A method and system for engineering quality monitoring based on BIM technology

Abstract

This application discloses a method and system for engineering quality monitoring based on BIM technology, relating to the field of building information modeling (BIM) technology. The method includes: converting defect information identified in construction images into three-dimensional defect bodies in the three-dimensional space of the BIM model through pose estimation and semantic mapping guided by BIM priors; inputting the attribute information of the three-dimensional defect bodies and the attributes of the attached BIM components into an acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal; acquiring on-site images from the same perspective after rectification, calculating a comprehensive acceptance index using a spatiotemporal dual-channel comparison algorithm based on BIM feature references, and automatically determining the acceptance conclusion; and automatically updating the quality status of the corresponding components in the BIM model based on the acceptance conclusion. This application aims to solve the problem of the disconnect between visual defect identification and quality control processes.

Description

Technical Field

[0001] This invention relates to the field of building information technology, and in particular to a method and system for monitoring engineering quality based on BIM technology. Background Technology

[0002] Currently, engineering quality monitoring technologies based on BIM and AI can automatically identify construction defects through images, point clouds, and other methods. However, existing solutions generally suffer from a broken process in the "identification-decision-execution" chain, specifically manifested as three isolated operations: First, the AI ​​system generates a defect detection report, but can only mark the defect type and location; second, quality management personnel need to manually review a large amount of design specifications and BIM model attribute information, and write rectification plans based on experience, a process that is time-consuming, error-prone, and inconsistent in standards; finally, after rectification is completed, manual comparison of before-and-after images with the BIM model is still required on-site to determine whether it is up to standard, forming the "last mile" automation bottleneck. This broken model prevents AI's powerful identification capabilities from being translated into immediate and accurate control instructions, resulting in a lengthy information transmission chain, slow response, and subjective verification of rectification results. Ultimately, this leads to a cycle from problem discovery to closed-loop rectification taking days, severely restricting the efficiency and digitalization level of construction quality control. Summary of the Invention

[0003] This invention provides a method for engineering quality monitoring based on BIM technology, comprising: Through BIM-guided pose estimation and semantic mapping, the defect information identified in the construction images is transformed into a three-dimensional defect volume in the three-dimensional space of the BIM model. The attribute information of the 3D defect body and the attributes of the attached BIM components are input into the acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal. After obtaining on-site images from the same perspective after rectification, a spatiotemporal dual-channel comparison algorithm based on BIM feature reference is used to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion. Based on the acceptance results, the quality status of the corresponding components in the BIM model is automatically updated.

[0004] The aforementioned engineering quality monitoring method based on BIM technology, through BIM-guided pose estimation and semantic mapping, transforms defect information identified in construction images into three-dimensional defect volumes in the three-dimensional space of the BIM model, including: Using the geometric features of BIM components as prior information, the six-degree-of-freedom pose of the camera that acquires images of the current construction area is calculated. By using the defect pixel contours obtained from semantic segmentation and combining them with the solved camera pose, the defects are mapped onto the corresponding 3D surface of the component in the BIM model, forming a 3D space with defect attribute labels.

[0005] The aforementioned engineering quality monitoring method based on BIM technology uses the geometric features of BIM components as prior information to calculate the six-degree-of-freedom pose of the camera acquiring images of the current construction area, including: Based on the on-site construction progress information, query and extract the three-dimensional geometric model of the components contained in the current construction area from the BIM model; Feature extraction is performed on the construction site images captured by the camera, and then matched with the projection features of the BIM component geometric model; Based on the matching relationship, the high-precision six-DOF pose of the camera is solved by iterative optimization.

[0006] The aforementioned engineering quality monitoring method based on BIM technology utilizes defect pixel contours obtained through semantic segmentation, combined with the calculated camera pose, to map defects onto the corresponding 3D surface of the component in the BIM model, forming a 3D spatial volume with defect attribute labels, including: A semantic segmentation network is used to process construction images and extract precise pixel-level contours of defect areas. Based on the camera pose, the defect pixel contours are mapped onto the surface of the 3D component in the BIM model through projection transformation. Defect attribute information is attached to the generated 3D defect patches to form a 3D defect body with attributes.

[0007] The aforementioned engineering quality monitoring method based on BIM technology inputs the attribute information of the 3D defect body and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal. This includes: Construct an acceptance knowledge graph that integrates BIM component attributes and construction quality acceptance specifications, and associate defect types, severity levels, specification clauses, and candidate rectification processes; Based on the attributes of the 3D defect body and the attributes of the BIM components to which the defect is attached, inference retrieval is performed in the acceptance knowledge graph to match applicable acceptance specification clauses and candidate rectification process sets. The rule engine reads the candidate rectification process set, combines the geometric features of defects with the characteristics of components, infers and generates quantitative rectification requirements, calculates the final rectification value, and generates a digital rectification order.

[0008] The aforementioned engineering quality monitoring method based on BIM technology acquires on-site images from the same perspective after rectification, utilizes a spatiotemporal dual-channel comparison algorithm based on BIM feature references to calculate a comprehensive acceptance index, and automatically determines the acceptance conclusion, including: The image after rectification is obtained through a mobile terminal. The six-degree-of-freedom pose of the camera is calculated based on the geometric features of the BIM model, and the consistency with the pose captured before rectification is determined. The time channel of the registered and rectified image is compared with that of the unrectified image. Local texture differences are calculated in the defect area to obtain the repair completeness index and measure the degree of visual restoration of the defect repair. Spatial channel comparison is performed between the registered and rectified images and the images before rectification to extract the actual component edge geometric features. The deviation is quantified between the actual component edge geometric features and the ideal geometric projection of the BIM model from the current perspective to obtain the geometric deviation index. Based on the repair completeness index output by the time channel and the geometric deviation index output by the space channel, the comprehensive acceptance index is calculated, and the acceptance conclusion is output.

[0009] A BIM-based engineering quality monitoring system includes: The defect 3D mapping module is used to transform defect information identified in construction images into 3D defect volumes in the 3D space of the BIM model through pose estimation and semantic mapping guided by BIM priors. The rectification order generation module is used to input the attribute information of the 3D defect body and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine, generate a structured digital rectification order, and send it to the construction terminal. The spatiotemporal comparison and acceptance module is used to acquire on-site images from the same perspective after rectification, and to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion using a spatiotemporal dual-channel comparison algorithm based on BIM feature reference. The BIM status update module is used to automatically update the quality status of the corresponding components in the BIM model based on the acceptance conclusion.

[0010] The beneficial effects achieved by this invention are as follows: Compared with existing technologies, it automates the entire process of identification, decision-making, execution, and acceptance, reducing the processing cycle of a single defect from the traditional "days" to "hours". By introducing BIM priors and knowledge graphs, it solves the problems of inconsistent rectification plans and strong subjectivity in acceptance, ensuring the compliance, consistency, and traceability of the control process. Overall, it realizes an unmanned closed loop from the discovery to the closure of construction quality problems, significantly improving the level of refined construction management. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0012] Figure 1 This is a flowchart of an engineering quality monitoring method based on BIM technology provided in Embodiment 1 of this application.

[0013] Figure 2 This is a schematic diagram of an engineering quality monitoring system based on BIM technology provided in Embodiment 2 of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example 1 like Figure 1 As shown, Embodiment 1 of this application provides an engineering quality monitoring method based on BIM technology, including: S1: Through pose estimation and semantic mapping guided by BIM priors, the defect information identified in the construction images is transformed into a three-dimensional defect body in the three-dimensional space of the BIM model. The process involves transforming defect information identified in construction images into attributed defect volumes in the 3D space of the BIM model through BIM-guided pose estimation and semantic mapping. This includes the following sub-steps: S11: Using the geometric features of BIM components as prior information, calculate the six-degree-of-freedom pose of the camera that acquires images of the current construction area; The process of calculating the six-degree-of-freedom pose of the camera acquiring the current construction area image, using the geometric features of BIM components as prior information, includes the following sub-steps: S111: Based on the on-site construction progress information, query and extract the three-dimensional geometric model of the components contained in the current construction area from the BIM model; Specifically, the construction management system can determine the set of BIM components corresponding to the camera's acquisition range based on the current construction section or process progress. Using this set identifier, it retrieves and loads the 3D geometric model data of these components from the BIM database, including precise geometric representations of component edges, corners, and surface features. The extracted geometric model serves as a geometrically complete 3D reference benchmark, providing a priori feature source for subsequent pose calculations without requiring on-site control points.

[0016] S112: Extract features from the construction site images captured by the camera and match them with the projection features of the BIM component geometric model; After acquiring color images of the construction site, feature point extraction is performed on these images to generate a set of two-dimensional feature point descriptors. Simultaneously, the three-dimensional geometric model of the BIM component is projected onto a two-dimensional plane based on an initial camera pose estimate, rendering a projected image of the BIM model's outline and significant geometric features. Corresponding projection feature descriptors are then extracted from this projected image. Subsequently, cross-matching is performed between the two sets of feature descriptors to establish multiple correspondences between two-dimensional pixel feature points in the construction images and projected feature points in the BIM model.

[0017] S113: Based on matching relationships, iteratively optimize the solution of the camera's high-precision six-DOF pose.

[0018] This paper constructs a problem to solve for camera position and orientation, using multiple sets of correspondences between 2D pixel feature points and projected feature points of the BIM model as constraints. The 3D coordinates of the geometric feature points of the BIM model and their corresponding pixel coordinates in the construction image are used as input. By repeatedly adjusting the camera pose parameters, the overall pixel distance between the projected feature points of the BIM model and the corresponding feature points extracted from the construction image is minimized when the BIM model is reprojected according to the current pose. When this distance converges to a preset threshold, the current pose parameters are determined as the final calculated high-precision six-DOF camera pose, and this six-DOF pose is associated with and stored with the defect identifier. This process utilizes the geometric information inherent in the BIM model throughout, without requiring any calibration objects or control points to be placed on-site.

[0019] S12: Using the defect pixel contours obtained from semantic segmentation, combined with the solved camera pose, the defects are mapped onto the corresponding 3D surface of the component in the BIM model to form a 3D space with defect attribute labels.

[0020] The process involves using the defect pixel contours obtained from semantic segmentation, combined with the calculated camera pose, to map the defects onto the corresponding 3D surface of the component in the BIM model, forming a 3D volume with defect attribute labels. This includes the following sub-steps: S121: Use a semantic segmentation network to process construction images and extract precise pixel-level contours of defect areas; The same construction site image is input into a pre-trained semantic segmentation network (e.g., DeepLabV3+). This network classifies each pixel in the image, identifying pixel sets belonging to various types of construction defects. Based on the pixel-by-pixel category predictions output by the network, a mask image the same size as the input image is generated, in which pixels representing defect areas are marked. By performing contour tracing on this mask image, a precise closed pixel contour line representing the boundary of the defect area is obtained.

[0021] S122: Based on the camera pose, the defect pixel contours are mapped onto the surface of the 3D component in the BIM model through projection transformation; Based on the camera's high-precision six-DOF pose, a projection relationship from the camera's imaging plane to the 3D space of the BIM model is constructed. Starting from the camera's optical center, a ray is emitted that passes through every boundary pixel on the defect pixel contour. The first intersection point of each ray with the 3D geometric model of the BIM component is calculated. Since the BIM model accurately represents the spatial position and surface shape of each component, these intersection points accurately fall on the surface of the component that the camera was focused on during the image capture, forming a set of 3D projection points on that surface. By sequentially connecting adjacent projection points, a 3D defect patch corresponding to the image defect area can be defined on the BIM model.

[0022] S123: Attach defect attribute information to the generated 3D defect patch to form a 3D defect body with attributes.

[0023] In the BIM model, a 3D defect patch is attached as a new quality defect object to the corresponding component. Attribute information is written to this object, including at least: the defect type and confidence level obtained from the semantic segmentation network recognition results; the actual physical size of the defect calculated based on the mask area and projection relationship; and the severity level assessed based on the defect type and location. Thus, a defect that originally existed isolated in a 2D image is transformed into a 3D defect body within the BIM model space, with accurate location, clear boundaries, and complete business attributes.

[0024] S2: Input the attribute information of the 3D defect body and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine to generate a structured digital rectification order and send it to the construction terminal; The process involves inputting the attribute information of the 3D defect and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal. This includes the following sub-steps: S21: Construct an acceptance knowledge graph that integrates BIM component attributes and construction quality acceptance specifications, and associate defect types, severity levels, specification clauses, and candidate rectification processes; Specifically, the attribute information of all components is extracted from the BIM model of the current construction area. For each component, a unique component identifier is generated, and the material type, strength grade, geometric thickness, design stress ratio, and environmental exposure category of the component are extracted as structured data and stored in the graph database in the form of component entity nodes. At the same time, attribute relationship edges between component entities and material entities and geometric entities are established, such as "Component A has C30 concrete material" and "Component A has a thickness of 200 mm".

[0025] The graph database includes attribute information of BIM components (such as material, strength, thickness, stress ratio, etc.) and knowledge of clauses related to defect judgment and rectification in the construction quality acceptance specifications. Subsequently, the clauses in the construction quality acceptance specifications were digitally decomposed. For each clause related to the judgment and rectification of quality defects, the applicable defect type, applicable material and strength conditions, rectification action description and benchmark quantitative requirements were extracted and constructed as condition entities, action entities and benchmark parameter entities. "Trigger" relationship, "required action" relationship and "have benchmark value" relationship were established in the graph database.

[0026] For example, the specification states that "when the width of a crack in a cast-in-place concrete structure is greater than 0.3 mm, it should be repaired by surface sealing, and the dry film thickness of the repair coating should not be less than 1.5 mm". The conditions "defect type = crack", "material = concrete", and "crack width > 0.3 mm" are extracted, the action is "surface sealing", and the reference thickness is "1.5 mm", and then linked through the relationship.

[0027] The aforementioned relationship linking refers to establishing labeled relationship edges between different entity nodes in a graph database, thereby enabling two originally independent data points to have a semantic association. Next, based on the defect type ontology node and severity level node, a "defect triggering" relationship edge is established between each defect type and the relevant specification clause node. According to the severity level classification of defects in the specification, the severity level attribute is attached to the relationship edge, thus forming a multi-hop reasoning path from defect type, severity, material, component category to specific acceptance clauses and rectification actions, and finally obtaining the acceptance knowledge graph.

[0028] For example, AI identifies a defect type as "crack" with a severity level of "2" (i.e., crack width exceeding 0.3mm) in a 3D defect body, and this defect is anchored to "Component A" (material: C30 concrete). In the knowledge graph, starting from the node "Defect Type = Crack", the system traverses along the "Defect Trigger" relationship edge to the relevant specification clause nodes. First, it filters out clause nodes whose severity level attribute matches the current defect severity level (level 2) on the relationship edge, obtaining a candidate clause set. Then, along the "Applicable Component Conditions" relationship edge, it filters out clause nodes that do not meet the requirement of "Material = C30 Concrete" (because crack repair requirements differ for different material grades). This confirms that the clause matches the C30 concrete condition for Component A.

[0029] Then, starting from the matched clause node, follow the "required action" relationship edge to jump to the "rectification action" node and obtain the rectification action "surface sealing repair".

[0030] Finally, starting from the rectification action node, follow the "with benchmark value" relationship edge to jump to the "benchmark parameter" node to obtain the benchmark quantification requirements. (The dry film thickness of the repair coating should not be less than 1.5 mm).

[0031] S22: Based on the attributes of the 3D defect body and the attributes of the BIM components to which the defect is attached, perform reasoning retrieval in the acceptance knowledge graph to match applicable acceptance specification clauses and candidate rectification process sets; First, the attributes of the 3D defect are analyzed, including reading the defect type code and severity level from the semantic tags of the 3D defect, and calculating the defect volume from the 3D mesh data of the defect. The projected area of ​​defects on the surface of the component and the maximum depth of the defect Simultaneously, by using the BIM component identifier anchored to the three-dimensional defect body, the attribute set of the component is obtained, which includes at least the material designation, strength grade, component thickness, component stress ratio, and environmental exposure category code.

[0032] In the acceptance knowledge graph, starting with the defect type code and severity level, the system traverses along the "defect triggering" relationship to the corresponding specification clause node set, and simultaneously filters out clause nodes that precisely match the current component material, strength grade, and other conditions. Then, for each matching clause node, the associated rectification action node is obtained along the "required action" relationship, and the baseline rectification parameters required for that action are obtained along the "have baseline value" relationship. This forms a set of candidate rectification processes, with each candidate item containing a description of the rectification process, material requirements, and at least one baseline quantitative value.

[0033] S23: The rule engine reads the candidate rectification process set, combines the defect geometric features and component characteristics, infers and generates quantitative rectification requirements, calculates the final rectification value, and generates a digital rectification order.

[0034] The entire set of candidate rectification processes is read into the working memory of the rule engine one by one. Each entry's rectification action identifier, rectification process description text, recommended materials list, and baseline parameters are stored as a single fact unit. Simultaneously, the obtained defect volume is... Projected area Maximum depth and BIM component properties (i.e., component thickness) Component stress ratio Environmental exposure category codes ), and inject them into the working memory at the same time.

[0035] Subsequently, the rule engine's pattern matcher extracts four facts from the working memory: defect type code, severity level, component material number, and component strength level. These facts are then compared one by one with all production rules pre-stored in the rule base. Each rule in the rule base is stored as a quadruple condition group (IF Defect Type == X AND Severity Level >= L AND Component Material == M AND Component Strength Level == S THEN Activate Remedial Action A). When all condition fields of a rule match the four fact values ​​in the working memory, the rule is marked as a successful match and added to the candidate rule set.

[0036] If only one rule in the candidate rule set matches successfully, it is directly selected as the winning rule; if multiple rules match successfully at the same time, conflict resolution is performed. Conflict resolution is performed in two levels: the first level compares the priority attributes of the rules, and the one with the higher priority value wins; the second level compares the specificity labels when the priorities are the same, and the one with more conditional constraints wins.

[0037] Once the winning rule is determined, the execution engine of the rule engine extracts the rectification action identifier from the THEN part of the rule, checks back the acceptance knowledge graph, and reads the corresponding complete rectification process description text, recommended material list, and benchmark quantitative parameters along the association edges between the rectification action node and the process description node, material node, and benchmark parameter node. , and instantiate the rectification actions.

[0038] For example, when the defect type is crack, the severity level is greater than or equal to 2, and the component material is C30 concrete, the "Surface Sealing Repair" action is activated and instantiated. .

[0039] The final quantitative indicators for rectification were then calculated based on the formula. and will The specific formula for the value used as the structured quantitative rectification standard for this defect item is as follows: ,in, The absolute minimum allowable rectification value specified in the normative clause is extracted by the rule engine from the knowledge graph; The component importance adjustment coefficient is obtained by mapping the importance level of BIM components. The benchmark rectification parameters are provided by the benchmark value entity that is attached to the action node corresponding to the matching clause in the knowledge graph, reflecting the standard rectification quantitative reference value given by the construction acceptance specification for the current defect type and component material conditions; The defect volume is calculated from the mesh or voxel data of the three-dimensional defect body. It reflects the actual space volume occupied by the defect inside the component. The larger the volume, the higher the amount of repair material and the stronger the process tend to be. Pre-calibrate the reference defect volume corresponding to this component type; Based on component thickness The determined critical depth is set at 0.3. The smaller of the allowable values; The maximum depth of the defect is obtained by traversing sampling points along the normal direction of the component surface in the three-dimensional defect geometric model. It reflects the maximum distance the defect penetrates into the component and is a key geometric indicator for judging whether the defect endangers the core of the structure. The environmental exposure category code is numerically mapped. The environmental exposure category extracted from the component attributes is converted into a continuous value according to a preset mapping table, which reflects the severity level of the environment in which the component is located. The higher the level, the more stringent the requirements for the durability and protection of the rectification layer. The area of ​​the inspected surface of the component is calculated from the BIM model; The projected area of ​​the defect on the component surface is calculated by integrating the projection of the three-dimensional defect surface nodes along the normal direction of the component surface, reflecting the size of the area covered by the defect on the component surface. The stress ratio of a component is read from the attribute data of the BIM component. It reflects the ratio of the actual load currently borne by the component to its design bearing capacity. It is a key structural safety parameter for determining whether a component is in a high stress state. The force influence factor is defined as follows: , To design the normal stress ratio threshold, it is usually taken as 0.6; These are constant coefficients determined through regression training on a large number of actual defect rectification cases; in this embodiment, we take... ;in, The defect volume scaling factor characterizes the effect of changes in the ratio of the defect volume to the reference defect volume on the quantitative indicators of rectification. Sensitivity to the overall amplification or reduction trend. The larger the value, the greater the impact of the volume factor. The more significant the contribution of the fundamental correction; The defect volume nonlinearity index reflects the attenuation characteristic of the defect volume influence gradually saturating as the volume increases. At that time, the increase in volume has an effect on The marginal impact diminishes, which aligns with the pattern that the demand for additional reinforcement slows down once the size of defects in an engineering project reaches a certain level. This is a depth sensitivity coefficient, which adjusts the intensity of the impact of the deviation of the maximum defect depth from the critical depth of the component on the amount of rectification required. The larger the depth, the more it exceeds the critical threshold. The more drastic the increase; The depth critical offset threshold characterizes the normalized starting point at which the defect depth begins to have a significant impact on the safety of the component. and The ratio is less than hour, The function's contribution to this term approaches zero, meaning that minor depth deviations do not trigger additional tightening measures. The environmental exposure amplification factor reflects the linear amplification factor of quantitative rectification indicators for each increase in environmental severity level. Determined from historical engineering data, this represents an additional safety margin reflecting the repair quality requirements under different erosion environments; This is the defect area ratio factor, which characterizes the basic sensitivity of the proportion of the defect projection area on the inspected surface of the component to the amount of rectification. The larger the area, the greater the impact on the same area percentage. The stronger the correction; This is a nonlinear exponent of the defect area, reflecting the marginal contribution saturation rate of the rectification requirements as the defect coverage area increases. When the area ratio increases to a certain extent, it has an impact on... The additional pulling gradually weakens, which is consistent with the process judgment logic of transitioning from partial repair to overall flipping.

[0040] The three-dimensional coordinates of the defective object are then used as the rectification location, and the rectification process is compared with the recommended materials and quantitative standards. The data, along with defect identifiers, component identifiers, and timestamps, is encapsulated into a structured digital rectification order. According to the correspondence between components and process responsible persons in the preset responsibility matrix, the rectification order is pushed to the mobile terminal of the construction responsible person. The rectification order is visualized in the mobile terminal as an augmented reality view with BIM model overlay. The responsible person can intuitively see the three-dimensional location of the defect, the quantitative rectification requirements, and the recommended process on the real scene.

[0041] S3: Obtain on-site images from the same perspective after rectification, and use a spatiotemporal dual-channel comparison algorithm based on BIM feature reference to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion. After acquiring on-site images of the rectification, the camera pose is calculated using the geometric features of the BIM model and compared with the pose taken before rectification. The images after rectification are then registered at the pixel level with the images before rectification. A spatiotemporal dual-channel comparison algorithm based on BIM feature reference is executed to calculate the repair completeness index of the time channel and the geometric deviation index of the spatial channel, respectively. A comprehensive acceptance index is obtained through a weighted fusion judgment formula, and the acceptance conclusion is automatically determined and the quality status of the BIM model is updated.

[0042] The process includes acquiring on-site images from the same perspective after rectification, using a spatiotemporal dual-channel comparison algorithm based on BIM feature references to calculate a comprehensive acceptance index, and automatically determining the acceptance conclusion. This process includes the following sub-steps: S31: Obtain the image after rectification through the mobile terminal, calculate the six-degree-of-freedom pose of the camera based on the geometric features of the BIM model, and determine the consistency with the pose captured before rectification. Specifically, when the construction supervisor completes on-site rectification and submits an acceptance request via mobile terminal, the terminal acquires an image of the construction area from the current perspective and calls the BIM prior-guided pose estimation algorithm described in step S1. Using the three-dimensional geometric features of the current construction area BIM model retrieved in step S1 as prior data, ORB feature points in the currently acquired image are extracted and matched with feature points of the synthetic image generated by rendering the BIM model under the preset perspective in that area. The three-dimensional position coordinates and three-dimensional orientation angle of the camera in the BIM model coordinate system are solved through the multi-point perspective projection equation to obtain the current camera pose.

[0043] The calculated current camera pose is then compared with the pre-rectification camera pose (high-precision six-DOF pose) recorded and stored in step S1 when identifying the same defect. The Euclidean distance between the two camera positions and the angle between the two line-of-sight directions are calculated. If the Euclidean distance is less than a preset position threshold and the line-of-sight angle is less than a preset angle threshold, the viewpoint consistency requirement is met, and the currently acquired image is used as a valid post-rectification image for subsequent processing. Otherwise, the construction manager is guided to move back to the pre-rectification shooting pose via the mobile terminal interface with direction and distance indicators, and the image is re-acquired until the viewpoint consistency condition is met.

[0044] By utilizing the mapping relationship between the pixel coordinates of the defective area in the pre-rectification image and the surface coordinates of the 3D component in the BIM model, and using the 3D surface of the BIM model as an intermediate reference plane, perspective transformation is used to remap the defective area in the post-rectification image (after passing the viewpoint consistency verification) back to the pixel coordinate system of the pre-rectification image. This generates a registered post-rectification image that is precisely aligned pixel-level with the pre-rectification image, denoted as [image name missing]. .

[0045] S32: Perform time channel comparison between the registered and rectified image and the image before rectification, calculate the local texture difference in the defect area, obtain the repair completeness index, and measure the degree of visual restoration of the defect repair. Specifically, the defect pixel mask corresponding to the defect in the image before rectification is read from the metadata of the 3D defect. The defect pixel mask is a binary image that marks all pixel positions covered by the original defect. A morphological dilation operation is performed on the defect pixel mask, with the dilation radius being 0.15 times the equivalent diameter of the defect region, to generate an expanded mask. The difference between the extended mask and the defective pixel mask is used as the edge transition zone. The area obtained after performing a morphological erosion operation on the defective pixel mask (with the erosion radius being 0.5 times the expansion radius) is used as the internal filling area. The edge transition zone and the internal filling area together constitute the complete region of interest for time channel alignment.

[0046] Within the internal filling area, the image before rectification is calculated pixel by pixel. Image after registration and rectification The local texture differences between them. For each pixel within the internal fill area, take a size of [value] centered on that pixel. Extract the images before rectification from the local window. and the image after registration and rectification Within this window, the local binary pattern feature vectors are analyzed, and the chi-square distance between the two feature vectors is calculated to form the internal texture difference value. .

[0047] At the same time, within the same local window and Calculate the local value of the structural similarity index . The deviations of the two windows in three dimensions—brightness, contrast, and local structure—are comprehensively compared, with values ​​ranging from [0, 1]. Values ​​closer to 1 indicate greater similarity between the two windows. The complement of the complement is taken as the structural degradation difference value at that window location, representing the degree of local structural mismatch between the repaired area and the original surface.

[0048] Internal texture difference values ​​for all pixels within the internal fill area Take the mean and denote it as , for structural degradation difference value Take the mean and denote it as .

[0049] exist Calculate pixel by pixel within the region and Differences in edge transition between them. For Calculate for each pixel within the range and The ratio of gradient magnitudes And the cosine of the angle between the gradient directions of the two images at that pixel. The closer the cosine value is to 1, the more consistent the edge direction and sharpness are before and after the rectification, resulting in a natural edge transition. The average gradient direction consistency index is calculated for all pixels within the range, denoted as . .

[0050] Subsequently, the completeness index of repair was calculated according to the formula. The specific calculation formula is as follows: ,in, for The average texture chi-square distance within the area reflects the overall degree of variation in the surface texture of the defective internal area before and after rectification; the smaller the value, the closer the repair material is to the original micro-texture of the base layer, and the less obvious the repair traces; for The average structural degradation difference value within the region reflects the degree of overall mismatch between the repaired region and the original region in three dimensions: brightness, contrast, and local structure. The smaller the value, the better the overall restoration of the structure after repair; for The average cosine of the gradient direction angle within the region characterizes the consistency of the edge orientation before and after rectification; when the repaired edge is completely consistent with the original component texture orientation... Approaching 1; if there is overflow of repair material, shrinkage cracks, or boundary misalignment at the edge, then Significantly reduced; Represents the texture difference sensitivity coefficient, controlling The rate of decay of the contribution to the repair completeness index; The larger the value, the greater the difference in texture quality. The stronger the punishment; in this embodiment, we take... ; This represents the structural degradation amplification factor, used to impose a joint amplification penalty when texture differences and structural degradation occur simultaneously; even if Lower, if at the same time Larger (i.e., textures appear similar but the structure has degraded), penalty item This will significantly lower the score of internal repair; in this embodiment, we take... ; Indicates the edge transition sharpness index, which can be adjusted. Nonlinear contribution to edge scoring; when hour, Even a slight decrease is amplified, reflecting the strict criterion in engineering quality acceptance that "poor edge treatment quality is unacceptable"; this embodiment takes... ; This represents the internal filling weighting coefficient, reflecting the proportion of the quality of internal defect repair in the overall evaluation; This represents the edge transition weight coefficient, reflecting the proportion of edge repair quality in the overall evaluation. Both factors must satisfy... Based on the differences in emphasis on edge treatment and internal filling for common defects such as cracks, honeycombing, and pitting in the construction quality acceptance specifications, this embodiment adopts... , ; The value range is [0, 1]; when When the value approaches 1, it indicates that the internal texture is highly restored and the edge transition is smooth and natural, indicating an extremely high degree of restoration completeness; when... When the value approaches 0, it indicates that there is a significant difference between the repaired area and the original component surface, and the rectification effect is unqualified.

[0051] The above constant coefficients and weight The value of is determined by collecting several defect rectification cases in historical projects that have been manually inspected and rated by the supervisor. The results of the manual rating are used as labels for regression training. In subsequent steps, the results of this inspection are fed back to the knowledge graph for continuous iterative optimization.

[0052] S33: Perform spatial channel comparison between the registered and rectified image and the image before rectification, extract the actual component edge geometric features, quantify the deviation between the actual component and the ideal geometric projection of the BIM model from the current perspective, and obtain the geometric deviation index. Specifically, using the current camera pose and camera intrinsic parameter matrix, the ideal 3D surface model of the BIM component anchored to the corresponding defect is transformed onto the image plane through perspective projection, generating a set of ideal edge projection lines for the component. . It includes the structural outline of the components, the boundary lines of the pre-reserved holes, and the two-dimensional projections of each geometric edge.

[0053] Image after registration and rectification Subpixel-level edge detection is performed. The second-order directional derivative is calculated on the grayscale image. Within regions where the gradient magnitude is greater than a preset threshold, the zero-crossing points of the second-order derivative are located along the gradient direction, yielding a set of actual component edge points with subpixel precision. .

[0054] right Centered in the defect pixel mask Iterate through every edge point in its surrounding and neighboring areas. For each projected line segment in the region, calculate the perpendicular Euclidean distance from that point to the nearest line segment, which is used as the geometric deviation value for that point. Then, take the root mean square of the geometric deviation values ​​for all valid edge points within that region to obtain the geometric deviation index of the spatial channel. . It characterizes the degree of dispersion of the actual surface shape of the repaired component relative to the design model; the smaller the value, the higher the geometric compliance.

[0055] S34: Calculate the comprehensive acceptance index based on the repair completeness index output by the time channel and the geometric deviation index output by the space channel, and output the acceptance conclusion.

[0056] Specifically, based on the restoration completeness index output by the time channel and the geometric deviation index output by the spatial channel, a comprehensive acceptance index is calculated using a formula. The specific calculation formula is as follows: ,in, Weighting of the time channel repair completeness index. For the geometric compliance weights of the spatial channel, both satisfy... In this embodiment, we take , ; The repair completeness index is used to adjust the repair completeness index. The degree of nonlinearity in the contribution to the overall score is taken in this embodiment. ; The allowable benchmark value for the geometric deviation index of the component is preset according to the BIM component type and construction acceptance specifications. For example, the allowable deviation of the flatness of the concrete surface is 2.0mm. The geometric deviation attenuation coefficient controls the geometric deviation index. The decay rate of the contribution of the scoring item is taken in this embodiment. ; This is the cross-penalty coefficient, used to apply a joint penalty when the repair is incomplete and the geometric deviation index exceeds the allowable range. In this embodiment, it is taken as... ; The nonlinear exponent for cross-penalty is used to adjust the rate of increase in the intensity of the joint penalty. In this embodiment, we take... .

[0057] The constant coefficients and weights mentioned above were determined by collecting several defect rectification cases from historical projects that were manually inspected by the supervisors, and then using the manual judgment conclusions as labels for regression training.

[0058] The calculated comprehensive acceptance index Compared with the preset pass / fail threshold Compare. If If the rectification is successful, the system will automatically determine that the rectification is acceptable; otherwise, it will determine that the rectification is unacceptable and a second rectification is required. Simultaneously, a quantitative deviation record will be output, including at least the repair completeness index. and geometric deviation index The specific numerical value is provided for quality traceability reference. The pass / fail threshold in this embodiment... Take 0.65.

[0059] S4: Based on the acceptance conclusion, automatically drive the update of the quality status of the corresponding components in the BIM model.

[0060] Obtain the comprehensive judgment conclusion of this acceptance. If the acceptance is deemed satisfactory, update the quality status attribute of the component anchored by the defect in the BIM model from "Pending Rectification" to "Accepted," and simultaneously change the status label of the 3D defect to "Closed Loop." If the acceptance is deemed unsatisfactory, update the component's quality status attribute from "Pending Rectification" to "Secondary Rectification," and simultaneously change the status label of the 3D defect to "Pending Secondary Rectification," and update the geometric deviation index calculated in this acceptance. and repair completeness index The information is written into the attribute fields of the 3D defect body as a quantitative reference for secondary rectification.

[0061] Based on the updated 3D defect status labels, the preset visual style rules within the BIM model are triggered: when the status label is "closed," the display color of the 3D defect is switched from red when the defect is marked to green, and its transparency parameter is increased to 70%, making it appear as a semi-transparent overlay on the component surface in the BIM model, indicating that the defect has been repaired and closed but can still be traced and viewed; when the status label is "awaiting secondary rectification," the display color of the 3D defect is switched from red to orange, maintaining an opacity of 100%, and a flashing boundary outline is overlaid around the defect to clearly identify the defect location that needs to be re-examined in the BIM model.

[0062] The occurrence node of the status change record is marked with the current system timestamp. A record is added to the component attribute change log of the BIM model, which includes the status value before and after the change, the acceptance conclusion that triggered the change, the acceptance time, and the associated defect identifier, thus completing the closed loop of the full life cycle traceability chain of the component quality status.

[0063] All data generated during this acceptance inspection, including pre-rectification images, post-rectification images, texture difference maps, repair completeness indicators, geometric deviation indicators, comprehensive acceptance index, final judgment conclusions, and structured digital rectification form content, will be linked with the unique identifier of the defect and the unique identifier of the anchored component. This will be persistently stored as the acceptance file for the defect throughout its entire lifecycle. Simultaneously, an index link pointing to this acceptance file will be created on the 3D defect object in the BIM model, and the acceptance file will be used for reverse indexing to the spatial location of the corresponding component in the BIM model, forming a bidirectional retrieval path between the BIM model and the acceptance data.

[0064] Next, the rectification case was structurally encapsulated, including: defect type and severity level, component material and strength level, rectification action identifier and benchmark parameters, quantitative rectification standards, the completeness of repair and geometric deviation in this acceptance, comprehensive acceptance index, and final judgment conclusion. This structured case data was then added as a new instance node to the acceptance knowledge graph, and semantic relationships were established between this case node and the corresponding specification clause node, rectification action node, and component material node.

[0065] The actual acceptance result (pass or fail) serves as feedback signal to dynamically adjust the priority attributes of relevant rules in the rule engine. If the rectification and acceptance are successful and... and If all performances are excellent, the priority weight of the rule under the condition of this component type will be increased; if the acceptance fails, the priority weight of the rule will be reduced and a review flag for the applicable conditions or recommended actions of the rule will be triggered.

[0066] Example 2 like Figure 2 As shown, Embodiment 2 of this application provides an engineering quality monitoring system based on BIM technology, including: 3D Defect Mapping Module 21: Used to transform defect information identified in construction images into 3D defect bodies in the 3D space of the BIM model through pose estimation and semantic mapping guided by BIM priors. Rectification order generation module 22: Used to input the attribute information of the 3D defect body and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine, generate a structured digital rectification order, and send it to the construction terminal; Spatiotemporal comparison and acceptance module 23: used to acquire on-site images from the same perspective after rectification, and to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion using a spatiotemporal dual-channel comparison algorithm based on BIM feature reference; BIM Status Update Module 24: Used to automatically update the quality status of the corresponding components in the BIM model based on the acceptance conclusion.

[0067] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for engineering quality monitoring based on BIM technology, characterized in that, include: Through BIM-guided pose estimation and semantic mapping, the defect information identified in the construction images is transformed into a three-dimensional defect volume in the three-dimensional space of the BIM model. The attribute information of the 3D defect body and the attributes of the attached BIM components are input into the acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal. After obtaining on-site images from the same perspective after rectification, a spatiotemporal dual-channel comparison algorithm based on BIM feature reference is used to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion. Based on the acceptance results, the quality status of the corresponding components in the BIM model is automatically updated.

2. The engineering quality monitoring method based on BIM technology according to claim 1, characterized in that, Through BIM-guided pose estimation and semantic mapping, defect information identified in construction images is transformed into 3D defect volumes in the 3D space of the BIM model, including: Using the geometric features of BIM components as prior information, the six-degree-of-freedom pose of the camera that acquires images of the current construction area is calculated. By using the defect pixel contours obtained from semantic segmentation and combining them with the solved camera pose, the defects are mapped onto the corresponding 3D surface of the component in the BIM model, forming a 3D space with defect attribute labels.

3. The engineering quality monitoring method based on BIM technology according to claim 2, characterized in that, Using the geometric features of BIM components as prior information, the six-degree-of-freedom pose of the camera acquiring images of the current construction area is calculated, including: Based on the on-site construction progress information, query and extract the three-dimensional geometric model of the components contained in the current construction area from the BIM model; Feature extraction is performed on the construction site images captured by the camera, and then matched with the projection features of the BIM component geometric model; Based on the matching relationship, the high-precision six-DOF pose of the camera is solved by iterative optimization.

4. The engineering quality monitoring method based on BIM technology according to claim 2, characterized in that, Using the defect pixel contours obtained from semantic segmentation, combined with the calculated camera pose, the defects are mapped onto the corresponding 3D surface of the component in the BIM model, forming a 3D volume with defect attribute labels, including: A semantic segmentation network is used to process construction images and extract precise pixel-level contours of defect areas. Based on the camera pose, the defect pixel contours are mapped onto the surface of the 3D component in the BIM model through projection transformation. Defect attribute information is attached to the generated 3D defect patches to form a 3D defect body with attributes.

5. The engineering quality monitoring method based on BIM technology according to claim 1, characterized in that, The attribute information of the 3D defect and the attributes of the attached BIM components are input into the acceptance knowledge graph and rule engine to generate a structured digital rectification order, which is then sent to the construction terminal, including: Construct an acceptance knowledge graph that integrates BIM component attributes and construction quality acceptance specifications, and associate defect types, severity levels, specification clauses, and candidate rectification processes; Based on the attributes of the 3D defect body and the attributes of the BIM components to which the defect is attached, inference retrieval is performed in the acceptance knowledge graph to match applicable acceptance specification clauses and candidate rectification process sets. The rule engine reads the candidate rectification process set, combines the geometric features of defects with the characteristics of components, infers and generates quantitative rectification requirements, calculates the final rectification value, and generates a digital rectification order.

6. The engineering quality monitoring method based on BIM technology according to claim 1, characterized in that, After obtaining on-site images from the same perspective following rectification, a spatiotemporal dual-channel comparison algorithm based on BIM feature references is used to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion, including: The image after rectification is obtained through a mobile terminal. The six-degree-of-freedom pose of the camera is calculated based on the geometric features of the BIM model, and the consistency with the pose captured before rectification is determined. The time channel of the registered and rectified image is compared with that of the unrectified image. Local texture differences are calculated in the defect area to obtain the repair completeness index and measure the degree of visual restoration of the defect repair. Spatial channel comparison is performed between the registered and rectified images and the images before rectification to extract the actual component edge geometric features. The deviation is quantified between the actual component edge geometric features and the ideal geometric projection of the BIM model from the current perspective to obtain the geometric deviation index. Based on the repair completeness index output by the time channel and the geometric deviation index output by the space channel, the comprehensive acceptance index is calculated, and the acceptance conclusion is output.

7. An engineering quality monitoring system based on BIM technology, characterized in that, include: The defect 3D mapping module is used to transform defect information identified in construction images into 3D defect volumes in the 3D space of the BIM model through pose estimation and semantic mapping guided by BIM priors. The rectification order generation module is used to input the attribute information of the 3D defect body and the attributes of the attached BIM components into the acceptance knowledge graph and rule engine, generate a structured digital rectification order, and send it to the construction terminal. The spatiotemporal comparison and acceptance module is used to acquire on-site images from the same perspective after rectification, and to calculate the comprehensive acceptance index and automatically determine the acceptance conclusion using a spatiotemporal dual-channel comparison algorithm based on BIM feature reference. The BIM status update module is used to automatically update the quality status of the corresponding components in the BIM model based on the acceptance conclusion.

Images