Large BIM model optimization system based on intelligent loading

Abstract

The invention relates to the technical field of building information model optimization, in particular to a large-volume BIM model optimization system based on intelligent loading, which comprises a model analysis unit, a function identification unit, a simplified execution unit and a model recombination unit, and is characterized in that the model analysis unit reads an IFC file and separates component data; the function identification unit outputs a decoration component (first class) set, a non-bearing structure (second class) set and a bearing component (third class) set through geometric and semantic dual-channel analysis, the simplified execution unit performs edge folding on the first class, performs vertex clustering on the second class and skips simplification on the third class, and the model recombination unit binds metadata and constructs a dynamic octree index. The problems that a traditional simplification strategy is single and core information is lost are solved, the model loading efficiency and application precision are improved, and the method is suitable for complex building scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building information model optimization, in particular to a large-scale BIM model optimization system based on intelligent loading. BACKGROUND

[0002] Building information model optimization is an important technology. With the increase of building volume and the improvement of complexity, efficient loading and accurate application of large-scale BIM models become the key, and the lightweight optimization technology is of great significance to reduce hardware resource consumption and improve collaboration efficiency. The traditional model processing method relying on unified simplification standard has been difficult to meet the differentiated needs of different functional components.

[0003] However, the traditional large-scale BIM model optimization technology has the core problems of single simplification strategy and core information loss. The existing scheme uses the same simplification algorithm for all components without distinguishing the functional differences of decorative components, non-load-bearing structures and load-bearing core components. Over-simplification of decorative components will lose visual details and affect the visualization effect. Insufficient simplification of non-load-bearing structures will lead to model redundancy and slow loading speed. Mis-simplification of load-bearing components will damage the mechanical topological characteristics and endanger the accuracy of structural analysis. At the same time, the simplification process lacks the cooperation of geometric shape and semantic attribute verification. When the component curvature mutation area or high-strength material component is blindly simplified, not only the model accuracy is reduced, but also the subsequent modification cost is increased. Even the core structure data distortion may cause engineering risks, making it difficult to realize efficient and lightweight large-scale BIM models under the premise of ensuring the integrity of key information, and restricting the deep application of BIM technology in complex building scenarios. In order to solve this technical problem, we provide a large-scale BIM model optimization system based on intelligent loading. SUMMARY

[0004] The purpose of the present application is to provide a large-scale BIM model optimization system based on intelligent loading to solve the problems raised in the background art.

[0005] To achieve the above purpose, a large-scale BIM model optimization system based on intelligent loading is provided, which includes a model analysis unit for reading an IFC format file of a BIM model, separating the triangular mesh of each component, component type code and material strength parameter in the IFC format file, characterized in that it further comprises: A function recognition unit inputs the component data output by the model analysis unit into a convolutional neural network model, which includes a parallel geometric analysis channel and a semantic analysis channel, for calculating and recognizing the component data, and outputting a first component set, a second component set and a third component set; The simplification execution unit performs differential processing on the first component set, the second component set and the third component set respectively: performing edge collapse algorithm on the components in the first type of component set, simplifying to a face density that meets the visual recognition threshold of decorative components; performing vertex clustering algorithm on the components in the second type of component set, retaining mechanical topological features and meeting the simplification standard of non-load-bearing structure; skipping simplification processing for the components in the third type of component set; The model reorganization unit reorganizes the component data processed by the simplification execution unit into a lightweight BIM model.

[0006] Compared with the prior art, the beneficial effects of the present application are: In a large-scale BIM model optimization system based on intelligent loading, in the function recognition link, with the help of geometric analysis and semantic analysis dual-channel cooperation, three types of component sets are accurately divided, the geometric analysis channel locates the visual sensitive area through curvature, face density and other features, the semantic analysis channel extracts the structure attribute combined with the material strength and type code, the dynamic weight distribution module adjusts the feature priority according to the scene demand, ensures the classification accuracy of decorative components, non-load-bearing structure and load-bearing components, provides reliable basis for differentiated simplification, avoids the simplification deviation caused by traditional single classification, in the simplification execution link, customized processing is implemented for different sets, for the first type of decorative component, edge collapse algorithm is used to reduce the face in the flat area, the folding step is dynamically adjusted combined with the contour feature comparison, the visual sensitive area details are retained while reducing the face density, which not only guarantees the completeness of the decorative effect, but also reduces the redundant data, for the second type of non-load-bearing structure, a dynamic protection zone is built around the key node of the force transmission path, vertex clustering is performed outside the protection zone, and the stiffness loss is checked by cooperating with finite element micro-simulation, which ensures that the mechanical topological features are maintained after simplification, balances the lightweight and structural stability, for the third type of load-bearing component, the simplification is skipped through digital watermark marking, the original triangular mesh and material parameters are completely retained, and the accuracy of structural analysis is ensured, the model reorganization unit binds the differentiated metadata tags, constructs a dynamic octree index and matches the LOD level, realizes intelligent loading of the model, only loads the component data required for the current scene when calling, greatly improves the loading efficiency, and supports reverse restoration of the original data of the load-bearing component through the watermark interface, meets the high-precision application demand. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The overall block diagram of the present application.

[0008] The meanings of various labels in the figure are: 1, model analysis unit; 2, function recognition unit; 3, simplification execution unit; 4, model reorganization unit. DETAILED DESCRIPTION

[0009] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0010] The present application provides a large-scale BIM model optimization system based on intelligent loading, please refer to Figure 1 As shown in the figure, it comprises a model analysis unit 1 for reading the IFC format file of the BIM model, separating the triangular mesh, component type code and material strength parameters of each component in the IFC format file, characterized in that it further comprises: A function recognition unit 2 inputs the component data output by the model analysis unit 1 into a convolutional neural network model, which contains a parallel geometric analysis channel and a semantic analysis channel, for calculating and recognizing the component data, and outputs a first component set, a second component set and a third component set; In order to accurately distinguish the components with different functions in the BIM model and provide a basis for subsequent differential simplification, the function recognition unit 2 realizes the generation of the three sets through the following steps, and the specific implementation is as follows: In the geometric analysis channel, the face density distribution data of the triangular mesh of the component is calculated, the area with a density change rate exceeding a preset threshold value in the face density distribution data is identified through a density gradient detection algorithm, the geometric attribute set of the area is extracted, the geometric attribute set is quantized into a geometric feature vector, and the geometric attribute set includes the area ratio, the number of curvature extreme points and the boundary contour complexity; The geometric analysis channel focuses on the spatial form feature of the component, quantizes the geometric attributes of the triangular mesh, and generates a geometric feature vector reflecting the form complexity of the component, and the specific steps are as follows: For each component triangular mesh output by the model analysis unit 1, the number of triangular facets in a unit area, i.e. the face density, is counted, and the distribution of the density on the surface of the component is recorded, such as the edge area face density of a certain wall component is 50 / m2, and the middle area is 20 / m2. The density gradient detection algorithm is used to analyze the density distribution data, and when the density change rate of adjacent areas exceeds a preset threshold value, the area is marked as a "form sensitive area", for example, the connecting part of the column and the beam, the face density increases from 30 / m2 to 60 / m2, the change rate is 100%, far exceeding the threshold value, and is marked as a form sensitive area. For the marked form sensitive area, three core geometric attributes are extracted: Area ratio: the ratio of the area of the region to the total area of the component, such as the connecting part accounts for 15% of the total area of the column; Number of curvature extreme points: the number of vertices with maximum or minimum curvature in the region, such as the bending vertices of the curved surface component, the more the number, the more complex the form; Boundary contour complexity: measured by the number of turns of the contour line, such as a straight line contour complexity is 1, and a contour with 3 corners has a complexity of 4.

[0011] Convert the three attributes into numerical values according to the preset rules, such as area ratio 15% corresponds to 0.15, curvature extreme point 5 corresponds to 0.5, and contour complexity 4 corresponds to 0.4, and combine to form a geometric feature vector (such as [0.15, 0.5, 0.4]).

[0012] In the semantic analysis channel, the structural attribute identifier of the component type code is analyzed, the material strength parameter is extracted according to the structural attribute identifier, the material strength parameter is mapped to the safety level coefficient according to the building specification, and the semantic feature vector is generated combined with the structural attribute identifier; The semantic analysis channel extracts functional information from the type code and material parameters of the component, and generates a semantic feature vector reflecting the structural attributes of the component. The specific steps are as follows: The component type code contains structured information reflecting the function, such as the code "C-02-03", "C" represents a load-bearing component, and "02" represents a concrete type. By analyzing the identifier in the code, the basic function category of the component is determined, such as load-bearing, decoration, and non-load-bearing. According to the structural attribute identifier, the material strength parameter is located, such as the compressive strength of concrete cube and the yield strength of steel. Then, according to the building specification, the parameter is mapped to the safety level coefficient: the higher the strength, the larger the safety level coefficient (such as strength 30MPa corresponding to coefficient 0.8, 50MPa corresponding to coefficient 1.2). The function category (such as load-bearing component marked as 1, decoration component marked as 0) corresponding to the structural attribute identifier and the safety level coefficient are combined to form a semantic feature vector (such as load-bearing component with strength coefficient 1.2, vector [1, 1.2]).

[0013] The geometric feature vector and the semantic feature vector are input into the dynamic weight distribution module of the convolutional neural network model, and the component belonging set is determined through priority evaluation. The specific steps are as follows: Based on the priority allocation result of the dynamic weight distribution module, the first, second and third component sets are output, and the module automatically adjusts the weights of geometric and semantic features according to the application scenario of the component. If the model is used for visual display, the weight of geometric features is increased (such as 0.6); if it is used for structural analysis, the weight of semantic features is increased (such as 0.7). For example, for decorative components, the weight of geometric features (form complexity) is 0.6, and the weight of semantic features (safety level) is 0.4. For load-bearing components, the weight of semantic features is 0.7, and the weight of geometric features is 0.3. The comprehensive score of the feature vector is calculated according to the weight, and the set is divided according to the score result: The first type of set: the components with high geometric feature scores (e.g., the proportion of morphologically sensitive regions is more than 50%) and low semantic feature scores (e.g., the safety level coefficient is less than 0.5), which are mostly decorative components (e.g., reliefs, patterns); The second type of set: the components with medium geometric feature scores (the proportion of morphologically sensitive regions is 20%-50%) and medium semantic feature scores (the safety level coefficient is 0.5-0.8), which are mostly non-load-bearing structures (e.g., partitions, secondary supports); The third type of set: the components with high semantic feature scores (the safety level coefficient is greater than 0.8), which are mostly load-bearing core components (e.g., main beams, columns). The final three-component set accurately reflects the simplification needs of different components, laying the foundation for subsequent differentiated optimization.

[0014] To accurately select decorative components with visual presentation as the core function, the generation of the first type of set is based on the coordination of geometric analysis and semantic analysis channels, and is realized through double verification of visual sensitive region identification and functional attribute filtering. The specific implementation is as follows: Based on the face density distribution data output by the geometric analysis channel, the visual key area of the component is further located through curvature feature, and the steps are as follows: The characteristic surface partitions of the component are divided by the curvature clustering algorithm, and the regions with similar curvature values are classified into the same characteristic surface (e.g., the curvature of the planar region is close to 0, and the curvature of the curved surface region is in the range of 5-10). For example, the surface of a column head component includes a planar base (curvature 0), an arc-shaped side (curvature 8), and a top carving (curvature 20-30). By clustering algorithm, it is divided into 3 characteristic surface partitions. A preset change rate threshold is set (e.g., the curvature change rate of adjacent regions exceeds 50%). The regions with a curvature change rate exceeding the preset change rate threshold are marked as visual sensitive regions. Such regions are usually visual focal points (e.g., the junction between the carving and the arc-shaped side, the curvature increases from 8 to 20, with a change rate of 150%, far exceeding the threshold). For example, the top carving area of the column head component and its junction with the side are marked as visual sensitive regions. The total area of all visual sensitive regions is calculated, and the proportion compared with the overall surface area of the component is obtained (e.g., the area proportion of the visual sensitive region of the column head component is 40%). At the same time, the text matching rule is activated in the semantic analysis channel, the component type code containing the preset keyword is shielded, and the component with a material strength higher than the decorative material standard value is filtered. The components with a visual sensitive region area proportion exceeding the preset visual sensitive threshold and passing the filtering are selected and listed in the first type of set, and the steps are as follows: The preset keyword list (such as "beam", "column", "load-bearing", representing load-bearing or structural functions) is checked by text matching to check the component type code. If the code contains these keywords, it is directly excluded (such as the code "Z-01-Load-bearing" is shielded because it contains the keyword "load-bearing"), the standard value of the decorative material is set, the material strength parameter corresponding to the component type code is extracted, and if the parameter is higher than the standard value, it is determined as a non-decorative component and filtered. The results of geometric analysis and semantic analysis are combined, and finally the components that meet the conditions are screened out and listed in the first set: Only components that meet the following two conditions are retained: The visual sensitive area ratio exceeds the preset visual sensitive threshold (such as 30%, that is, the visual sensitive area needs to account for more than 30% of the total area of the component), and is filtered through the semantic analysis channel, that is, the code does not contain the preset keyword, and the material strength does not exceed the standard value of the decorative material. For example, the visual sensitive area ratio of a certain carved swallow component is 50% (exceeding the 30% threshold), the type code is "Q-02-Decoration" (not containing the keyword), and the wood strength is 8MPa (lower than the 10MPa standard value). The component is listed in the first set as a simplified object that focuses on visual retention.

[0015] In order to accurately select non-load-bearing structural components suitable for using vertex clustering algorithm for simplification, the generation of the second set combines material strength filtering in the semantic analysis channel with curvature feature recognition in the geometric analysis channel, ensuring that the selected components have no load-bearing function and have the feasibility of shape simplification. The specific implementation is as follows: In the semantic analysis channel, for components whose component type code is explicitly non-load-bearing structural category (such as codes containing "partition wall", "filling", "secondary support" identifiers), a dynamic filtering mechanism for material strength parameters is established, and the steps are as follows: From the attribute data associated with the component type code, the concrete strength grade and steel yield strength parameters are extracted, which directly reflect the structural bearing potential of the component. Non-bearing structures usually do not require high strength, and according to the material requirements for non-bearing structures in the building code, the allowable value is preset. The setting logic of the allowable value is that non-bearing structures only need to meet their own weight and slight additional load, and do not need to reach the strength standard of bearing structures. When the component type code belongs to the non-bearing structure category, the concrete strength grade and steel yield strength parameters are extracted and compared with the preset allowable value of non-bearing structures. The components with strength values lower than the standard values are filtered out. For example, a certain non-bearing partition wall has a concrete strength grade of C20 (lower than C25) and a steel yield strength of 210 MPa (lower than 235 MPa), so it is filtered out. If another non-bearing component has a concrete strength grade of C30 (higher than C25), it is excluded due to excessive strength. At the same time, the overall curvature distribution of the component is calculated in the geometric analysis channel, and the components with smooth curvature changes and no sudden characteristics are marked. The components that meet both of the above requirements are listed in the second set, and at the same time, the components that pass the semantic filtering are analyzed for morphological characteristics in the geometric analysis channel to determine whether they are suitable for simplification. The steps are as follows: Based on the triangular mesh data of the component, the curvature value of each vertex is calculated (e.g. the curvature of a planar vertex is 0, and the curvature of a curved surface vertex is 5-10), and the distribution of the curvature value on the surface of the component is counted. The curvature change rate is detected by a sliding window algorithm. If the curvature change rate of any adjacent region on the surface of the component is lower than the preset mutation threshold, and there is no curvature extreme point (a point where the curvature value suddenly increases to 20 or more), it is determined that the curvature changes smoothly and has no sudden characteristics. For example, a certain non-bearing partition wall is a planar structure with a curvature value of 0 and a change rate of 0, which meets the characteristics. However, a partition wall with complex corners does not meet the characteristics because of the sudden change in curvature (e.g. the curvature at the corner increases from 0 to 30). The components that meet the "smooth curvature change and no sudden characteristics" are marked, indicating that their morphology is simple and they are not likely to lose key features during simplification. The low-strength non-bearing components filtered by the semantic analysis channel are compared with the components with smooth curvature marked by the geometric analysis channel, and only the components that meet both conditions are retained and listed in the second set. For example, a certain non-bearing infill wall: When the semantic analysis shows that the concrete strength grade of the component is C20 (in line with the non-load-bearing strength standard) and the geometric analysis shows that the surface of the component is a plane with a curvature change rate of 0 (in line with the flat feature), the component is included in the second type set as an object of subsequent vertex clustering simplification. The semantic analysis channel ensures the adaptability of the non-load-bearing property and the material strength of the component, and the geometric analysis channel ensures the simplification feasibility of the component form. The second type set generated by the cooperation of the two channels accurately locks the component suitable for simplification by using the vertex clustering algorithm, avoids over-simplification of high-strength non-load-bearing components, and prevents distortion caused by simplification of components with complex forms, thereby providing a reliable basis for subsequent differentiated processing.

[0016] The simplification execution unit 3 performs differentiated processing on the first type component set, the second type component set, and the third type component set, respectively. The edge folding algorithm is performed on the components in the first type component set, and the components are simplified to a face density that meets the visual recognition threshold of the decorative component. In order to realize model lightweighting under the premise of ensuring the visual recognition degree of the decorative component, when the edge folding algorithm is performed on the components in the first type component set, the simplifiable area is located through curvature analysis, and the folding strategy is dynamically adjusted in combination with contour comparison. The specific implementation is as follows: The curvature distribution data of the component surface is calculated, the flat area with a curvature value below the visual perception threshold is identified, and based on the triangular mesh model of the component, the vertices of each triangular facet are traversed to calculate the curvature value at the vertex, which reflects the degree of curvature of the curved surface. The smaller the value, the closer it is to a plane. Set the visual perception threshold (such as curvature value ≤ 5, the human eye cannot perceive the slight curvature of the area), mark all areas with a curvature value below the threshold as flat areas, and these areas are the main objects of edge folding simplification due to their simple shape. For example, the background plane part of a certain relief component has a curvature value of 2 (below the threshold of 5), which is identified as a flat area; while the relief pattern itself has a curvature value of 15, which belongs to the visual sensitive area that cannot be simplified, then the candidate triangular facets for edge folding are located based on the flat area, and in the marked flat area, triangular facets with larger area and a distance from the visual sensitive area boundary exceeding a predetermined safety distance (such as 5 cm, to avoid the folding affecting the shape of the sensitive area) are selected as candidates. For example, in the background plane of the above-mentioned relief component, triangular facets far from the relief pattern are selected as candidates, ensuring that the folding operation only affects areas that do not affect the visual effect, then a real-time generation of the component's front projection contour map is performed after each folding operation, and the current contour map and the original contour map are input into a convolutional neural network model for skeleton feature comparison. The edge folding operation is to merge the common edge of two adjacent triangular facets, reducing the number of facets (such as merging two adjacent triangles into a quadrilateral, reducing 1 facet). After each folding, a two-dimensional contour map is generated from the front of the component, and the skeleton features of the contour (such as the turning points and line directions of the contour) are extracted and compared with the original contour map skeleton features before folding. For example, after folding a triangular facet in a flat area, the coincidence degree of the skeleton lines of the generated contour map and the original map needs to be kept at a high level, and when the feature similarity is lower than the similarity threshold, a folding step adaptive reduction mechanism is triggered, which reduces the number of triangular facets processed each time. For example, from 5 to 3, to avoid contour deformation due to excessive folding. For example, after a certain folding, the similarity drops to 75% (lower than 80%), so the next folding only processes 3 facets, making the contour change more gently. Finally, the above folding and comparison process is iterated until the facet density of the component meets the visual recognition threshold of the decorative component. The visual recognition threshold refers to the minimum facet density at which the human eye can clearly recognize the decorative details (such as 30 facets per square meter), and when the facet density of the folded component reaches the threshold, the operation is stopped. For example, a certain component has an initial facet density of 100 per square meter, which is reduced to 30 after multiple foldings, and the contour skeleton similarity always remains above 80%, indicating that the simplification is complete.

[0017] To avoid excessive edge folding operations causing the contour of the first type of component (decorative component) to distort, the folding step adaptive reduction mechanism adjusts the folding scale by calculating a compensation coefficient in real time, and dynamically updates the simplified area in combination with curvature re-detection. The specific implementation is as follows: The folding compensation coefficient is calculated according to the feature similarity difference value between the current contour map and the original contour map, and the feature similarity difference value refers to the difference degree of the original contour skeleton feature and the current folded contour skeleton feature (for example, the similarity decreases from 90% to 70%, and the difference value is 20%). The compensation coefficient is positively correlated with the difference value: the larger the difference value, the smaller the compensation coefficient; the smaller the difference value, the larger the compensation coefficient. For example, when the similarity difference value is 10%, the compensation coefficient is set to 0.8; when the difference value is 30%, the compensation coefficient decreases to 0.3. The area ratio of the total area of the triangular patches folded by the current edge folding operation to the total area of the flat region of the component is extracted, that is, the area ratio, for example, 5 triangular patches are currently folded, the total area is 2 square meters, and the total area of the flat region of the component is 20 square meters, the area ratio is 10%, and the area ratio is multiplied by the folding compensation coefficient to obtain the next folding step, that is, the triangular patch area ratio that can be processed by the next folding operation, for example, the area ratio of 10% multiplied by the compensation coefficient of 0.8, the next folding step is 8%, that is, the total area of the folded patch is not more than 8% of the flat region. In this way, the folding step is dynamically adjusted according to the contour distortion degree: the more the contour similarity decreases, the smaller the step, and the further deformation of the contour is avoided. Multiply the area ratio by the folding compensation coefficient to obtain the next folding step, set a preset period and a preset folding compensation coefficient threshold, and if the compensation coefficient of each folding in the period is less than the threshold, it indicates that the current flat region folding is close to the boundary of the visual sensitive area, and continuing to fold the original region may affect the decorative details. At this time, the curvature re-detection module is activated, the curvature distribution data of the component surface is recalculated, and the flat region distribution map is updated: The over-folded region is removed, and a new flat region is marked, such as the region that has not been folded before and the curvature is still below the visual perception threshold. After 5 foldings, the compensation coefficient continues to be low, and the curvature re-detection finds that the edge of the original flat region has reached the relief pattern (visual sensitive area). The updated flat region distribution map excludes the edge part and only keeps the area away from the sensitive area, ensuring that subsequent folding operations will not damage the decorative details. The folding step adaptive reduction mechanism dynamically controls the folding amplitude through the product of the compensation coefficient and the area ratio, avoids contour distortion, and updates the simplified region in real time through curvature re-detection to ensure that the folding is always within a safe range. Finally, while efficiently reducing the number of patches, the visual features of the decorative component are maximized.

[0018] The vertex clustering algorithm is performed on the components in the second component set to retain the mechanical topological features and meet the simplified standard of non-load-bearing structure; In order to simplify the non-load-bearing structure components while retaining their mechanical topological features, the vertex clustering algorithm is performed on the second component set through the cooperative operation of key node protection and clustering verification to achieve safe and efficient simplification. The specific implementation is as follows: The component mechanical topology features output by the loading function recognition unit 2 include the force transmission path of the component, such as the main route of force transmission from one end of the component to the other end and the key point information of force bearing, and the key node coordinate set in the force transmission path is recognized according to the component mechanical topology features. These nodes are the turning points or concentration points of force transmission, such as the connection points of non-load-bearing partitions and frames, the end points of secondary supports, and are crucial to maintaining the basic mechanical properties of the component. For example, the force transmission path of a certain non-load-bearing partition is from the top two ends to the bottom support, and the top two end points and the bottom support point are marked as key nodes, and the coordinates are recorded as the key node coordinate set. A protection zone with a dynamically expanding radius is constructed around each node in the key node coordinate set. The protection zone is a three-dimensional area around the node, which is used to prohibit vertex clustering operations to avoid the loss of mechanical features caused by the merging of key nodes. The radius of the protection zone adopts a dynamic expansion mechanism: the initial radius is set according to the force of the node, such as 10 centimeters for nodes with large forces and 5 centimeters for nodes with small forces. If the subsequent clustering operation is close to the edge of the protection zone (such as less than 2 centimeters), the radius is automatically expanded (such as from 10 centimeters to 12 centimeters), ensuring that the mechanical topology structure around the key node is not damaged. For example, the initial protection zone radius of a certain key node is 10 centimeters, and when the clustering operation involves a vertex that is 9 centimeters away, the radius is expanded to 12 centimeters to prevent the vertex from being merged to affect the node function. The vertex clustering algorithm is executed in the outer area of the protection zone, and adjacent vertices that meet the non-load-bearing component simplification standard are merged to reduce the number of vertices, such as merging an area with originally 100 vertices into 50, while preserving the overall shape of the component. After each clustering, the change rate of the component stiffness matrix is verified by finite element micro-simulation. The stiffness matrix reflects the ability of the component to resist deformation, and the smaller the change rate, the smaller the impact of clustering on the mechanical properties. For example, the stiffness matrix change rate after clustering is 5%, indicating that the simplification has little effect on the stiffness of the component; if the change rate reaches 15%, it may affect the stability of the non-load-bearing structure. When the stiffness loss exceeds the pre-set stiffness loss threshold, the vertex clustering operation is rolled back to ensure that the mechanical properties of the component meet the non-load-bearing structure simplification standard, i.e., the simplified component can still bear its own weight and slight additional loads. After a certain clustering, the stiffness loss reaches 12% (exceeding the 10% threshold), and the clustering is cancelled, the merged vertices are restored, and vertices that are farther apart are selected for clustering to avoid structural risks caused by excessive simplification.

[0019] In order to avoid the non-load-bearing component from being insufficient in stiffness due to excessive simplification during the vertex clustering simplification process, when the finite element micro-simulation detects that the stiffness loss exceeds the pre-set threshold, precise positioning of the abnormal area is required, the clustering operation is rolled back, and verification is strengthened to ensure that the mechanical properties of the component meet the standard. The specific implementation is as follows: After the finite element micro-simulation detects that the stiffness loss exceeds the limit, the spatial coordinates of the stiffness matrix change rate anomaly of the component are extracted, the coordinates are matched with the vertex clustering operation log, and the affected mesh area of the last three clustering operations is located; Based on the spatial distribution characteristics of the affected mesh area abnormal coordinates, the protection zone range is expanded along the force transmission path direction of the component, and the last three vertex clustering operations acting on the abnormal area are revoked, while the protection zone expansion operation is performed, the mesh area after rollback and the expanded protection zone range are input into the vertex clustering algorithm for re-simplification, and the re-simplified component is started for intensive finite element micro-simulation verification.

[0020] In the finite element micro-simulation, when the component stiffness loss exceeds the preset threshold, first extract the spatial coordinates of the stiffness matrix change rate anomaly, i.e. the specific location of the component where the stiffness has decreased significantly, such as the coordinates of the middle area of a non-load-bearing partition wall, then call the vertex clustering operation log, which records the vertex coordinates involved in each clustering operation, the information of the merging time, match the abnormal coordinates with the clustering area in the log, and locate the mesh area affected by the abnormal area in the last three clustering operations, such as the last three clustering operations involving the vertices of the middle area of the partition wall, the stiffness of the middle area of the partition wall is abnormal, and the log matching finds that the last three clustering operations have merged the adjacent vertices of the area, it is determined that these clustering is the cause of the stiffness loss, for the located affected mesh area, expand the key node protection zone range along the force transmission path direction of the component: the force transmission path is the main route of force transmission in the component, such as the force of the partition wall from the top to the bottom support, when expanding the protection zone, the key node surrounding area on the force transmission path needs to be covered, the protection zone radius of the partition wall top node is expanded from 10 cm to 15 cm, to ensure that the mesh area on the force transmission path is not over-clustered, at the same time, revoke the last three vertex clustering operations acting on the abnormal area, i.e. split the merged vertices back to the state before clustering, restore the area merged into 50 vertices to 100 vertices, for example, the middle area of a partition wall has 50% less vertices due to three clustering, after rollback, the number of vertices in the area is restored, the stiffness is partially restored, and the mesh area after rollback and the expanded protection zone range are input into the vertex clustering algorithm for re-simplification operation: In the area outside the new protective zone, vertices that are farther apart are selected for clustering, such as merging only vertices that are more than 5 cm apart to avoid excessively dense merging. The number of vertices in each cluster is reduced, for example, from 10 vertices to 5 vertices each time. After simplification, enhanced finite element micro-simulation verification is initiated. Compared with conventional verification, enhanced verification increases the load test scenario, such as applying 1.2 times the self-weight load to the surface of the component, and extends the simulation time to ensure that the simplified component can still meet the stiffness requirements under more stringent conditions. For example, after a partition wall is re-clustered, the enhanced verification confirms that the stiffness loss is 7%, which meets the simplification standard for non-load-bearing structures. The simplification is completed. Finally, while reducing the number of vertices, the mechanical topological characteristics of non-load-bearing components are not destroyed, meeting the dual requirements of lightweighting and structural stability.

[0021] Simplification processing is skipped for components in the third type of component set; To ensure the integrity of the original data of the core load-bearing components in the third set and to avoid the impact of simplification on structural safety, a method of embedding digital watermark identifiers is used to skip simplification for these components. The specific implementation method is as follows: While the model parsing unit 1 reads the IFC format file, special processing is simultaneously applied to components that have been marked as the third type of set by the function identification unit 2: A hash value is generated from the component's type code, a fixed-length character sequence that uniquely identifies the code. For example, "ZL-01-load-bearing" is converted into a specific character combination. At the same time, the component's material strength parameters are extracted. The hash value and material strength parameters are concatenated in a fixed format (such as the string form "hash value + strength parameter") to form a digital watermark identifier. This identifier is unique; different components have different type codes or strength parameters, resulting in different generated identifiers. The generated digital watermark identifier is secretly embedded in the component's triangular mesh data, such as embedding it into the decimal places of the mesh vertex coordinates, without affecting normal data reading. This completes the marking of the third type of component. For example, if the type code hash value of a main beam component is "ABC123" and the strength parameter is "50MPa", the identifier "ABC123+50MPa" is generated and embedded in its mesh data, marking it as a component that needs to be skipped for simplification.

[0022] When the simplified execution unit 3 processes component data, it performs real-time detection of the digital watermark identifier for each component: The presence state of the digital watermark identifier is detected in real time, the watermark identifier that may be hidden in the component data is extracted through a special algorithm, and whether the format is correct is verified, such as whether it meets the splicing rule of "hash value + strength parameter", whether the content matches the component type code and material parameter, whether the strength parameter in the identifier is consistent with the parameter extracted by the analysis unit, and if a valid watermark identifier is detected, the skipping program is triggered, and any simplification operation of the component is stopped, such as edge folding, vertex clustering, etc., to ensure that the original triangular mesh data of the component, such as vertex coordinates and the number of facets, remains unchanged. For example, a certain column component is detected to have a valid watermark, and the simplification execution unit 3 directly skips the processing, and the original 1000 triangular facet data is completely retained. After the skipping program is triggered, the original triangular mesh data of the skipped component is directly transmitted to the model reorganization unit 4, and the data is not modified during transmission. At the same time, the system automatically records the component information that is skipped, and records the total number of components that are skipped and the spatial coordinates, which are used for verification in subsequent model reorganization, such as confirming whether all the third type of components are loaded according to the original data, and model version tracing, such as checking the number and position of the third type of components in a certain version of the model.

[0023] The model reorganization unit 4 reorganizes the component data processed by the simplification execution unit 3 into a lightweight BIM model.

[0024] In order to integrate the three types of components that have been processed differently into a complete lightweight BIM model, while retaining the processing characteristics and fast access capability of each component, the model reorganization unit 4 realizes reorganization through the process of metadata binding, spatial index construction and hierarchical matching, and the specific implementation is as follows: The model reorganization unit 4 first receives the three types of component sets output by the simplification execution unit 3, and binds different metadata tags to each component in the three types of component sets. The content of the different metadata tags is the processing characteristics of the functional identification unit 2 and the simplification execution unit 3. The processing characteristics of the functional identification unit 2 include: the component set category (first type / second type / third type), the visual sensitive area ratio (for the first type), the mechanical topological feature keywords (for the second type and the third type), the material strength parameter level, etc. The processing characteristics of the simplification execution unit 3 include: the edge folding number and the final facet density of the first type of component, the change in the number of vertices before and after vertex clustering and the stiffness loss rate of the second type of component, the "unsimplified" identification and the original facet number of the third type of component. For example, the metadata tag of a certain first type of carved ornament component is: "set category: first type; visual sensitive area ratio: 40%; edge folding number: 8 times; final facet density: 30 per square meter". The tag of a certain third type of main beam component is: "set category: third type; mechanical topological feature: main force transmission path; processing state: unsimplified; original facet number: 1000". These tags are like the "identity file" of the component, which not only records the processing process, but also provides a feature basis for subsequent model application; When constructing the lightweight packaging structure, assign the components to the leaf nodes by recursive space division, and attach the precision control tags in the nodes. Automatically match the LOD levels according to the octree leaf node tags, and finally output the lightweight BIM model carrying the octree index, data block and precision rule configuration file, which supports reverse restoration of the original data of the load-bearing components through the watermark interface.

[0025] Based on the component space coordinates, construct a dynamic octree index. The octree index is a structure that recursively divides a three-dimensional space into eight sub-regions. The specific steps are as follows: Take the spatial range of the entire model as the root node, and divide it uniformly into eight cubic sub-regions. Check the number of components in each sub-region. If the number of components in a sub-region exceeds the preset threshold, continue to divide the sub-region into eight smaller sub-regions until the number of components in each sub-region (leaf node) does not exceed the threshold. For example, the root node range of a certain building model is X=0-100, Y=0-100, Z=0-50. After three recursive divisions, each leaf node corresponds to a spatial range of 5x5x3 meters, containing 10-15 components. According to the spatial coordinates of each component, assign it to the corresponding leaf node. For example, a partition wall with coordinate range X=10-15, Y=20-25, Z=0-3 is assigned to the leaf node. Add precision control tags in each leaf node. The tag content is the simplification precision level of the components in the node. For example, the first type of component corresponds to "visual precision level", the second type corresponds to "non-load-bearing precision level", and the third type corresponds to "original precision level". For example, a leaf node containing carved components is marked as "visual precision level", a node containing a partition wall is marked as "non-load-bearing precision level", and a node containing a main beam is marked as "original precision level". The role of the dynamic octree index is to realize fast retrieval and loading of the model. For example, when viewing a floor, only the components of the corresponding leaf node of the floor are loaded. The precision control tags provide a basis for model invocation in different scenarios. Based on the dynamic octree index, construct the packaging structure of the lightweight BIM model. The specific steps are as follows: According to the accuracy control mark of the leaf node, the corresponding LOD level is automatically matched, the "original accuracy level" corresponds to LOD300 (high accuracy), the "non-load-bearing accuracy level" corresponds to LOD200 (medium accuracy), and the "visual accuracy level" corresponds to LOD100 (basic accuracy). For example, the main beam is matched with LOD300, the partition wall is matched with LOD200, and the carving is matched with LOD100. The component data (including geometric data and metadata tags) in each leaf node is packaged as an independent data block, the data block is compressed by using a compression algorithm to reduce the volume, but the key information is retained, the original triangular mesh data of the third type of component is not compressed, an accuracy rule configuration file is created, the accuracy standards corresponding to each LOD level, the storage path of the data block and the structure information of the octree index are recorded, and finally the lightweight BIM model is output, which includes three parts: a dynamic octree index, a compressed component data block and an accuracy rule configuration file. At the same time, the model supports reverse restoration of the original data of the third type of component through a watermark interface. When high-precision load-bearing component data is needed, the original triangular mesh data not simplified is extracted from the data block by reading the watermark identifier in the metadata tag, so that the accuracy of the structural analysis scene is ensured, the storage and running costs of the large-volume BIM model are significantly reduced while the core accuracy is ensured, and the application demand of intelligent loading is met.

[0026] In the present application, the model analysis unit 1 reads the IFC file and separates the component data, the function recognition unit 2 analyzes through the geometric and semantic double channels and outputs the decoration component (the first type), the non-load-bearing structure (the second type) and the load-bearing component (the third type) set, the simplification execution unit 3 performs edge folding on the first type, performs vertex clustering on the second type and skips simplification on the third type, the model reorganization unit 4 binds metadata and constructs a dynamic octree index to generate a lightweight model, solves the problems of single traditional simplification strategy and core information loss, improves the model loading efficiency and application accuracy, and is suitable for complex building scenes.

[0027] The basic principles, main features and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A large-scale BIM model optimization system based on intelligent loading, comprising a model analysis unit (1) for reading an IFC format file of a BIM model, separating the triangular mesh of each component, component type code and material strength parameters in the IFC format file, characterized in that, Also comprising: a function recognition unit (2) that inputs the component data output by the model analysis unit (1) into a convolutional neural network model containing a geometric analysis channel and a semantic analysis channel in parallel, for computational recognition of the component data, outputting a first set of components, a second set of components, and a third set of components; a simplification execution unit (3) that performs differential processing on the first set of components, the second set of components, and the third set of components, respectively: performing edge folding algorithm on the components in the first set of components to simplify to a face density that meets the visual recognition threshold of decorative components; performing vertex clustering algorithm on the components in the second set of components to retain the mechanical topological features and meet the non-load-bearing structure simplification standard; skipping simplification processing for the components in the third set of components; a model reorganization unit (4) that reorganizes the component data processed by the simplification execution unit (3) into a lightweight BIM model.

2. The large-scale BIM model optimization system based on intelligent loading according to claim 1, wherein, The function recognition unit (2) realizes the generation of the three sets by the following steps: In the geometric analysis channel, calculate the face density distribution data of the triangular mesh of the component, identify the area where the density change rate exceeds the preset threshold in the face density distribution data through the density gradient detection algorithm, extract the geometric attribute set of the area, quantify the geometric attribute set into a geometric feature vector, and the geometric attribute set includes area proportion, curvature extreme point number, and boundary contour complexity; In the semantic analysis channel, analyze the structural attribute identifier of the component type code, extract the material strength parameter according to the structural attribute identifier, map the material strength parameter to a safety level coefficient according to the building code, and generate a semantic feature vector combining the structural attribute identifier; input the geometric feature vector and the semantic feature vector into the dynamic weight distribution module of the convolutional neural network model, and output the first set of components, the second set of components, and the third set of components based on the priority allocation result of the dynamic weight distribution module.

3. The large-scale BIM model optimization system based on intelligent loading of claim 2, wherein, The generation of the first set further comprises: based on the face density distribution data output by the geometric analysis channel, divide the feature surface partition of the component through the curvature clustering algorithm, mark the area where the curvature change rate exceeds the preset change rate threshold as the visual sensitive area, activate the text matching rule in the semantic analysis channel at the same time, shield the component type code containing the preset keyword, and filter the component whose material strength is higher than the standard value of decorative material, screen out the components whose visual sensitive area proportion exceeds the preset visual sensitive threshold and pass the filtering, and list them in the first set.

4. The large-scale BIM model optimization system based on intelligent loading of claim 2, wherein, The generation of the second set further comprises: establish a dynamic filtering mechanism for material strength parameters in the semantic analysis channel, when the component type code belongs to the non-load-bearing structure category, extract the concrete strength grade and steel yield strength parameter, and compare them with the preset non-load-bearing structure allowed value, screen out the components whose strength value is lower than the standard value, and at the same time, calculate the overall curvature distribution of the component in the geometric analysis channel, mark the components with flat curvature change and no mutation characteristics, and list the components that meet the above two requirements in the second set.

5. The large-scale BIM model optimization system based on intelligent loading of claim 1, wherein, Performing edge folding algorithm on the components in the first set of components, specifically as follows: The method comprises the following steps: calculating the surface curvature distribution data of the component, identifying the flat area with a curvature value lower than a visual perception threshold, locating the candidate triangular facets for performing edge folding based on the flat area, generating a component orthographic contour map in real time after each folding operation, inputting the current contour map and the original contour map into a convolutional neural network model for skeleton feature comparison, triggering a folding step adaptive reduction mechanism when the feature similarity is lower than a similarity threshold, and iteratively repeating until the visual recognition threshold of the decorative component is met.

6. The large-scale BIM model optimization system based on intelligent loading of claim 5, wherein, The specific implementation of the folding step adaptive reduction mechanism is as follows: A folding compensation coefficient is calculated according to the feature similarity difference between the current contour map and the original contour map, the area proportion of the triangular facets in the current edge folding operation is extracted, and the area proportion is multiplied by the folding compensation coefficient to obtain the next folding step. When the folding compensation coefficients of the folding operations in a preset period are all less than a preset folding compensation coefficient threshold, a curvature re-detection module is activated to update the flat area distribution map.

7. The large-scale BIM model optimization system based on intelligent loading of claim 1, wherein, The vertex clustering algorithm is performed on the components in the second type component set, and the specific implementation is as follows: The component mechanical topology features output by the function recognition unit (2) are loaded, the key node coordinate set in the force transmission path is identified according to the component mechanical topology features, and a protection zone with a dynamically expanding radius is constructed with the nodes in the key node coordinate set as the center; The vertex clustering algorithm is performed in the outer region of the protection zone, adjacent vertices that meet the non-load-bearing component simplification standard are merged, and the component stiffness matrix change rate is verified through finite element micro-simulation after each clustering. When the stiffness loss exceeds a preset stiffness loss threshold, the vertex clustering operation is rolled back.

8. The large-scale BIM model optimization system based on intelligent loading of claim 7, wherein, The method for rolling back the vertex clustering operation when the stiffness loss exceeds the preset stiffness loss threshold is as follows: After the finite element micro-simulation detects that the stiffness loss is out of limit, the component space coordinates with abnormal stiffness matrix change rate are extracted, the coordinates are matched with the vertex clustering operation log, and the affected grid area of the last three clustering operations is located; Based on the spatial distribution characteristics of the abnormal coordinates of the affected grid area, the protection zone range is expanded along the force transmission path direction of the component, the last three vertex clustering operations acting on the abnormal area are cancelled, and the protection zone expansion operation is performed. The rolled-back grid area and the expanded protection zone range are input into the vertex clustering algorithm for re-simplification, and the re-simplified component is subjected to intensive finite element micro-simulation verification.

9. The large-scale BIM model optimization system based on intelligent loading of claim 1, wherein, The method for skipping the simplification process for the components in the third type component set is as follows: When the model analysis unit reads the IFC format file, a digital watermark identifier is embedded in the component marked as the third type set. The identifier is generated by splicing the hash value of the component type code and the material strength parameter. When the simplification execution unit (3) processes the component data, the existence state of the digital watermark identifier is detected in real time. If a valid watermark identifier is detected, the skipping program is triggered, the original triangular mesh data of the skipped component is directly transmitted to the model reorganization unit (4), and the total number and spatial coordinates of the skipped components are recorded.

10. The large-scale BIM model optimization system based on intelligent loading of claim 1, wherein, The model reorganization unit (4) reorganizes the lightweight BIM model through the following steps: Receiving the three types of component sets output by the simplified execution unit (3), binding a differentiated metadata tag for each component in the three types of component sets, the content of the differentiated metadata tag is the processing characteristics of the function identification unit (2) and the simplified execution unit (3); Based on the component space coordinates, a dynamic octree index is constructed, the components are allocated to the leaf nodes through recursive space division, and the precision control marks are attached in the nodes. When constructing a lightweight encapsulation structure, the LOD level is automatically matched according to the octree leaf node marks, and finally a lightweight BIM model carrying the octree index, data block and precision rule configuration file is output, which supports reverse restoration of the original data of the load-bearing component through the watermark interface.

Images