16-dimensional node features of the starting node and the ending node, and perform an arithmetic mean according to the corresponding dimensions to form the 16-dimensional node features corresponding to the two ends of the candidate connection. Extract candidate connections Corresponding twelve-dimensional connectivity features The twelve-dimensional connection feature Distance between nodes , direction difference Difference in burial depth Projection overlap Boundary separation value Anchor bolt direction consistency Intersection value with boundary The twelve link coding components obtained by the candidate link coding layer mapping, if the candidate link The starting node is the existing anti-buoyancy anchor data node. Then directly extract the existing anti-buoyancy anchor data nodes. The sixteen-dimensional node features are used as source reference features. If candidate connection The starting node is not a node in the existing anti-buoyancy anchor data. Then according to the distance threshold Direction threshold and burial depth threshold Search for existing anti-buoyancy anchor data nodes that meet the conditions and have the smallest center-plane distance. And extract the corresponding 16-dimensional node features as source reference features. If there are no existing anti-buoyancy anchor data nodes that meet the conditions Then directly remove the candidate connection. In the candidate connection boundary separation value Calculate all boundary nodes of the base plate structure In the middle, select the boundary separation value. The boundary node of the base plate structure that contributes the most And extract the boundary nodes of the base plate structure. The sixteen-dimensional node features are used as the separating reference features. ,in The boundary node number of the base plate structure is... , , and Sequentially assembled to form a 60-dimensional interconnected judgment input. ;

[0080] Input the sixty-dimensional interconnection judgment. The input anchor penetrates the latent variable layer, which includes source branches, continuation branches, split branches, and fusion units. The source branches consist of twenty-four fully connected neurons and eight fully connected neurons. The input is... , and The source branch is based on the candidate connection starting point to the existing anti-buoyancy anchor data node. Center plane distance, candidate connection direction and design direction The included angle, the depth of the candidate connection starting point and the design burial depth The difference is mapped to form an eight-dimensional source branch output. ,Will The arithmetic mean of the eight components is used to form the source branch judgment value. The source consistency relationship is defined as the ability of the candidate connection starting point to trace back to the existing anti-buoyancy anchor data node. Furthermore, the direction of tracing back aligns with the design direction. Consistency, backtracking depth and design burial depth A consistent relational state;

[0081] The continuation branch consists of thirty-two fully connected neurons and eight fully connected neurons, with the input being... and The branch continues to jointly determine the orientation, burial depth, and length of the nodes at both ends of the candidate connection: taking the extension direction of the starting node as a reference, it calculates whether the ending node is located within the coverage area of ​​the extension direction of the starting node, and calculates the directional difference between the nodes at both ends of the candidate connection. and burial depth difference Calculate the projection overlap along the candidate connection direction. Then, the continuity of length is determined by whether the projection intervals of the starting node length and the ending node length in the candidate connection direction are continuous. The source node is the existing anti-buoyancy anchor data node. When the length is continuous, the design length is used. The coverage relationship between the endpoint node and the projection starting point is determined, and the source node is the detected anomaly segment. When the length is continuous, the starting length is used. With end length The connection relationship of the projection intervals is determined, and the continuation branches form an eight-dimensional continuation branch output. and will The arithmetic mean of the eight components forms the continuation branch judgment value. The continuation relationship is defined as the relationship state in which the two ends of the candidate connection simultaneously satisfy the direction connection, burial depth connection and length coverage connection under the continuous pointing relationship.

[0082] The dividing branch consists of sixteen fully connected neurons and eight fully connected neurons, with the input being... and The branches are separated according to the boundary nodes of the base plate structure. The coded components corresponding to the boundary center plane position, boundary orientation, thickness variation, and interface category, as well as candidate connections. Corresponding boundary separation value Determine the boundary nodes of the base plate structure Whether to sever continuous pointing relationships, and to create an eight-dimensional split branch output. and will The arithmetic mean of the eight components forms the dividing branch determination value, and the dividing relationship is defined as the boundary node of the base plate structure. For candidate connections The state of blocking continuous pointing relationships;

[0083] The source branch decision value, continuation branch decision value, and separation branch decision value are threshold-filtered. When the source branch decision value is not lower than the source branch threshold, the continuation branch decision value is not lower than the continuation branch threshold, and the separation branch decision value is not higher than the separation branch threshold, the eight-dimensional source branch is output. Eight-dimensional continuous branch output and eight-dimensional branch output The data are sequentially concatenated into a 24-dimensional fusion input and input into a fusion unit. The fusion unit consists of twelve fully connected neurons and one output neuron, which connects candidate connection start nodes to corresponding existing anti-buoyancy anchor data nodes. The distance between the centers of the plane is denoted as The depth of the candidate connection starting node is compared with the design burial depth. The absolute value of the difference is denoted as When candidate connections The starting node is the existing anti-buoyancy anchor data node. At that time, take ,Pick The parameters of the same rod through the fusion unit are calculated according to the following formula. :

[0084] ;

[0085] In the formula, Indicates candidate connection The same rod through parameters, , , , , and This indicates a preset positive weighting coefficient. This indicates the candidate connection starting node to the corresponding existing anti-buoyancy anchor data node. The distance between the centers of the plane, Indicates the distance threshold. Indicates the consistency of anchor bolt direction. Indicates the depth of the candidate connection starting node relative to the design burial depth. The absolute value of the difference Indicates the burial depth threshold. Indicates the projection overlap. Indicates the difference in direction. Indicates the direction threshold. Indicates the difference in burial depth. This represents the boundary separation value. Each ratio term is obtained by dividing the quantities of the same dimension to obtain the dimensionless value. , and It is a dimensionless value;

[0086] The connection condition is defined as: candidate connection Source reference features have been obtained The source branch determination value is not lower than the source branch threshold, the continuation branch determination value is not lower than the continuation branch threshold, the split branch determination value is not higher than the split branch threshold, and the same rod connection parameter... Not lower than the penetration threshold When the same rod has the same through parameters When the connection condition is met, the corresponding candidate connection is retained. and retain the candidate connections Together with the starting node, ending node, connection direction, and through-rod parameters Together they are referred to as a continuous chain segment. When the same rod has the same through parameters If the connection condition is not met, the corresponding candidate connection is removed. Connecting links Defined as the directed valid connection retained after filtering through the hidden variable layer of anchor bolt penetration;

[0087] All connected links The input is a continuous relation update layer, which consists of two heterogeneous relation propagation units. The first heterogeneous relation propagation unit is for features that satisfy the same source reference feature. The starting point and the ending point are connected and the difference in their connection directions is no greater than the direction threshold. Through-chain segment Perform a localized through-through update and use the through-through parameters of the same rod. As a connection retention weight to correct the local connection direction, the second-layer heterogeneous relation propagation unit updates the locally connected chain segment. Continue updating the path continuously, ensuring that the distance between the endpoint center planes is no greater than the distance threshold. The difference in burial depth at the endpoints should not exceed the burial depth threshold. The difference in connection direction is no greater than the direction threshold. Multi-segment through-chain Connect them to form a continuous relationship diagram. Connection diagram All connections retained in the text are through connections on the same pole that meet the through connection conditions.

[0088] In this embodiment, step S4 specifically includes:

[0089] From the connection diagram Extract the endpoint node as the proposed excavation boundary node. Through-chain segment and will connect such links As a boundary connectivity entry point. A diagram showing the interconnected relationships. Indicates the proposed excavation boundary node. Indicates a continuous chain segment. Indicates the sequence number of the continuous chain segment, along each continuous chain segment Reverse tracing of the connection direction of the preceding through-chain segment Front-drive through-chain segment The determination criterion is: the preceding continuous chain segment The endpoint node and the current connected chain segment The starting node is the same node, and there are two connected chain segments. The connection direction difference is not greater than the direction threshold. Two connecting chain segments The difference in burial depth between the endpoints is not greater than the burial depth threshold. , Indicates the direction threshold. This indicates the burial depth threshold, when multiple preceding continuous chain segments exist. At that time, each leading branch is retained, and when the tracking branch reaches the existing anti-buoyancy anchor data node... At that time, the existing anti-buoyancy anchor data nodes will be used. To the proposed excavation boundary node An ordered sequence of nodes and an ordered sequence of connected chain segments are denoted as a connected path. , This indicates the existing anti-buoyancy anchor data nodes. Indicates the through path, This indicates the path sequence number, indicating when the tracing branch cannot reach the existing anti-buoyancy anchor data node. When this happens, delete the tracking branch. To avoid loop tracking, within the same tracking branch, perform a cleanup of the entered through-chain segment. Mark the visited connected links. Do not re-enter;

[0090] Each connecting path The input is the connected relationship update layer. The first layer of heterogeneous relationship propagation units in the connected relationship update layer only handles the connected paths. Anomaly detection fragments within Existing anti-buoyancy anchor data nodes Boundary nodes of the base plate structure Proposed Excavation Boundary Nodes and connecting segments Perform a local connectivity update. Indicates the detection of abnormal segments, Indicates the boundary nodes of the base plate structure. Indicates the sequence number of the detected anomaly segment. This indicates the boundary node number of the base plate structure. During local connectivity updates, each continuous link segment will be... Through-rod parameters As connection retention weights, the features of adjacent starting nodes, ending nodes, connection directions, and endpoint burial depths are weighted and propagated to form a locally connected representation. , Indicates the parameters for continuity within the same rod. Indicates the through path The partial connectivity indicates that the second-layer heterogeneous relation propagation unit of the connectivity relation update layer follows the connectivity path. From existing anti-buoyancy anchor data nodes Pointing to the proposed excavation boundary node The path is continuously updated sequentially, and the locally connected representation is represented. The source connectivity, directional connectivity, burial depth connectivity, and boundary proximity connectivity are mapped to four coded groups to obtain a sixteen-dimensional through-path feature. , This represents the sixteen-dimensional through-path characteristics. The first four dimensions are continuous source coding, the middle four dimensions are continuous direction coding, the next four dimensions are continuous burial depth coding, and the last four dimensions are boundary proximity coding. The continuous source coding is derived from the through path. Each connecting chain segment Through-rod parameters Together with the connection order of the source node, the directional continuous encoding is formed by adjacent connected chain segments. The connection direction difference is formed by the accumulation of segments, and the continuous encoding of burial depth is formed by adjacent through-chain segments. The difference in burial depth at the endpoints is formed by the accumulation of segments, and the boundary proximity coding is formed by the through path. End node to planned excavation boundary node The planar positional relationship of the corresponding side segments, the directional relationship of the side segments, and the relationship of the excavation depth are all combined to form the overall structure.

[0091] Extraction and connection path Corresponding planned excavation boundary node Sixteen-dimensional node features , Indicates the proposed excavation boundary node Sixteen-dimensional node features, sixteen-dimensional node features From the coordinates of the boundary center plane and , Side segment direction Length of side segment and excavation depth It is formed through node coding layer mapping, where the first four dimensions represent the boundary center plane coordinate coding, the middle four dimensions represent the edge segment direction coding, the next four dimensions represent the edge segment length coding, and the last four dimensions represent the excavation depth coding. and Represents the coordinates of the boundary center plane. Indicates the direction of the edge segment. Indicates the length of the side segment. Indicates the excavation depth, and displays the sixteen-dimensional through-path features. With 16-dimensional node features The components are assembled in a fixed order to form a 32-dimensional input. , This represents the 32-dimensional input that traverses the decision layer, with the first 16 dimensions representing the 16-dimensional through-path features. The latter sixteen dimensions are sixteen-dimensional node features. ;

[0092] Input the 32D model The input is used to determine the traversal decision layer, which consists of 24 fully connected neurons and 3 output neurons. The second output neuron regresses to the proposed excavation boundary node. The starting point of the inscissor zone is obtained by corresponding one-dimensional arc length coordinates on the edge segment. The third output neuron regresses to the proposed excavation boundary node. The one-dimensional arc length coordinates on the corresponding edge segment are used to obtain the endpoint of the inseverable zone. , Indicates the starting point of the inseverable band. Indicates the end point and the beginning point of the indivisible zone. and the end of the indivisible zone All based on the proposed excavation boundary nodes The starting point of the corresponding side segment is the zero point of the arc length, and the length of the side segment is... The upper limit of the arc length will be determined by the boundary node of the proposed excavation. Directly connected through-chain segments The connection direction is denoted as It will be connected with the proposed excavation boundary node. Directly connected detection anomaly fragments The depth of burial is recorded as , Indicates the direction of connection of the end through chain segment. This indicates the depth of the abnormal segment detected at the end; the first output neuron determines the depth based on the penetration path. End geometry and boundary projection validity form crossing value :

[0093] ;

[0094] In the formula, Indicates the through path The value of crossing Indicates the boundary node with respect to the proposed excavation. Directly connected detection anomaly fragments The depth of burial, Indicates the proposed excavation boundary node The excavation depth corresponding to the edge segment, Indicates the burial depth threshold. Indicates the boundary node with respect to the proposed excavation. Directly connected through-chain segments The connection direction, Indicates the proposed excavation boundary node The direction of the corresponding edge segment. Indicates the starting point of the inseverable band. Indicates the end of the indivisible band. Indicates the proposed excavation boundary node The length of the corresponding edge segment This represents a preset positive stable term, where the first term is used to calculate the through path. The degree of consistency between the end burial depth and the excavation depth; the second item is used to calculate the breakthrough path. End direction and planned excavation boundary node The effectiveness of the crossing angle in the corresponding side segment direction; the third term is used to calculate the extent to which the incisive zone falls into the proposed excavation boundary node. The effective projection ratio within the corresponding edge segment range, all three terms are dimensionless values, and the crossing value As a dimensionless value, the crossing value Indestructible starting point and the end of the indivisible zone They are collectively denoted as crossing parameters. , Indicates the crossing parameters;

[0095] The crossing condition is defined as: crossing value Not lower than the crossing threshold The starting point cannot be cut off. Falling into the range The end point cannot be cut. Falling into the range And the starting point cannot be cut off. Less than the end of the indivisible zone , This represents the crossing threshold, when the crossing parameter... When the crossing conditions are met, determine the through path. Crossing the proposed excavation boundary node Corresponding edge segments, and the arc length coordinate range The section is determined to be a valid projection segment when the crossing parameters are... If the crossing conditions are not met, the path will be opened. Excluding the process of generating the indivisible zone, the effective projection segment is projected as the indivisible zone. At that time, along the planned excavation boundary node The position of the arc length parameter corresponding to the edge segment, with an arc length of... The boundary points are determined as indivisible zones. The starting point, with an arc length of The boundary points are determined as indivisible zones. The endpoint is determined, and the boundary section between the two points is defined as an indivisible zone. , Indicates a non-cuttable strip;

[0096] For the same proposed excavation boundary node Multiple uncut strips on the corresponding edge segment To merge sections, start from the point where the section cannot be cut. Sort by size from smallest to largest, then compare adjacent indivisible bands in turn. The starting and ending arc length coordinates, when the latter is an indivisible zone The starting point is not greater than the previous indivisible band. At the end point, the two uncut strips Merge into a new indivisible zone, with the new starting point taking the smaller starting point and the new ending point taking the larger ending point, when the latter indivisible zone... The starting point is greater than the previous indivisible band. At the endpoint, it is retained as an independent, uncuttable strip, and the arc length of the starting point of the edge segment is determined. Arc length of the endpoint of the side segment The starting and ending points of the merged incisors are arranged in ascending order. The proposed excavation boundary is divided into multiple boundary segments according to the coordinates of adjacent arc lengths. For each boundary segment, the arc length coordinates of its midpoint are used to determine whether they fall within any incisor. Using the arc length range as the criterion, if the boundary segment falls within the range, it is determined as an uncuttable segment; if it does not fall within the range, it is determined as a cuttable segment. The starting arc length coordinates, ending arc length coordinates, segment status, and corresponding connection path of each boundary segment are recorded. Crossing parameters Establish a correspondence, and record the segment positions and segment states of all boundary sub-segments together as the contour constraint result. , This represents the contour constraint result. Defined as an indivisible zone The segmented description of the proposed excavation boundary constraints includes a unique arc length range, a unique segment state, and corresponding crossing parameters for each boundary sub-segment. .

[0097] In this embodiment, step S5 specifically includes:

[0098] Based on the contour constraint results Extract all boundary segments on the proposed excavation boundary. This indicates the contour constraint result, including all indivisible bands. The covered boundary segment is denoted as the restricted edge segment. ,in Indicates a non-cuttable strip. Indicates that it cannot be cut and has a serial number. Indicates a restricted edge segment. Indicates the index of the restricted edge segment, and sets the index of two adjacent restricted edge segments. Between, and not by any indivisible band A continuous boundary segment covered by the boundary is denoted as an adjustable edge segment. ,in Indicates an adjustable segment. This indicates that the edge segment number can be adjusted, and the edge segments that do not belong to the restricted edge segments will be included. It is also not an adjustable segment. The boundary segment is denoted as the reserved edge segment. ,in This indicates that the edge segment should be retained. This indicates that the edge segment number should be retained and the strip cannot be cut. The start and end points are determined using contour constraints. The incision point of the uncut band recorded in the middle and the end of the indivisible zone Confirmed, among which Indicates the starting point of the inseverable band. Indicates the end of the indivisible band;

[0099] To restrict the edge segment As a boundary constraint, for each adjustable edge segment To adjust the position, first calculate the directed area sign based on the ordered vertices of the proposed excavation boundary. When the directed area is positive, the proposed excavation boundary is determined to have a counterclockwise closing direction, and the right normal of each side segment is determined to be the outer normal. When the directed area is negative, the proposed excavation boundary is determined to have a clockwise closing direction, and the left normal of each side segment is determined to be the outer normal. For each adjustable side segment... Retain both ends and adjacent restricted edge segments Or retain the edge segment The connecting anchor points can only move along the outer normal direction of the adjustable side segment. The internal vertices of are denoted by the distance moved. The search increment is denoted as ,in Indicates the distance to adjust the position. Indicates the position adjustment increment, let Start from zero Increasing, in each In terms of position, the moved internal vertices are connected in the original vertex order and then connected to the two anchor endpoints to form candidate adjustment edges. ,in Indicates candidate adjustment edge segments, candidate adjustment edge segments The determination criterion is: candidate adjustment edge segment All internal vertices are offset by a distance along the outer normal. Candidate adjustment segments With any indivisible band No intersection points exist; candidate adjustment segment With any indivisible band There are no overlapping segments, candidate adjustment segments The two endpoints are respectively connected to the adjacent restricted edge segments Or retain the edge segment Keep the beginning and end connected;

[0100] Adjust all candidate segments All restricted edges and all reserved edges Boundary segments are spliced ​​in order of the proposed excavation arc length to form boundary chains. Adjacent segments within the boundary chains are then connected. During connection determination, the direction of the adjacent line segment at the endpoint of the previous segment is extracted as the endpoint tangential direction, and the direction of the adjacent line segment at the starting point of the next segment is extracted as the starting point tangential direction. The angle between the endpoint tangential direction and the starting point tangential direction is calculated as the connection angle. Finally, the Euclidean distance between the endpoint of the previous segment and the starting point of the next segment is calculated as the connection distance. The connection angle must not exceed an angle threshold. And the connection distance is not greater than the distance threshold. When adjacent edges are connected to form a continuous boundary, a closed boundary is formed when the starting point of the first segment and the ending point of the last segment simultaneously satisfy the same judgment condition. ,in Indicates the angle threshold. Indicates the distance threshold. This indicates a closed boundary. If any connection position does not meet the connection determination condition, the corresponding candidate adjustment segment is returned. The position adjustment process, and in the current position Continue on the basis Increasing;

[0101] For closed boundaries With all incuttable bands Intersection determination is performed, using a segment-by-segment detection method for closed boundaries. Each line segment and each indivisible band Each corresponding boundary segment is checked for intersections or overlapping sections. If no intersections or overlapping sections exist, the boundary is closed. Determined to adjust the excavation outline ,in This indicates that when adjusting the excavation outline, if there are intersecting sections, the results should be based on the outline constraint. Determine the adjustable edge corresponding to the location of the intersecting section. And adjust the corresponding candidate segments. Continue to increase the position and adjust the distance Then repeat the boundary splicing, connection determination, and intersection determination until the boundary is closed. With all incuttable bands There are no intersecting sections, forming an indivisible zone. Non-intersecting adjusted excavation profile .

[0102] In this embodiment, step S6 specifically includes:

[0103] The excavation outline will be adjusted. Represented as a sequence of boundary vertices arranged in closed order ,in This indicates an adjustment to the excavation outline. Indicates the first A boundary vertex, Indicates the boundary vertex index. This represents the total number of boundary vertices, and each boundary vertex... Using planar coordinates It means that, among them Indicates the first The lateral coordinates of each boundary vertex Indicates the first The vertical coordinates of each boundary vertex will be used to connect adjacent boundary vertices. and Connect them sequentially, and connect the boundary vertices. With boundary vertices Connect to form a closed cutting path ,in This represents a closed cut path. Decomposed into contour-cut segments according to the connection order. Each contour cutting segment Determined by a pair of adjacent boundary vertices, Indicates the first Cutting segments along the contour, cutting segments along the contour. and The segment merging determination is performed based on two conditions: first, two contour cutting segments. and The included angle is not greater than the preset direction merging threshold, and there are two contour cutting segments. The common endpoints are the same boundary vertex. When both conditions are met simultaneously, the two contour segments will be cut. Merge into a new contour cutting segment The starting point of the new contour cutting segment is taken from the starting point of the previous contour cutting segment, and the ending point of the new contour cutting segment is taken from the ending point of the next contour cutting segment, thus obtaining a sequence of contour cutting segments connected end to end.

[0104] The sequence of contour cutting segments is adjusted according to the excavation contour. The boundary vertices are arranged in a sequential order and aligned, and the first contour cutting segment is taken. The starting point is used as the cutting starting point. ,in Indicates the starting point of the cut, and cuts segments for each contour. Record the starting point of the cut Cutting endpoint and cutting direction ,in Indicates the first The starting point for cutting the outline segment. Indicates the first The cutting endpoint of the outline cutting segment. Indicates the first The outline cutting segment starts from the cutting point Pointing to the cutting end point When aligning the cutting direction and sequence, the segments are cut based on the current contour. Cutting endpoint As the next contour cutting segment The connection reference point is the next contour cutting segment. When the starting point coincides with the connecting reference point, maintain the next contour cutting segment. The endpoint order remains unchanged, and the next contour segment... When the endpoint coincides with the connecting reference point, swap the next contour cutting segment. The starting and ending points are determined so that the exchanged cutting starting point is consistent with the connection reference point, and the contour cutting segments are aligned sequentially. Connect them sequentially to form a continuous cutting trajectory. ,in Indicates a continuous cutting trajectory;

[0105] For continuous cutting trajectory Perform a closure determination and extract the first contour segment. Cutting start point and the last contour cutting segment Cutting endpoint ,in Indicates the total number of contour segments. This indicates the starting point of the first contour cutting segment. This indicates the cutting endpoint of the last contour segment, and the cutting start point. The plane coordinates are as follows , will cut the endpoint The plane coordinates are as follows ,in Indicates the horizontal coordinate of the cutting start point. Indicates the longitudinal coordinate of the cutting start point. Indicates the horizontal coordinate of the cut endpoint. The vertical coordinate of the cutting endpoint is represented. The planar distance between the cutting start point and the cutting endpoint is calculated based on the difference between the horizontal and vertical coordinates, and the planar distance is then compared with the closure threshold. In comparison, among them This represents the closure threshold, when the interplanar spacing is not greater than the closure threshold. At that time, determine the continuous cutting trajectory The closure condition is met when the interplanar spacing is greater than the closure threshold. At that time, re-execute the contour cutting segment. Alignment of sequence, continuous cutting trajectory Once the closure condition is met, the cutting equipment is controlled to cut segments according to each contour. Cutting start point Cutting endpoint and cutting direction Perform contour cutting sequentially, and make adjacent contour cutting segments Continuous entry and exit at common connection points create a continuous cutting trajectory. The enclosed base area is separated from the surrounding base.

[0106] After the base plate area is separated from the surrounding base plate, the excavation profile will be adjusted. The enclosed planar area is defined as the excavation boundary, and the bottom slab area within this boundary is excavated. During the excavation, the continuous cutting trajectory is first removed. The base plate entity within the enclosed area is then adjusted by modifying the excavation outline. Boundary vertex sequence The defined planar boundary is the excavation boundary. The soil within the planar boundary is excavated to form the equipment foundation pit. ,in This indicates the equipment foundation pit.

[0107] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for excavating foundation pits based on graph neural networks and interface fusion, characterized in that, include: S1. Obtain the non-destructive testing results of the area to be excavated, and extract the detection anomaly segments corresponding to the area to be excavated from the non-destructive testing results; S2. Obtain existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary, construct the detected anomaly segments, existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary into different types of nodes in the heterogeneous relationship graph network, and establish candidate connections according to spatial continuity to form a candidate connection graph; S3. Embed the anchor bolt penetration hidden variable layer into the candidate connection in the candidate connection diagram. Calculate the same bolt penetration parameter based on the continuous pointing relationship between the two end nodes of the candidate connection, the existing anti-buoyancy anchor bolt data and the bottom plate structure boundary. When the same bolt penetration parameter meets the penetration condition, retain the corresponding candidate connection and form a penetration chain segment. When the same bolt penetration parameter does not meet the penetration condition, remove the corresponding candidate connection to obtain the penetration relationship diagram. S4. According to the connection relationship diagram, trace the connection path along the connection chain segment connected to the corresponding node of the proposed excavation boundary, calculate the crossing parameters of each connection path relative to the proposed excavation boundary, and when the crossing parameters meet the crossing conditions, project the corresponding connection path as an uncut strip, and divide the proposed excavation boundary into segments by the uncut strip to form the contour constraint result. S5. Based on the contour constraint results and the proposed excavation boundary, adjust the proposed excavation boundary under the constraint of the inseverable zone to form an adjusted excavation contour that does not intersect with the inseverable zone. S6. Cut the bottom plate according to the adjusted excavation outline to form an equipment foundation pit that avoids the anti-buoyancy anchor.

2. The method for excavating foundation pits based on graph neural networks and interface fusion as described in claim 1, characterized in that, S1 specifically refers to: Obtain the non-destructive testing results corresponding to the area to be excavated, and determine the target detection range in the non-destructive testing results based on the boundary of the area to be excavated; The detection responses that are continuously distributed along the same extension direction within the target detection range are defined as the continuous response region, and the boundary of the continuous response region is located to obtain the start and end positions of the continuous response region. The segment's center plane position, burial depth, length, orientation, response width, response intensity, and attenuation difference at both ends are calculated based on the start and end positions of the continuous response region to form the segment characteristics corresponding to the continuous response region. Anomaly detection is performed on the segment features. When the response intensity meets the anomaly threshold, the length meets the length threshold, and the continuity of the direction meets the connectivity threshold, the corresponding continuous response region is identified as the detected anomaly segment, and the segment features are used as the node input features for the detected anomaly segment.

3. The method for excavating foundation pits based on graph neural networks and interface fusion as described in claim 1, characterized in that, S2 specifically refers to: Obtain existing anti-buoyancy anchor data, and extract the anchor starting point plane position, design direction, design length, design burial depth and spacing from the existing anti-buoyancy anchor data to form a six-dimensional node input feature; Obtain the boundary of the base slab structure and the boundary to be excavated. Extract the boundary center plane position, boundary orientation, thickness variation and interface category from the boundary of the base slab structure to form five-dimensional node input features. Extract the boundary center plane position, edge direction, edge length and excavation depth from the boundary to be excavated to form five-dimensional node input features. The center plane position, burial depth, length, orientation, response width, response intensity, and attenuation difference between the two ends of the detected anomaly segment are determined as the input features of the eight-dimensional node. The detected anomaly segment, existing anti-buoyancy anchor data, bottom plate structure boundary, and proposed excavation boundary are respectively constructed as different types of nodes in the heterogeneous relation graph network. A node encoding layer is used to encode the eight-dimensional node input features, six-dimensional node input features, and five-dimensional node input features respectively. The node encoding layer includes a first layer of thirty-two fully connected neurons and a second layer of sixteen fully connected neurons to obtain sixteen-dimensional node features corresponding to different types of nodes. Based on the sixteen-dimensional node features, node pairs that meet the distance threshold, direction threshold, and burial depth threshold are selected. The node pairs are then calculated for the node pairs for the distance, direction threshold, burial depth threshold, projection overlap, boundary separation value, anchor bolt direction consistency, and boundary intersection value, forming a seven-dimensional candidate connection feature. A candidate connection coding layer is used to encode the seven-dimensional candidate connection features. The candidate connection coding layer consists of a first layer of twenty-four fully connected neurons and a second layer of twelve fully connected neurons, resulting in twelve-dimensional connection features. Based on the sixteen-dimensional node features and twelve-dimensional connection features, when the anchor direction consistency meets the consistency threshold, the projection overlap meets the overlap threshold, the boundary separation value meets the separation threshold, and the boundary intersection value meets the intersection threshold, the spatial continuity relationship is determined and candidate connections are established to form a candidate connection map.

4. The method for excavating foundation pits based on graph neural networks and interface fusion as described in claim 1, characterized in that, S3 specifically refers to: Each candidate connection in the candidate connection diagram is used as a processing unit of the anchor bolt penetration hidden variable layer. The sixteen-dimensional node features corresponding to the two ends of the candidate connection, the twelve-dimensional connection features corresponding to the candidate connection, the sixteen-dimensional node features corresponding to the existing anti-buoyancy anchor bolt data, and the sixteen-dimensional node features corresponding to the bottom plate structure boundary are extracted to form a sixty-dimensional penetration judgment input. The sixty-dimensional through-determination input anchor is connected to the hidden variable layer. The hidden variable layer includes source branches, continuation branches, separation branches and fusion units. The source branches include twenty-four fully connected neurons and eight fully connected neurons, the continuation branches include thirty-two fully connected neurons and eight fully connected neurons, and the separation branches include sixteen fully connected neurons and eight fully connected neurons. The source branch is calculated to determine the source consistency relationship between the candidate connection end and the existing anti-buoyancy anchor data, and the source branch output is generated. The continuation branch is calculated, and the continuation relationship of the detected anomaly segment along the continuous pointing relationship is determined based on the direction, burial depth and length of the nodes at both ends of the candidate connection, and the continuation branch output is formed; The separation branches are calculated, and the separation relationship of the bottom plate structure boundary to the continuous pointing relationship is determined according to the relative position of the bottom plate structure boundary and the candidate connection, and the separation branch output is formed. The source branch output, the continuation branch output, and the split branch output are input into the fusion unit. The fusion calculation is performed using twelve fully connected neurons and one output neuron to obtain the one-dimensional same-rod connection parameter. When the same-rod connection parameter meets the connection condition, the corresponding candidate connection is retained and a connection chain segment is formed. When the same-rod connection parameter does not meet the connection condition, the corresponding candidate connection is eliminated. The candidate connection input is retained into the connection relationship update layer. The connection relationship update layer includes two layers of heterogeneous relationship propagation units. The first layer of heterogeneous relationship propagation units performs local connection updates on the connection chain segments, and the second layer of heterogeneous relationship propagation units performs path continuous updates on the connection chain segments to form a connection relationship graph.

5. The method for excavating foundation pits based on graph neural networks and interface fusion according to claim 1, characterized in that, S4 specifically refers to: Extract the connecting chain segments that are directly connected to the nodes corresponding to the proposed excavation boundary from the connecting relationship diagram, and trace along the connecting chain segments to the termination node connected to the existing anti-buoyancy anchor data according to the connection direction and node type of the candidate connections retained in the connecting relationship diagram, thus forming each connecting path. Each connection path is input into the connection relationship update layer. The first layer of heterogeneous relationship propagation unit is used to update the local connectivity relationship between the detected abnormal segments, existing anti-buoyancy anchor data, bottom plate structure boundary and proposed excavation boundary to form a local connection representation. Then, the second layer of heterogeneous relationship propagation unit is used to continuously update the local connection representation to obtain the sixteen-dimensional connection path features corresponding to each connection path. Extract the 16-dimensional node features corresponding to the nodes of the proposed excavation boundary for each through path, and concatenate the 16-dimensional through path features with the 16-dimensional node features to form a 32-dimensional input; The 32-dimensional input is fed into the traversal decision layer, which consists of 24 fully connected neurons and 3 output neurons. The 32-dimensional input is calculated to obtain the traversal value, the start point of the uncut zone, and the end point of the uncut zone. The three outputs together constitute the traversal parameters. The crossing parameters are determined. If the crossing value meets the crossing threshold and the starting point and ending point of the incisional zone fall on the corresponding edge segment of the proposed excavation boundary, the corresponding through path is determined to cross the proposed excavation boundary, and the boundary segment between the starting point and the ending point of the incisional zone is determined as the effective projection segment. The effective projection section is projected as an uncut zone. When the crossing value does not meet the crossing threshold, the corresponding through path is excluded from the uncut zone generation process. When multiple uncut zones fall into the same proposed excavation boundary section, the multiple uncut zones are merged. The proposed excavation boundary is segmented according to the starting point and ending point of the inseverable zone, and the segmented proposed excavation boundary is correlated with the crossing parameters of the corresponding through path to form the contour constraint result.

6. The method for excavating foundation pits based on graph neural networks and interface fusion according to claim 1, characterized in that, S5 specifically refers to: Based on the contour constraint results, extract the corresponding boundary segments of the incisional zone start point, incisional zone end point and proposed excavation boundary, and determine the restricted and adjustable boundary segments in the proposed excavation boundary. Using the restricted edge segment as the boundary constraint, the position of the adjustable edge segment is adjusted so that the adjustable edge segment is located outside the indivisible strip and remains connected to the restricted edge segment from beginning to end, thus forming a candidate adjustable edge segment. The candidate adjustment edge segment is spliced ​​with the proposed excavation boundary that is not covered by the uncut zone, and the adjacent edge segments after splicing are connected. When the connection angle meets the angle threshold and the connection distance meets the distance threshold, a closed boundary is formed. Intersection determination is performed between the closed boundary and the uncuttable zone. If there is no intersecting section, the closed boundary is determined as the adjusted excavation profile. If there is an intersecting section, the adjustable edge segment corresponding to the intersecting section is further adjusted according to the profile constraint result to form an adjusted excavation profile that does not intersect with the uncuttable zone.

7. The method for excavating foundation pits based on graph neural networks and interface fusion as described in claim 1, characterized in that, Step S6 is as follows: The closed cutting path is determined by adjusting the excavation profile, and the closed cutting path is decomposed into profile cutting segments that are connected end to end. Align the contour cutting segments sequentially, determine the cutting start point, cutting end point and cutting direction of each contour cutting segment, and form a continuous cutting trajectory; The continuous cutting trajectory is closed. When the distance between the cutting start point and the cutting end point meets the closure threshold, the bottom plate is contour cut according to the continuous cutting trajectory, so that the bottom plate area enclosed by the continuous cutting trajectory is separated from the surrounding bottom plate. The separated bottom plate area is excavated to form an equipment foundation pit corresponding to the adjusted excavation outline.

Images