A BIM model structural design review method and system based on large model

Abstract

The present invention discloses a BIM model structural design review method and system based on a large model, which relates to the field of building information modeling technology, including BIM model data preprocessing based on an AI large model; using an AI large model's built-in rule engine to review the BIM model's structural design specifications, and automatically generating a review report based on natural language processing technology; performing BIM model collision detection based on a deep learning algorithm, identifying potential structural conflicts, combining three-dimensional visualization to display error points, and providing adjustment suggestions based on historical data and optimization algorithms. The method of the present invention improves data quality and consistency, makes key information in the BIM model clearer and more concise, improves the accuracy of the review process, enables potential problems in the structural design process to be discovered and handled in a timely manner, improves the overall intelligence level of the BIM model, and makes the architectural design process more accurate, intelligent, and efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of building information modeling, and in particular to a BIM model structural design review method and system based on a large model. Background Art

[0002] In recent years, with the widespread adoption of Building Information Modeling (BIM) technology, the construction industry has gradually evolved towards intelligent and digital development. Through three-dimensional digitalization, BIM technology enables collaborative design across multiple disciplines, including structural, mechanical, and electrical engineering, and HVAC systems, improving the design efficiency and construction quality of construction projects. However, with the increasing complexity of BIM models, the accuracy, standardization, and optimization capabilities of structural design have become crucial for project quality assurance. To ensure that BIM models comply with relevant architectural design standards and specifications, the industry has gradually developed a series of BIM model review methods, including automated review based on rule matching, collision detection based on geometric analysis, and optimization analysis based on numerical simulation. While these methods have improved the compliance and constructability of BIM models to a certain extent, they still rely on manually set rules or thresholds, making them difficult to adapt to evolving architectural design requirements. Recent developments in artificial intelligence (AI), particularly deep learning and natural language processing, have provided new insights into BIM model review. AI algorithms can automatically learn and extract design specifications, improving the automation and accuracy of review processes.

[0003] Existing BIM model review methods still have many limitations in their application. Existing methods mainly rely on rule-based review engines, whose rule bases are usually manually defined and require continuous maintenance and updating by professionals, resulting in limited coverage of review rules and difficulty in adapting to dynamic changes in building design standards. Due to the certain ambiguity in building design specifications, rule systems based on hard matching have difficulty handling complex design constraints, resulting in limited accuracy of review results and weak collision detection and structural optimization capabilities. Traditional methods usually perform structural collision detection based on geometric calculations, which can only identify explicit geometric conflicts and cannot analyze potential structural bearing capacity issues. At the same time, for detected conflict points, existing technologies mainly rely on manual adjustments and lack intelligent optimization suggestions based on historical data, which affects the optimization efficiency of BIM models. The generation method of review reports is relatively simple. Existing technologies usually use templates to output review results, lack in-depth analysis of design problems, and cannot provide readable reports in natural language, which affects designers' understanding and correction of problems. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the existing BIM model structural design review method based on large models has the problem that the rule base relies on manual maintenance, it is difficult to dynamically adapt to changes in design specifications, the collision detection capability is limited, it is impossible to perform deep structural optimization, the review report generation method is single, and there is a lack of intelligent analysis. The problem also arises as to how to combine the AI ​​large model to realize efficient data preprocessing of BIM models, intelligent rule review, and deep learning optimized automated BIM design review.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a BIM model structural design review method based on a large model, comprising preprocessing BIM model data based on an AI large model; using a built-in rule engine in the AI ​​large model to review the structural design specifications of the BIM model, and automatically generating a review report based on natural language processing technology; performing BIM model collision detection based on a deep learning algorithm, identifying potential structural conflicts, combining three-dimensional visualization to display error points, and providing adjustment suggestions based on historical data and optimization algorithms; the rule engine comprises constructing a BIM structural design knowledge graph based on Ontology semantic modeling, extracting rules from architectural design standards, and converting them into logical reasoning rules; generating a review report comprises automatically generating a BIM review report based on natural language processing technology, extracting technical parameters using a BERT pre-trained language model, generating a review report summary through a Seq2Seq structure, and screening high-risk review points using a TextRank algorithm; providing adjustment suggestions comprises combining GNN and LSTM prediction models to enable the BIM model to identify conflicts in the design, making dynamic optimization suggestions based on historical building data, and generating a 3D interactive optimization solution.

[0007] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the data preprocessing includes: using the outlier detection algorithm of the AI ​​large model, training RNN based on historical BIM design data, automatically identifying outliers in BIM structural parameters, and automatically filling them; using the Transformer model to convert BIM structural data into a unified format, parsing BIM data generated by different software into the IFC standard format; using the semantic parsing model to clean up unstructured text information, eliminate redundant design annotations, and remove duplicate BIM component labels.

[0008] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the rule engine includes, based on Ontology semantic modeling, constructing a BIM structural design knowledge graph, extracting the rules of architectural design standards, and converting the rules of architectural design standards into logical reasoning rules; the knowledge structure of the knowledge graph includes a design specification layer, a rule conversion layer, and a BIM data layer; the design specification layer includes regulatory clauses; the rule conversion layer includes logical reasoning rules; and the BIM data layer includes actual design parameters extracted from BIM data.

[0009] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the structural design specification review includes: rule mapping using a decision tree algorithm to perform specification matching based on the attribute values ​​of the BIM structural model; based on FuzzyLogic optimization, for BIM structural parameters that do not meet the fixed threshold, a fuzzy logic method is used to provide optimization suggestions within the tolerance range and output the review results.

[0010] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the automatic generation of the review report based on natural language processing technology includes: using the BERT pre-trained language model to parse the BIM design manual and extract core technical parameters, including load standards and structural constraints; automatically generating a review report summary through a Seq2Seq structure to summarize BIM structural design issues; and using the TextRank algorithm to automatically screen high-risk review points and provide key correction suggestions.

[0011] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the identification of potential structural conflicts includes using the PointNet++ deep learning framework to analyze 3D point cloud data of the BIM structural model to detect beam-column collisions and floor penetration problems.

[0012] As a preferred solution of the large-model-based BIM model structural design review method described in the present invention, the provision of adjustment suggestions includes: using GNN to reason about BIM component relationships, identifying the mechanical relationships between floor slabs, beams, and walls, and providing adjustment plans; using an LSTM prediction model combined with historical building project data to provide BIM structural optimization suggestions based on big data analysis and generate a 3D interactive optimization plan.

[0013] Another object of the present invention is to provide a BIM model structural design review system based on a large model, which can review the structural design specifications of the BIM model by adopting the built-in rule engine of the AI ​​large model, thereby solving the problem that the current BIM model structural design review technology based on a large model contains a lack of intelligent analysis.

[0014] As a preferred solution of the BIM model structural design review system based on the big model described in the present invention, it includes: a data processing module, a review module, and an optimization module; the data processing module is used to preprocess the BIM model data based on the AI ​​big model; the review module includes a rule engine module and a report generation module, the rule engine module is used to use the built-in rule engine of the AI ​​big model to review the structural design specifications of the BIM model, and the report generation module is used to automatically generate a review report based on natural language processing technology; the optimization module includes a conflict identification module and an adjustment module, the conflict identification module is used to perform BIM model collision detection based on a deep learning algorithm and identify potential structural conflicts, and the adjustment module is used to combine three-dimensional visualization to display error points and provide adjustment suggestions based on historical data and optimization algorithms.

[0015] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a step of a BIM model structural design review method based on a large model.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a BIM model structural design review method based on a large model.

[0017] Beneficial effects of the present invention: The BIM model structural design review method based on a large model provided by the present invention performs BIM model data preprocessing based on the AI ​​large model, improves data quality and consistency, reduces the risk of manual intervention, uses the AI ​​large model's built-in rule engine to review the BIM model's structural design specifications, and automatically generates a review report based on natural language processing technology, avoiding human omissions and subjective judgment deviations, and improving the accuracy and effectiveness of decision-making. BIM model collision detection is performed based on a deep learning algorithm to identify potential structural conflicts, and error points are displayed in combination with three-dimensional visualization. Adjustment suggestions are provided based on historical data and optimization algorithms. Designers can view error points in real time, make decisions quickly and correct problems, thereby improving the scientificity and accuracy of design optimization. The present invention achieves better results in terms of BIM model data processing efficiency, structural design specification review accuracy, and structural conflict identification and optimization suggestion generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 An overall flow chart of a large-model-based BIM model structural design review method provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0021] Example 1, reference Figure 1 , as one embodiment of the present invention, provides a BIM model structural design review method based on a large model, comprising:

[0022] S1: Preprocess BIM model data based on AI big model.

[0023] Furthermore, data preprocessing includes using the outlier detection algorithm of the AI ​​large model to train RNN based on historical BIM design data, automatically identifying outliers in BIM structural parameters and automatically filling them in; converting BIM structural data into a unified format through the Transformer model, parsing BIM data generated by different software into the IFC standard format; and using a semantic parsing model to clean up unstructured text information, eliminate redundant design annotations, and remove duplicate BIM component labels.

[0024] It should be noted that through the BIM model data preprocessing step based on the AI ​​big model, first through the outlier detection algorithm of the AI ​​big model, use the historical BIM design data to train RNN (recursive neural network) to automatically identify outliers in BIM structural parameters. This process can quickly and accurately detect the parts of the data that do not conform to the normal rules and automatically fill them in, avoiding the tedious process of manual inspection and correction in traditional methods, improving the accuracy and reliability of BIM model data, and thus laying a solid foundation for subsequent design review, collision detection and other steps; through the Transformer model, the BIM structural data is converted into a unified format, Parsing BIM data generated by different software into a unified IFC standard format can ensure that the BIM data generated by different software can be seamlessly connected, achieve data compatibility between different software platforms, and improve the efficiency of cross-platform collaborative work. It not only reduces the tedious work of manual data conversion, but also ensures that subsequent steps can be processed in a unified format, avoiding errors caused by format differences; by using a semantic parsing model to clean up unstructured text information, eliminating redundant design comments and repeated BIM component labels, the data quality is further optimized, making the key information in the BIM model clearer and more concise, which is conducive to improving the efficiency of subsequent analysis and review.

[0025] S2: Use the built-in rule engine of the AI ​​big model to review the structural design specifications of the BIM model, and automatically generate a review report based on natural language processing technology.

[0026] Furthermore, the rule engine includes building a BIM structural design knowledge graph based on Ontology semantic modeling, extracting the rules of architectural design standards, and converting the rules of architectural design standards into logical reasoning rules; the knowledge structure of the knowledge graph includes a design specification layer, a rule conversion layer, and a BIM data layer; the design specification layer includes regulatory clauses; the rule conversion layer includes logical reasoning rules; and the BIM data layer includes actual design parameters extracted from BIM data.

[0027] It should also be noted that a preferred solution for constructing a BIM structural design knowledge graph specifically includes extracting specification clauses from architectural design specifications and converting each clause into a standardized node. The specification clauses include the design requirements for structural components such as beams, columns, and walls; using natural language processing technology to extract the logical relationships between the clauses and mapping these relationships into the ontology structure to form a knowledge graph between the design specification layer and the BIM data layer. The logical relationships include the dependency relationships between design requirements, such as "if the span of a beam exceeds 6 meters, additional steel bars are required"; mapping the structural data in the BIM model to the nodes of the knowledge graph. The BIM model data includes design parameters such as the dimensions, materials, and loads of beams, columns, and walls; associating BIM data with design specification requirements through the knowledge graph to provide a basis for structural design review; a preferred solution for extracting the rules of architectural design standards specifically includes extracting explicit rules from architectural design specifications. Each rule includes the structural parameters of the design requirements. The structural parameters include the maximum span of the beam, the minimum thickness of the wall, etc.; using data mining technology to mine potential rules from BIM design data. The potential rules include the common range of structural parameters in historical design data. The data mining technology uses an association rule mining algorithm such as the Apriori algorithm; combining the explicit rules extracted from the design specifications with the potential rules mined from BIM data to form a rule library; matching the rule library with the design data of the BIM model to provide a basis for structural review and generate optimization suggestions.

[0028] It should be noted that the design specification review includes that the rule mapping uses a decision tree algorithm to perform specification matching according to the attribute values of the BIM structural model; based on FuzzyLogic optimization, for BIM structural parameters that do not meet the fixed threshold, a fuzzy logic method is used to provide optimization suggestions within the tolerance range and output the review results.

[0029] It should also be noted that the input features of the decision tree algorithm include parameters such as the dimensions, material types, and load requirements of each structural element in the BIM model. The parameters reflect whether the structural design meets the requirements of the design specifications. The structural elements include beams, columns, walls, etc.; the processing flow of the decision tree algorithm includes: using the labeled BIM design data for training to generate a decision tree model; based on the rules in the design specifications, the splitting nodes of the decision tree correspond to different parameters of the structural elements. The parameters include the span of the beam, the thickness of the wall, etc.; according to the classification results generated by the decision tree, outputting structural review opinions and providing necessary design optimization suggestions; continuously optimizing the decision tree model to improve the review accuracy of the structural design.

[0030] It should also be noted that the automatic generation of review reports based on natural language processing technology includes using the BERT pre-trained language model to parse BIM design specifications and extract core technical parameters, including load standards and structural constraints; automatically generating review report summaries through the Seq2Seq structure to summarize BIM structural design issues; and using the TextRank algorithm to automatically screen high-risk review points and provide key correction suggestions.

[0031] It should also be noted that the source and format of the Seq2Seq model training data. A preferred solution specifically includes collecting a training data set containing BIM model design data and corresponding review reports. The BIM design data includes parameters such as the size, load, and material of each structural element, and the review report includes design compliance assessment, error point marking, modification suggestions, etc.; converting the BIM design data into a standardized format in JSON or XML format, and each data item includes the attributes and design parameters of the structural element; converting the review report into a text format as the target output of the Seq2Seq model, and performing word segmentation and stop word removal on the output text; standardizing the BIM design data and normalizing the numerical data to ensure that the scales of different features are consistent; using the Seq2Seq model for training, the training process uses an encoder to encode the input data and generates the review report text through a decoder.

[0032] It should also be noted that an optimal solution for automatically screening high-risk review points using the TextRank algorithm specifically includes pre-processing the BIM design manual with the TextRank algorithm, removing redundant characters and irrelevant content, and performing sentence segmentation and standardization; treating each sentence in the BIM design manual as a node in the graph, calculating the similarity between sentences, and calculating the semantic similarity between sentences through a word vector model; constructing a sentence graph based on sentence similarity, in which the edges between nodes represent the similarity between sentences; using the PageRank algorithm to iteratively calculate the sentence graph to obtain the score of each sentence, thereby determining the most critical sentences in the text; based on the calculated scores, screening out sentences with high scores as high-risk review points, and providing key correction suggestions.

[0033] It should also be noted that the rule engine built into the AI ​​large model realizes the automatic extraction and specification review of building design standards through the BIM structural design knowledge map built based on Ontology semantic modeling; by converting building design standards into logical reasoning rules, the system can accurately match the structural design specifications according to the specific data and attribute values ​​of the BIM model, and quickly identify parts that do not meet the design requirements. This process not only reduces the workload of manual inspection, but also automatically reviews a large amount of design data through efficient rule mapping (based on decision tree algorithm), greatly improving efficiency and accuracy; at the same time, the method based on FuzzyLogic optimization can use fuzzy logic to identify BIM structural parameters that do not meet the design specifications. The method provides optimization suggestions within the tolerance range, avoids excessive corrections caused by strict standards, and improves the feasibility and practicality of the design scheme; the automatic review report generation based on natural language processing technology uses the BERT pre-trained language model to parse the BIM design manual, extract core technical parameters, and automatically summarize possible problems in the design; the review report summary is generated through the Seq2Seq structure, and the TextRank algorithm is used to screen out high-risk review points and automatically propose correction suggestions, which greatly simplifies the report generation process and improves the quality and pertinence of the report; this automated review report process not only speeds up the project progress, but also improves the accuracy of the review process, so that potential problems in the structural design process can be discovered and handled in a timely manner.

[0034] S3: Performs BIM model collision detection based on deep learning algorithms, identifies potential structural conflicts, displays error points in combination with 3D visualization, and provides adjustment suggestions based on historical data and optimization algorithms.

[0035] Furthermore, identifying potential structural conflicts involves analyzing 3D point cloud data of BIM structural models using the PointNet++ deep learning framework to detect beam-column collisions and floor penetration issues.

[0036] It should also be noted that a preferred solution for 3D point cloud data analysis specifically includes extracting three-dimensional geometric data from the BIM model, including the geometric shape and spatial coordinates of structural elements; converting the extracted geometric data into point cloud data, wherein the outline and surface information of each structural element is converted into a set of discrete points containing XYZ coordinates, and the point cloud data is generated through three-dimensional geometric transformation; denoising and downsampling the generated point cloud data to remove redundant points and reduce data complexity; standardizing the processed point cloud data into a data format suitable for deep learning algorithms, such as an XYZ coordinate matrix; using a spatial collision detection algorithm to detect whether there is a collision between beams and columns. The collision detection determines whether there is spatial overlap by calculating the relative positions of the beam and column nodes in the point cloud data; using a geometric detection algorithm to detect whether the floor slab penetrates other structures. The penetration detection determines whether there is a penetration phenomenon by calculating the distance between the floor slab and other structural elements (such as columns and walls).

[0037] It should be noted that providing adjustment suggestions includes using GNN to reason about BIM component relationships, identifying the mechanical relationships between floor slabs, beams, and walls, and providing adjustment plans; using LSTM prediction models combined with historical building project data to provide BIM structural optimization suggestions based on big data analysis and generate 3D interactive optimization plans.

[0038] It should also be noted that a preferred solution for constructing a GNN graph structure specifically includes defining each structural element in the BIM model as a node in the graph, where the node includes attribute information of the structural element, including the size, material, load requirements, etc. of the structural element; defining edges between nodes, where the edges represent the mechanical relationship or geometric connection relationship between the nodes, and the attributes of the edges include the transferred stress, the distance between adjacent nodes, etc.; constructing an adjacency matrix of the graph based on the relationship between the structural elements in the BIM model, where the connection between each node and other nodes represents the corresponding mechanical or geometric relationship; converting the constructed graph data into an input data format suitable for graph neural networks, including the adjacency matrix, node feature matrix, and the like. Matrix and edge feature matrix; an optimal solution for the LSTM training process specifically includes collecting structural design data from historical building projects, including the structural parameters of each project and its optimization measures; converting historical data into a time series data format suitable for a long short-term memory network (LSTM), where each data point includes a time series representing various parameters of the structural design; constructing an LSTM model, with the input of the model being the historical data of the structural design and the output being the predicted structural optimization solution; using historical building project data to train the LSTM model and optimizing the model parameters through a backpropagation algorithm; using the trained LSTM model to predict new BIM design data and generate a structural design optimization solution.

[0039] It should also be noted that BIM model collision detection is performed based on deep learning algorithms, and the PointNet++ deep learning framework is used to analyze the 3D point cloud data of the BIM structural model to identify potential structural conflicts. In particular, for common structural problems such as beam-column collisions and floor penetrations, the deep learning model can automatically identify these problems and output the precise error location through training on a large amount of structural data, thereby improving the efficiency and accuracy of collision detection; by combining 3D visualization technology, the system can directly display error points in 3D space, allowing engineers to intuitively see the problem location, avoiding conflicts that are difficult to find in traditional 2D drawings; by combining historical buildings Project data and the LSTM prediction model, along with an optimization algorithm based on big data analysis, provide targeted adjustment suggestions for BIM structures. Based on the patterns and trends in historical data, they predict the best structural adjustment plan and provide the optimized adjustment plan in a three-dimensional interactive interface, thereby improving the efficiency and accuracy of design adjustments. This adjustment suggestion not only provides solutions to existing problems, but also takes into account historical data and big data analysis, providing predictive solutions to possible structural problems in the future and improving the adaptability and flexibility of the project. Through this combination of deep learning and big data analysis, the overall intelligence level of the BIM model is improved, making the architectural design process more accurate, intelligent, and efficient.

[0040] Example 2 is an embodiment of the present invention, which provides a BIM model structural design review method based on a large model. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0041] First, BIM models were imported from multiple architectural design software. The data formats were different. In order to process these data, they were pre-processed based on the AI ​​large model. The specific operations included: using the RNN algorithm to detect outliers in the structural parameters of the BIM model (such as the span of the beam, the thickness of the wall, etc.); automatically identifying and filling out outliers through training on historical design data, reducing manual intervention and ensuring data accuracy; converting BIM model data generated by different software into a unified IFC format, and using the Transformer model to standardize the data format; when working across platforms, the compatibility between data is improved; text information cleaning: using the semantic parsing model to clean up unstructured text information, such as redundant design comments, repeated BIM component labels, etc., to make subsequent analysis more efficient; structural design review After data pre-processing, the BIM model is sent to the review module, and the data is reviewed using the model based on Ontology's knowledge graph converts architectural design standards into inference rules. Combined with the actual data of the BIM model (such as structural parameters), a decision tree algorithm is used to match structural design specifications. The BERT pre-trained model is used to extract core technical parameters from the design manual, and a Seq2Seq structure is used to generate a review report summary. The TextRank algorithm is used to screen high-risk review points and automatically provide correction suggestions. This process significantly improves the accuracy and efficiency of design reviews. The PointNet++ deep learning framework is used to analyze the 3D point cloud data of the BIM model, automatically identifying potential beam-column collisions, floor penetrations, and other issues. Based on historical project data and the LSTM prediction model, a graph neural network (GNN) is combined to provide targeted adjustment solutions, presenting error points to designers through 3D visualization and providing real-time optimization suggestions. Refer to Table 1 for a record and analysis of the experimental data.

[0042] Table 1 Experimental data record table

[0043]

[0044] Experiments show that method B (the system of the present invention) successfully reduces the data preprocessing time from 120 seconds in method A (the existing system) to 75 seconds, a reduction of 37.5%, by adopting an outlier detection algorithm based on an AI large model and format conversion technology using the Transformer model. This improvement stems from the introduction of automated processing using the AI ​​model, which avoids the multi-step manual operations required in traditional methods. Automated outlier detection not only improves data accuracy but also reduces the time for manual intervention and potential errors. In addition, the application of the Transformer model in BIM data format conversion eliminates format compatibility issues that arise during manual conversion. It provides a unified standard format for subsequent analysis, further improving overall efficiency. In terms of error detection accuracy, the accuracy of Method B (the system of the present invention) has increased from 85% of the existing technology to 95%, an increase of 11.76%. This significant improvement is mainly due to the application of AI large models, especially the RNN algorithm. Through training on historical BIM data, Method B (the system of the present invention) can more accurately identify abnormalities in structural parameters. Compared with traditional methods, Method B can effectively reduce the incidence of false detection and missed detection, thereby providing more reliable data support for subsequent structural review and optimization. In terms of format conversion time, Method B (the system of the present invention) successfully reduces the time required for format conversion. The conversion time is shortened from 300 seconds in the existing technology to 120 seconds, a reduction of 60%. This improvement shows that through automated IFC format conversion, not only the time cost brought by manual conversion is eliminated, but also the consistency and compatibility of data are enhanced. In scenarios where BIM models involve data generated by multiple software, automated conversion can avoid errors caused by format differences, thereby ensuring the effectiveness and uniformity of data in cross-platform collaboration; the collision detection and optimization suggestion generation links also show obvious advantages. Method B (the system of the present invention) successfully reduced the collision detection time from 250 seconds to 150 seconds through the deep learning framework PointNet++ and LSTM model, which is an improvement. The time required to generate optimization suggestions was reduced from 180 seconds to 120 seconds, representing improvements of 40% and 33.33% respectively. The PointNet++ framework can automatically identify potential structural conflicts such as beam-column collisions and floor penetrations by analyzing the 3D point cloud data of BIM models. The LSTM model combines historical building project data to provide optimization suggestions based on big data analysis. Compared with traditional methods, Method B (the system of the present invention) not only improves the accuracy of collision detection, but also provides more precise structural optimization solutions, improving the efficiency of design adjustments. In terms of total processing time, the comprehensive processing time of Method B (the system of the present invention) was reduced from 850 seconds in the existing system to 565 seconds, a reduction of 33.33%.This overall optimization, 53%, is the result of the interaction of technological advances across multiple links, demonstrating the powerful advantages of integrating AI, big data, and deep learning technologies. Through the synergistic effects of data preprocessing, format conversion, error detection, collision detection, and optimization, Method B (the system of the present invention) not only improves processing efficiency but also ensures higher processing accuracy and a more intelligent design.

[0045] Compared with the existing technology, the present invention not only improves work efficiency, but also enhances the intelligent processing capability of the system, providing a more efficient and accurate BIM model processing solution for the architectural design industry, which has great practical application value and market potential.

[0046] Example 3 is an embodiment of the present invention, which provides a BIM model structural design review system based on a large model, including a data processing module, a review module, and an optimization module.

[0047] Among them, the data processing module is used to preprocess BIM model data based on the AI ​​large model.

[0048] It should be noted that the operation of the system starts with the data processing module, which pre-processes the BIM model data through AI large model-based technology. First, the data processing module detects outliers on the input BIM model data, automatically identifies and fills in missing or abnormal data in the model, and ensures the integrity and accuracy of the data; then, the data processing module performs data format conversion generated by different software, unifies the BIM model data into the IFC standard format, and ensures data compatibility between different platforms; at the same time, unstructured text information is cleaned and optimized through the semantic parsing model, eliminating redundant design annotations and duplicate component labels, and improving the efficiency and data quality of subsequent processing. These preprocessing steps provide a clean, standardized and accurate basic data for subsequent review module and optimization module operations.

[0049] The review module includes a rule engine module and a report generation module. The rule engine module is used to review the structural design specifications of the BIM model using the built-in rule engine of the AI ​​large model. The report generation module is used to automatically generate review reports based on natural language processing technology.

[0050] It should be noted that after preprocessing by the data processing module, the processed data enters the review module. The rule engine module first reviews the BIM model for structural design specifications based on the built-in rule engine of the AI ​​large model and the constructed BIM structural design knowledge graph. By matching the BIM model with the architectural design standards, the system can identify non-compliant parts and provide optimization suggestions within the tolerance range for these non-compliant parameters based on decision tree algorithms and fuzzy logic. The rule engine module ensures a comprehensive inspection of the model data and generates structural review results. Based on the review process, the report generation module automatically generates the review report. Using natural language processing technology, the BERT pre-trained language model parses the design specifications, automatically extracts core technical parameters and key information, and uses a Seq2Seq structure to generate a review report summary. The TextRank algorithm screens high-risk points and provides key correction suggestions. This information is then aggregated into a detailed review report, providing clear and precise feedback to designers. The optimization module receives the information provided by the review module and conducts subsequent structural conflict detection and optimization adjustments.

[0051] The optimization module includes a conflict identification module and an adjustment module. The conflict identification module is used to perform BIM model collision detection based on a deep learning algorithm to identify potential structural conflicts. The adjustment module is used to combine three-dimensional visualization to display error points and provide adjustment suggestions based on historical data and optimization algorithms.

[0052] It should be noted that after the review module generates a report, the system enters the optimization phase. The conflict identification module uses deep learning algorithms, such as the PointNet++ framework, to perform collision detection on the three-dimensional point cloud data of the BIM model. This module can automatically identify potential structural conflict problems, such as beam-column collisions, floor penetrations, etc., and identify and locate error points in the model. Through the training of deep learning models, the conflict identification module improves the accuracy and efficiency of collision detection. After identifying structural conflicts, the adjustment module combines three-dimensional visualization technology to present the error points to designers in an intuitive manner. Designers can clearly see the problem points in the model through an interactive interface. At the same time, the adjustment module combines historical building data with the LSTM prediction model to provide targeted adjustment suggestions based on the optimization algorithm. These suggestions not only propose solutions based on existing conflict problems, but also predict possible structural problems through analysis of historical data and generate adjustment plans. The optimization suggestions at this stage improve the accuracy and adaptability of BIM model design.

[0053] The data processing module provides a unified data input for the optimization module by converting the original BIM data into a standardized format. The conflict identification and adjustment suggestions of the optimization module rely on the accuracy and consistency of the data. If the data processing module fails to correctly handle outliers, format conversion or text cleaning, the results of the optimization module may be affected, resulting in incorrect conflict identification or inaccurate adjustment suggestions. During the operation of the system, the suggestions and conflict detection results output by the optimization module can be fed back to the data processing module for further data optimization. For example, during the conflict detection process, certain errors or abnormal data may be identified and marked as problems by the optimization module. The system can feed this data back to the data processing module for further processing, thereby achieving continuous improvement in data optimization and accuracy.

Claims

1. A BIM model structural design review method based on a large model, characterized by: include: Preprocess BIM model data based on AI big model; Use the built-in rule engine of the AI ​​big model to review the structural design specifications of the BIM model and automatically generate review reports based on natural language processing technology; Perform BIM model collision detection based on deep learning algorithms, identify potential structural conflicts, display error points with 3D visualization, and provide adjustment suggestions based on historical data and optimization algorithms; The rule engine includes building a BIM structural design knowledge graph based on Ontology semantic modeling, extracting rules from architectural design standards, and converting them into logical reasoning rules; Generating review reports includes automatically generating BIM review reports based on natural language processing technology, using the BERT pre-trained language model to extract technical parameters, generating review report summaries through the Seq2Seq structure, and using the TextRank algorithm to screen high-risk review points; Providing adjustment suggestions includes combining GNN and LSTM prediction models to enable the BIM model to identify conflicts in the design, make dynamic optimization suggestions based on historical building data, and generate 3D interactive optimization solutions; Data preprocessing include, Adopting the outlier detection algorithm of the AI ​​large model, the RNN is trained based on historical BIM design data to automatically identify outliers in BIM structural parameters and automatically fill them in. Use the Transformer model to convert BIM structural data into a unified format, parsing BIM data generated by different software into the IFC standard format; Use semantic parsing models to clean up unstructured text information, remove redundant design comments, and remove duplicate BIM component tags; The rule engine includes, Based on Ontology semantic modeling, a BIM structural design knowledge graph is constructed to extract the rules of architectural design standards and convert them into logical reasoning rules; The knowledge structure of the knowledge graph includes the design specification layer, rule conversion layer, and BIM data layer; The design specification layer includes regulatory provisions; The rule conversion layer includes logical reasoning rules; The BIM data layer includes actual design parameters extracted from BIM data; Structural design code review includes: Rule mapping uses a decision tree algorithm to perform specification matching based on the attribute values ​​of the BIM structure model; Based on FuzzyLogic optimization, for BIM structural parameters that do not meet fixed thresholds, fuzzy logic methods are used to provide optimization suggestions within the tolerance range and output review results; Automatically generate review reports based on natural language processing technology, including: Use the BERT pre-trained language model to parse BIM design specifications and extract core technical parameters, including load standards and structural constraints; Automatically generate review report summaries through Seq2Seq structure to summarize BIM structural design issues; Use the TextRank algorithm to automatically screen high-risk review points and provide key correction suggestions; Identifying potential structural conflicts includes, The PointNet++ deep learning framework is used to analyze 3D point cloud data of BIM structural models to detect beam-column collisions and floor penetration issues.

2. The method for reviewing a BIM model structure design based on a large model according to claim 1, wherein: Providing adjustment suggestions includes: Use GNN to reason about BIM component relationships, identify the mechanical relationships between floor slabs, beams, and walls, and provide adjustment solutions; The LSTM prediction model is combined with historical building project data to provide BIM structural optimization suggestions based on big data analysis and generate 3D interactive optimization solutions.

3. A large-model-based BIM model structural design review system, using the large-model-based BIM model structural design review method according to any one of claims 1 to 2, characterized in that: Including data processing module, review module, and optimization module; The data processing module is used to pre-process BIM model data based on the AI ​​large model; The review module includes a rule engine module and a report generation module. The rule engine module is used to review the structural design specifications of the BIM model using the built-in rule engine of the AI ​​large model. The report generation module is used to automatically generate a review report based on natural language processing technology. The optimization module includes a conflict identification module and an adjustment module. The conflict identification module is used to perform BIM model collision detection based on a deep learning algorithm to identify potential structural conflicts. The adjustment module is used to combine three-dimensional visualization to display error points and provide adjustment suggestions based on historical data and optimization algorithms.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the large-model-based BIM model structural design review method according to any one of claims 1 to 2 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the large-model-based BIM model structural design review method according to any one of claims 1 to 2 are implemented.

Images