Multimode construction plan special intelligent auditing method and device

Abstract

This invention provides a multimodal intelligent review method and device for construction schemes. It adopts a layered processing architecture, including three core parts: an input parsing layer, a specialized review engine layer, and an output layer. By deeply integrating general artificial intelligence technology with civil engineering expertise, it achieves multi-dimensional automated review of construction schemes. This solves the problem that existing general AI review tools lack understanding of professional calculation logic, engineering drawing semantics, and industry standard references in civil engineering scenarios. It provides a specialized review capability that deeply integrates the civil engineering knowledge system, significantly improving the professionalism, reliability, and efficiency of construction scheme review.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of information technology in civil engineering construction and artificial intelligence, and specifically relates to a method and equipment for intelligent review of multimodal construction schemes. Background Technology

[0002] Currently, the review of construction plans mainly relies on manual methods, where professional engineers cross-check the textual descriptions, calculation sheets, engineering drawings, and other contents of the plan documents, and conduct compliance reviews in accordance with relevant standards. This method has the following obvious drawbacks:

[0003] 1. Inefficient and costly: A complete construction plan may contain hundreds of pages of content, and manual review takes several days or even weeks, consuming a lot of time from senior technical personnel;

[0004] 2. Consistency in review is difficult to guarantee: If the same proposal is reviewed by different engineers, they may reach different conclusions due to differences in experience; if the same engineer reviews it at different times, they may also make oversights due to differences in status.

[0005] 3. Limited knowledge coverage: It is difficult for a single engineer to fully grasp all relevant standards (national standards, industry standards, local standards, etc.), especially when new and old standards are being replaced, key clauses are easily overlooked;

[0006] 4. Difficulty in associating multimodal information: It is difficult for manual methods to efficiently verify cross-modal consistency between drawing annotations, text descriptions, and calculation parameters, which can easily lead to hidden problems such as discrepancies between drawings and text, and calculation errors. Summary of the Invention

[0007] The purpose of this invention is to provide a method and equipment for intelligent review of multimodal construction schemes.

[0008] To address the above problems, this invention provides a specialized intelligent review method for multimodal construction schemes, comprising:

[0009] The parsing layer receives and processes multimodal input content of construction plans in the field of civil engineering, and transforms the multimodal input content into structured extraction data that can be processed by the subsequent review engine;

[0010] Metadata annotation and association are performed on the structured extracted data to obtain a multimodal associated dataset;

[0011] The multimodal association dataset is input into the special audit engine, and the computational consistency audit, image and text consistency audit, and standard compliance audit are performed in parallel to obtain three types of marking results: computational inconsistency marking results, image and text inconsistency marking results, and standard non-compliance marking results.

[0012] The three types of tagging results are integrated, classified, deduplicated, and precisely located to generate content features, including: precise location, problem classification, review basis, and specific modification suggestions.

[0013] Furthermore, in the above method, the parsing layer receives and processes multimodal input content of construction schemes in the field of civil engineering, transforming the multimodal input content into structured extraction data that can be processed by the subsequent review engine, including:

[0014] Obtain multimodal raw data of the construction plan to be reviewed;

[0015] The original multimodal data is formatted and preprocessed to obtain processed data in a unified format.

[0016] Multimodal content structure extraction is performed on data processed in a uniform format to obtain structured extracted data.

[0017] Furthermore, in the above method, the format of the multimodal raw data is standardized and preprocessed, including:

[0018] For scanned PDFs, OCR technology is used to convert them into editable text;

[0019] For CAD drawings, vector graphics information and annotation text are extracted using a professional parsing engine.

[0020] Furthermore, in the above method, multimodal content structure extraction is performed on the uniformly formatted data to obtain structured extracted data, including:

[0021] By using a rule engine and machine learning models, different content areas in the document are identified, including: technical parameter tables, calculation formula areas, engineering drawing areas, text description areas, and reference list areas.

[0022] Furthermore, in the above method, metadata annotation and association are performed on the structured extracted data to obtain a multimodal associated dataset, including:

[0023] To add associated metadata tags to each extracted content element, including: content type, location information in the document, project section, and potential review concerns, a multimodal associated dataset is obtained.

[0024] Furthermore, in the above method, performing a computational consistency audit includes:

[0025] Mechanical, geometric, and material parameters are extracted from the multimodal correlation dataset and converted into computable numerical values.

[0026] Based on the construction calculation knowledge graph, the corresponding minimal calculation module is matched according to the calculation task type. The calculation modules include, but are not limited to: beam, slab and column load calculation module, steel component stability verification, steel component strength verification, and foundation bearing capacity calculation. The calculation knowledge graph associations between the modules include: load design value and strength verification, load standard value and stability verification, scaffolding calculation and steel component strength verification, scaffolding calculation and steel component stability verification, and scaffolding calculation and foundation verification.

[0027] Input the calculable values ​​into the corresponding minimized calculation module to perform standardized calculations and obtain standard calculation results. Compare the standard calculation results with the calculation results in the original plan. If the results exceed the preset error range, they are marked as calculation inconsistencies. Record the problem location, erroneous parameters, deviation values, and correct results.

[0028] Furthermore, in the above method, the consistency audit of images and text is performed, including:

[0029] Based on computer vision technology, the semantic understanding model for civil engineering drawings identifies component numbers, cross-sectional dimensions, elevations, reinforcement markings, and material specifications in CAD drawings and converts them into structured data for the drawings.

[0030] A cross-modal RAG mechanism is adopted to semantically align the structured data of the drawings with the text data of the scheme according to the component number and position, unify the units and expressions, and obtain the semantic alignment result. Based on the semantic alignment result, the text parameters of the same component are compared with the annotation parameters of the drawings one by one. If the values ​​are inconsistent, it is marked as a problem of inconsistency between the drawing and the text, and the drawing position and text position are recorded accordingly.

[0031] Furthermore, in the above methods, the performance of compliance audits includes:

[0032] Construct a structured knowledge base in the field of civil engineering, including: a database of completed civil engineering-related standards, a semantic network of clauses to establish relationships, and a database of engineering cases;

[0033] Based on RAG retrieval enhancement generation technology, corresponding clauses are retrieved from the civil engineering-specific code knowledge base. Hybrid vector retrieval and multi-level semantic matching are used to improve recall accuracy and obtain a code matching result set. The scheme parameters are compared with the code matching result set item by item. If the requirements are not met, it is marked as a code non-compliance issue, and the code number, clause content, judgment basis and rectification requirements are recorded.

[0034] According to another aspect of the present invention, a computer-readable storage medium is also provided, having stored thereon computer-executable instructions, wherein when executed by a processor, the computer-executable instructions cause the processor to perform the method as described in any of the preceding claims.

[0035] According to another aspect of the present invention, a calculator device is also provided, comprising:

[0036] Processor; and

[0037] A memory configured to store computer-executable instructions, which, when executed, cause the processor to:

[0038] Perform the method described in any of the above steps.

[0039] This invention provides a multimodal intelligent review method for construction schemes in the field of civil engineering. It adopts a layered processing architecture, including three core parts: an input parsing layer, a special review engine layer, and an output layer. By deeply integrating general artificial intelligence technology with civil engineering expertise, it achieves multi-dimensional automated review of construction schemes.

[0040] This invention aims to address the shortcomings of existing general-purpose AI review tools in understanding specialized computational logic, engineering drawing semantics, and industry standard references in civil engineering scenarios. It provides a specialized review capability that deeply integrates with the civil engineering knowledge system, significantly improving the professionalism, reliability, and efficiency of construction plan review. Specific objectives include:

[0041] 1. To achieve automatic identification and verification of professional calculation content in construction plans;

[0042] 2. Implement automatic consistency checks between engineering drawings and text descriptions;

[0043] 3. Implement automated compliance reviews based on industry standards;

[0044] 4. Provide complete audit reports with location markers to support engineers in quickly locating and correcting problems.

[0045] Compared to current methods for inputting carbon emission calculation parameters, the main advantages of this invention are as follows:

[0046] (1) The professionalism of the field has been significantly improved. By deeply integrating the professional knowledge of civil engineering, the illusion problem of general AI tools in the application of professional fields has been solved, and the professionalism and reliability of the audit results have reached the level of engineering practicality.

[0047] (2) The review efficiency has been greatly improved. Automated review can shorten the review process that originally took several days to be completed in a few hours, freeing up the time of senior technical personnel;

[0048] (3) Enhanced consistency in audit quality: Based on a unified knowledge base and audit rules, the audit differences between different engineers have been eliminated, ensuring the consistency of audit standards;

[0049] (4) Multi-dimensional cross-validation capability enables comprehensive review of construction plans, including multiple dimensions such as calculation correctness, consistency of graphics and text, and compliance with standards, covering hidden issues that are easily overlooked by manual review.

[0050] (5) Knowledge accumulation and inheritance: The system’s built-in knowledge base and audit rules can be continuously accumulated and optimized to form corporate knowledge assets and reduce reliance on the experience of individual experts;

[0051] (6) Continuous learning and adaptability: The system can continuously optimize the audit rules through the feedback mechanism to adapt to the updates of standards and the development of new technologies. Attached Figure Description

[0052] Figure 1 This is a flowchart of a multimodal construction scheme intelligent review method according to an embodiment of the present invention. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] While some general-purpose document review AI tools exist on the market, these tools primarily target areas such as legal contracts and financial statements. They lack a deep understanding of civil engineering terminology, computational logic, engineering drawing semantics, and industry standards, thus failing to meet the professional review needs of construction plans. Specifically, existing technologies have the following shortcomings:

[0055] 1. Insufficient understanding of professional calculation logic: Unable to identify and verify professional calculations in construction plans, such as structural mechanics calculations and material usage calculations;

[0056] 2. Insufficient understanding of the semantics of engineering drawings: Unable to understand the professional annotations in CAD drawings and connect them to the text descriptions;

[0057] 3. Insufficient understanding of industry standards: Unable to intelligently match and reference relevant standard clauses based on specific engineering scenarios;

[0058] 4. Insufficient multimodal information association capability: It cannot effectively achieve cross-validation between text, tables, formulas, and drawings.

[0059] like Figure 1 As shown, this invention provides a specialized intelligent review method for multimodal construction schemes, comprising:

[0060] Step S1: The parsing layer receives and processes multimodal input content of construction schemes in the field of civil engineering, including text, tables, formulas, CAD drawings, etc., and transforms the multimodal input content into structured extraction data that can be processed by the subsequent review engine.

[0061] Specifically, step S1 includes:

[0062] Step S11: Obtain multimodal raw data of the construction plan to be reviewed. The multimodal raw data includes: text, tables, formulas and CAD drawings, in formats including PDF, Word, DWG, DXF and drawings.

[0063] Step S12: Unify and preprocess the format of the multimodal raw data, perform OCR optical character recognition on the scanned document, perform CAD professional parsing on the DWG / DXF format, and perform structured decomposition on the PDF and Word formats to obtain data with unified format processing.

[0064] Here, steps S11 and S12 involve unified processing of multiple document formats: the system supports construction plan documents in various formats such as PDF, Word, and CAD (DXF / DWG). For scanned PDFs, OCR (Optical Character Recognition) technology is used to convert them into editable text; for CAD drawings, a professional parsing engine extracts vector graphics information and annotation text.

[0065] Step S13: Perform multimodal content structured extraction on the uniformly formatted data. Based on the rule engine and the civil engineering-specific named entity recognition model, identify the technical parameter area, calculation formula area, engineering drawing area, text description area and standard reference area, extract component number, cross-sectional dimensions, material specifications, load parameters, standard and reinforcement ratio professional parameters, convert character parameters into calculable physical quantities, and obtain structured extracted data.

[0066] This step involves content structure extraction: using a rule engine and machine learning model, different content areas in the document are identified, including: technical parameter tables, calculation formula areas, engineering drawing areas, text description areas, and lists of referenced standards. Specifically tailored to the characteristics of the civil engineering field, the system can identify structural patterns using professional expressions such as "concrete strength grade C30" and "beam cross-section dimensions 300×600".

[0067] Step S2 involves metadata annotation and association of the structured extracted data. This includes adding associated metadata tags to each extracted content element, including content type (text, table, formula, drawing), location information in the document (page number, paragraph position), part of the project (underground structure, main structure), and potential review concerns (calculation parameters, specification references, drawing annotations), resulting in a multimodal associated dataset.

[0068] Step S3: Input the multimodal association dataset into the special audit engine and perform computational consistency audit, image and text consistency audit, and standard compliance audit in parallel to obtain three types of marking results: computational inconsistency marking results, image and text inconsistency marking results, and standard non-compliance marking results.

[0069] Here, step S3 includes three audit modules specifically designed for the civil engineering field. Each module deeply integrates general AI technology with civil engineering expertise to form a domain-specific audit capability.

[0070] Preferably, a computational consistency audit is performed, which specifically includes:

[0071] Step S311: Extract mechanical parameters, geometric parameters, and material parameters from the multimodal correlation dataset, and convert them into computable numerical values ​​respectively;

[0072] Here, the civil engineering-specific parameter recognition engine: the calculation consistency audit module has a built-in parameter recognition model for the civil engineering field, which can accurately identify professional parameters in the text, such as: standard load values ​​(e.g., "live load 2.5kN / m²"), concrete strength grade (e.g., "C30"), steel reinforcement strength (e.g., "HRB400"), reinforcement ratio and other mechanical parameters; interface dimensions, span, length, height and other geometric parameters; as well as concrete mix proportions, steel specifications, formwork type and other material parameters.

[0073] Example explanation: When the system recognizes "beam cross-section 300×600, concrete C30, longitudinal tensile reinforcement using 4Φ25", it can automatically extract results such as: cross-section width 300mm, height 600mm, concrete compressive strength design value 14.3N / mm² (converted according to C30), and total reinforcement area 1964mm² (4 25mm diameter steel bars), instead of the traditional named entity recognition method that only extracts and labels the original text content.

[0074] Step S312: Based on the construction calculation knowledge graph, match the corresponding minimized calculation module according to the calculation task type. The calculation module includes: beam, slab and column load calculation, steel component strength verification, steel component stability verification, foundation bearing capacity verification, scaffolding verification and formwork support verification.

[0075] Here, a built-in civil engineering calculation model library is a key innovation of this invention. The system incorporates engineering-verified civil engineering calculation models. This civil engineering calculation model library encapsulates the calculation logic and formulas unique to the civil engineering field, and associates the relationships between various calculation modules and the task blocks included in the calculation tasks through a construction calculation knowledge graph. This ensures the minimization of model calculation modules and the flexibility of module combination and calculation. Its core calculation modules include, but are not limited to: beam, slab and column load calculation module, steel component stability verification, steel component strength verification, and foundation bearing capacity calculation. The calculation knowledge graph associations between each module include: "load design value → strength verification", "load standard value → stability verification" or "scaffolding calculation → steel component strength verification", "scaffolding calculation → steel component stability verification", "scaffolding calculation → foundation verification".

[0076] Compared to traditional specialized calculation tools, this civil engineering professional calculation model library provides minimum calculation modules and module guidelines for subsequent automatic calculation modules to call. For example, if calculating scaffolding requires separate strength verification, stability verification, foundation verification, and wall tie verification for steel components, then when encountering this task, these four calculation modules can be called to complete the calculation tasks for each item.

[0077] Step S313: Input the calculable value into the corresponding minimized calculation module to perform standardized calculation and obtain the standard calculation result; compare the standard calculation result with the original calculation result of the scheme; if it exceeds the preset error range, mark it as a calculation inconsistency problem and record the problem location, error parameter, deviation value and correct result.

[0078] Here, the consistency verification module for calculations is problematic because traditional AI tools cannot understand the professional logic of construction-specific calculations and lack the performance to provide stable data calculation results for complex calculations, thus failing to verify the correctness of the calculation results.

[0079] Automatic calculation verification and deviation detection: The system identifies the calculation modules that need to be completed based on the specific calculation task, inputs the calculable values ​​into the corresponding professional calculation model, recalculates the results, and then compares them with the calculation results in the original plan.

[0080] Workflow example: The original plan calculated the scaffolding. The model, based on the construction project knowledge of civil engineering, confirmed that four calculation tasks need to be completed: strength verification of steel components, stability verification, foundation verification, and wall tie verification. Through parameter identification and calling the built-in civil engineering calculation model library, the corresponding modules are recalculated. Then, the results are compared with those in the original plan to confirm the deviation of the results and whether the calculation items have been completed.

[0081] Preferably, a consistency audit of the text and images is performed, specifically including:

[0082] Step S321: Based on the semantic understanding model of civil engineering drawings, identify and understand professional annotation symbols in CAD drawings, extract key engineering information from the drawings, and understand the spatial relationships and construction logic in the drawings;

[0083] Here, the semantic understanding model for civil engineering drawings, based on computer vision technology, can identify component numbers, cross-sectional dimensions, elevations, reinforcement markings, and material specifications in CAD drawings and convert them into structured data for the drawings.

[0084] For example, for a beam reinforcement drawing, the system can identify: beam number "KL-1", cross-sectional dimensions "300×600", elevation "+3.000", and reinforcement markings "4Φ25" and "Φ8@200", etc.

[0085] Step S322: Using the cross-modal RAG mechanism, the structured data of the drawings and the text data of the scheme are semantically aligned according to the component number and position, and the units and expressions are unified to obtain the semantic alignment result. Based on the semantic alignment result, the text parameters of the same component are compared with the annotation parameters of the drawings one by one. If the values ​​are inconsistent, it is marked as a problem of inconsistency between the drawing and the text, and the drawing position and text position are recorded accordingly.

[0086] Here, the cross-modal RAG (Retrieval Enhanced Generation) mechanism is another key innovation of this invention. Traditional multimodal AI can only process images and text separately, making it difficult to establish semantic connections between the two. This invention introduces a cross-modal RAG mechanism, specifically tailored to the characteristics of the civil engineering field, to unify semantic construction, engineering semantic-based association retrieval, and intelligent alignment and comparison.

[0087] For example: The system extracts the annotation information from the drawing as "KL-1 beam section: 300×600", while the content retrieved from the system text description is "second-floor frame beam KL-1, section size is 300×550". The system performs semantic alignment, setting the height of the drawing annotation to 600mm and the height of the text description to 550mm. Finally, the system performs consistency judgment and marks it as inconsistent between the drawing and the text, and clearly marks the drawing location and the text location in the review report.

[0088] Preferably, compliance audits are conducted, specifically including:

[0089] Step S331: Based on the project type, structural type, component type, and construction procedure, construct the corresponding civil engineering-specific specification knowledge base;

[0090] Here, we construct a dedicated knowledge base for civil engineering: a structured knowledge base specifically for the field of civil engineering, including a database of completed civil engineering-related standards, a semantic network of clauses to establish relationships, and a database of engineering cases.

[0091] Step S332: Based on RAG retrieval enhancement generation technology, retrieve the corresponding clauses from the civil engineering-specific specification knowledge base. Use hybrid vector retrieval and multi-level semantic matching to improve recall accuracy and obtain a specification matching result set (specification number, clause number, original clause text, limit / requirement). Compare the scheme parameters with the specification matching result set item by item. If the requirements are not met, mark it as a specification non-compliance issue and record the specification number, clause content, judgment basis, and rectification requirements.

[0092] Here, the intelligent specification matching based on RAG is used: it adopts retrieval enhancement generation technology, but has been specially optimized for the application scenario of civil engineering specifications. The system analyzes the specific context of the construction plan, dynamically determines the applicable specification scope, and improves the accuracy of knowledge retrieval by using multi-level semantic matching and hybrid vector retrieval. At the same time, it dynamically updates the specifications and performs conflict detection.

[0093] For example: The system recognizes the text description as "cantilevered balcony slab, cantilever length 4.2m, slab thickness 200mm". Based on the RAG code knowledge base for the "Code for Design of Concrete Structures" GB 50010-2010, Article 9.1.2 states: "The minimum thickness of the cantilevered slab shall not be less than 1 / 12 of the span". The system determines that the slab thickness in the text should not be less than 350mm. Therefore, the system concludes that the code does not meet the compliance requirements and marks the specific clause and calculation basis.

[0094] Here, the construction plan needs to comply with a large number of national and industry standards. Engineers find it difficult to remember all the clauses, and the standards are updated frequently, making it easy to refer to outdated or inapplicable standards. The compliance review module of this standard introduces the above-mentioned innovative solutions.

[0095] The consistency review module for text and image: Inconsistencies between the text descriptions in the construction plan and the annotations on the CAD drawings are a common problem. Traditional methods require manual verification one by one, which is inefficient and prone to omissions.

[0096] Step S4 involves integrating, classifying, deduplicating, and precisely locating the three types of marked results to generate content features including: precise location (specific page number, paragraph number, and character position range for each review issue), issue classification (including type classifications such as "inconsistent calculations," "inconsistent text and images," and "non-compliant with regulations"), review basis (including detailed calculation review process, specific original text of the regulations cited, and text and image comparison), and specific modification suggestions.

[0097] A structured audit report that includes a list of issues, audit basis, calculation process, standard clauses, and suggested modifications.

[0098] Here, the output layer integrates the results of each audit module to generate a structured audit report.

[0099] This invention belongs to the interdisciplinary field of civil engineering construction informatization and artificial intelligence, specifically involving a method and system for automated intelligent review of professional dimensions such as structural calculation consistency, drawing-text consistency, and compliance with specification clauses in construction plans.

[0100] This invention provides a multimodal intelligent review method for construction schemes in the field of civil engineering. It adopts a layered processing architecture, including three core parts: an input parsing layer, a special review engine layer, and an output layer. By deeply integrating general artificial intelligence technology with civil engineering expertise, it achieves multi-dimensional automated review of construction schemes.

[0101] This invention aims to address the shortcomings of existing general-purpose AI review tools in understanding specialized computational logic, engineering drawing semantics, and industry standard references in civil engineering scenarios. It provides a specialized review capability that deeply integrates with the civil engineering knowledge system, significantly improving the professionalism, reliability, and efficiency of construction plan review. Specific objectives include:

[0102] 1. To achieve automatic identification and verification of professional calculation content in construction plans;

[0103] 2. Implement automatic consistency checks between engineering drawings and text descriptions;

[0104] 3. Implement automated compliance reviews based on industry standards;

[0105] 4. Provide complete audit reports with location markers to support engineers in quickly locating and correcting problems.

[0106] Compared to current methods for inputting carbon emission calculation parameters, the main advantages of this invention are as follows:

[0107] (1) The professionalism of the field has been significantly improved. By deeply integrating the professional knowledge of civil engineering, the illusion problem of general AI tools in the application of professional fields has been solved, and the professionalism and reliability of the audit results have reached the level of engineering practicality.

[0108] (2) The review efficiency has been greatly improved. Automated review can shorten the review process that originally took several days to be completed in a few hours, freeing up the time of senior technical personnel;

[0109] (3) Enhanced consistency in audit quality: Based on a unified knowledge base and audit rules, the audit differences between different engineers have been eliminated, ensuring the consistency of audit standards;

[0110] (4) Multi-dimensional cross-validation capability enables comprehensive review of construction plans, including multiple dimensions such as calculation correctness, consistency of graphics and text, and compliance with standards, covering hidden issues that are easily overlooked by manual review.

[0111] (5) Knowledge accumulation and inheritance: The system’s built-in knowledge base and audit rules can be continuously accumulated and optimized to form corporate knowledge assets and reduce reliance on the experience of individual experts;

[0112] (6) Continuous learning and adaptability: The system can continuously optimize the audit rules through the feedback mechanism to adapt to the updates of standards and the development of new technologies.

[0113] According to another aspect of the present invention, a computer-readable storage medium is also provided, having stored thereon computer-executable instructions, wherein when executed by a processor, the computer-executable instructions cause the processor to perform the method as described in any of the preceding claims.

[0114] According to another aspect of the present invention, a calculator device is also provided, comprising:

[0115] Processor; and

[0116] A memory configured to store computer-executable instructions, which, when executed, cause the processor to:

[0117] Perform the method described in any of the above steps.

[0118] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0119] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0120] Obviously, those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A multi-modal construction plan special intelligent auditing method, characterized in that, include: The parsing layer receives and processes multimodal input content of construction plans in the field of civil engineering, and transforms the multimodal input content into structured extraction data that can be processed by the subsequent review engine; Metadata annotation and association are performed on the structured extracted data to obtain a multimodal associated dataset; The multimodal association dataset is input into the special audit engine, and the computational consistency audit, image and text consistency audit, and standard compliance audit are performed in parallel to obtain three types of marking results: computational inconsistency marking results, image and text inconsistency marking results, and standard non-compliance marking results. The three types of tagging results are integrated, classified, deduplicated, and precisely located to generate content features, including: precise location, problem classification, review basis, and specific modification suggestions.

2. The multi-modal construction plan special intelligent auditing method of claim 1, wherein, The parsing layer receives and processes multimodal input content from construction plans in the civil engineering field, transforming it into structured extraction data that can be processed by the subsequent review engine, including: Obtain multimodal raw data of the construction plan to be reviewed; The original multimodal data is formatted and preprocessed to obtain processed data in a unified format. Multimodal content structure extraction is performed on data processed in a uniform format to obtain structured extracted data.

3. The multi-modal construction plan special intelligent auditing method of claim 2, wherein, The raw multimodal data undergoes format standardization and preprocessing, including: For scanned PDFs, OCR technology is used to convert them into editable text; For CAD drawings, vector graphics information and annotation text are extracted using a professional parsing engine.

4. The multi-modal construction plan special intelligent auditing method of claim 2, wherein, Multimodal content structure extraction is performed on uniformly formatted data to obtain structured extracted data, including: Through rule engines and machine learning models, different content areas in the document are identified, including: technical parameter tables, calculation formula areas, engineering drawing areas, text description areas, and reference specification lists.

5. The multi-modal construction plan special intelligent auditing method of claim 1, wherein, Metadata annotation and association are performed on the structured extracted data to obtain a multimodal associated dataset, including: To add associated metadata tags to each extracted content element, including: content type, location information in the document, project section, and potential review concerns, a multimodal associated dataset is obtained.

6. The intelligent review method for multimodal construction schemes as described in claim 1, characterized in that, Perform a computational consistency audit, including: Mechanical, geometric, and material parameters are extracted from the multimodal correlation dataset and converted into computable numerical values. Based on the construction calculation knowledge graph, the corresponding minimal calculation module is matched according to the calculation task type. The calculation modules include, but are not limited to: beam, slab and column load calculation module, steel component stability verification, steel component strength verification, and foundation bearing capacity calculation. The calculation knowledge graph associations between the modules include: load design value and strength verification, load standard value and stability verification, scaffolding calculation and steel component strength verification, scaffolding calculation and steel component stability verification, and scaffolding calculation and foundation verification. Input the calculable values ​​into the corresponding minimized calculation module to perform standardized calculations and obtain standard calculation results. Compare the standard calculation results with the calculation results in the original plan. If the results exceed the preset error range, they are marked as calculation inconsistencies. Record the problem location, erroneous parameters, deviation values, and correct results.

7. The intelligent review method for multimodal construction schemes as described in claim 1, characterized in that, Perform a consistency audit of images and text, including: Based on computer vision technology, the semantic understanding model for civil engineering drawings identifies component numbers, cross-sectional dimensions, elevations, reinforcement markings, and material specifications in CAD drawings and converts them into structured data for the drawings. A cross-modal RAG mechanism is adopted to semantically align the structured data of the drawings with the text data of the scheme according to the component number and position, unify the units and expressions, and obtain the semantic alignment result. Based on the semantic alignment result, the text parameters of the same component are compared with the annotation parameters of the drawings one by one. If the values ​​are inconsistent, it is marked as a problem of inconsistency between the drawing and the text, and the drawing position and text position are recorded accordingly.

8. The intelligent review method for multimodal construction schemes as described in claim 1, characterized in that, Perform compliance audits, including: Construct a structured knowledge base in the field of civil engineering, including: a database of completed civil engineering-related standards, a semantic network of clauses to establish relationships, and a database of engineering cases; Based on RAG retrieval enhancement generation technology, corresponding clauses are retrieved from the civil engineering-specific code knowledge base. Hybrid vector retrieval and multi-level semantic matching are used to improve recall accuracy and obtain a code matching result set. The scheme parameters are compared with the code matching result set item by item. If the requirements are not met, it is marked as a code non-compliance issue, and the code number, clause content, judgment basis and rectification requirements are recorded.

9. A computer-readable storage medium having stored thereon computer-executable instructions, wherein, When the computer-executable instructions are executed by the processor, the processor causes the processor to perform the method as described in any one of claims 1 to 8.

10. A calculator device, wherein, include: processor; as well as A memory configured to store computer-executable instructions, which, when executed, cause the processor to: Perform the method as described in any one of claims 1 to 8.

Images