Engineering drawing compliance intelligent review method and system constructed based on knowledge base and large model

Abstract

The invention discloses an engineering drawing compliance intelligent review method and system constructed based on a knowledge base and a large model. The method mainly comprises the following steps: intelligently analyzing a design specification in a natural language form by utilizing a large language model, and dynamically constructing a machine-readable rule base and a knowledge base; uniformly converting the multi-format engineering drawing into a structured intermediate format; functional areas and primitives in the drawing are recognized through a computer vision model, and natural language information describing attributes of the functional areas and the primitives is generated; and finally, in combination with the knowledge base and natural language description, performing compliance judgment by utilizing the reasoning ability of the large language model, and generating an interpretable review report. According to the method, computer vision and a large language model technology are fused, so that the defects that traditional manual review is low in efficiency and prone to making mistakes and a traditional automatic tool lacks semantic understanding ability are overcome, and efficient and accurate engineering drawing automatic compliance review with deep semantic understanding ability is achieved. Figure 1 of the abstract is a system architecture block diagram.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence technology and engineering drawing examination, and particularly relates to an engineering drawing compliance intelligent examination method and system based on a knowledge base and a large language model. BACKGROUND

[0002] Engineering drawings (such as architectural, mechanical, and electrical drawings) are the fundamental basis for engineering project implementation, and their compliance examination (i.e., examining whether the drawings comply with relevant laws, regulations, design specifications, and industry standards) is a key link to ensure engineering quality and safety. Currently, the mainstream examination methods mainly have the following two types:

[0003] 1. Manual examination: relying on the personal experience and professional knowledge of the examination engineer to check the drawings item by item. This method has obvious defects: low examination efficiency, especially for large and complex projects; the examination result is affected by personal subjective factors, and errors and omissions are easily made; the examination cost is high, and it is difficult to ensure standardization.

[0004] 2. Traditional automated tool examination: some existing computer-aided design (CAD) software or independent examination tools have certain automatic checking functions, usually based on pre-defined hard-coded rules. However, these tools have poor flexibility and are difficult to adapt to rapidly updated specification provisions and diversified customized examination needs. More importantly, traditional tools lack semantic understanding ability for drawings. They can only identify the geometric properties of graphics (such as line length, angle), but cannot understand the actual engineering meaning represented by the graphics (for example, they cannot identify whether a rectangular area represents a "load-bearing wall" or a "normal partition"), resulting in the inability to effectively examine complex specification clauses involving semantic information.

[0005] Therefore, there is an urgent need for an automated examination solution that can balance examination efficiency, accuracy, flexibility, and deep semantic understanding ability. SUMMARY

[0006] The purpose of the present application is to overcome the deficiencies of the prior art and provide an engineering drawing compliance intelligent examination method and system based on a knowledge base and a large model construction to solve the problems of low efficiency, easy errors of manual examination, and poor flexibility and inability to identify graphic semantics of existing automated tools.

[0007] To achieve the above purpose, the present application adopts the following technical solutions:

[0008] In a first aspect, the present application provides an engineering drawing compliance intelligent examination system based on a knowledge base and a large model construction, characterized in that the system comprises:

[0009] The rule base and knowledge base construction module is configured to use a large language model to intelligently analyze and structure engineering design specification documents in natural language form, extract review items, constraints and their logical relationships, and construct a rule base and knowledge base that can be understood and reasoned by computers.

[0010] The format conversion module is configured to receive engineering drawing files in various original formats and convert them into a unified, structured intermediate format file to eliminate format differences;

[0011] The drawing parsing and semantic understanding module is configured to use a computer vision model to process the drawings in the unified intermediate format, identify and determine the functional areas in the drawings and the drawing data within the areas, and construct natural language description information describing the drawing elements and their attributes based on the identification results.

[0012] The review execution module is configured to associate the natural language description information output by the drawing parsing and semantic understanding module with the rules in the rule base and knowledge base, use the reasoning ability of the large language model to make compliance judgments on each element in the drawing, and generate a review report containing specific violations and location information.

[0013] Secondly, the present invention provides an intelligent review method for the compliance of engineering drawings based on a knowledge base and a large model, characterized in that the method includes the following steps:

[0014] S1: Utilize large language models to analyze engineering design specification documents, extract review rules and knowledge, and build a machine-readable rule base and knowledge base;

[0015] S2: Convert the received multi-format original engineering drawings into a unified intermediate format;

[0016] S3: Use computer vision models to parse drawings in a unified intermediate format, identify functional areas and graphic elements in the drawings, and generate corresponding natural language description information.

[0017] S4: Based on the rule base, knowledge base, and natural language description information, use a large language model to conduct a compliance review of the drawing elements and output the review results.

[0018] Preferably, in step S1, analyzing the specification using a large language model includes: inputting the specification document into the large language model in segments, and the instruction model identifying and structurally outputting the "review object", "constraints", "compliance standards" and "reference clauses".

[0019] Preferably, in step S3, the computer vision model is used to complete and determine each area of ​​the drawing and the graphic data in the area, which specifically includes: using an object detection model to identify functional areas such as rooms, doors, windows, and equipment in the drawing; using a graphic element recognition model to identify basic graphic elements such as walls, lines, and text in the area; and combining the recognition results with the drawing layers and attribute information to generate structured graphic data.

[0020] Preferably, in step S3, constructing natural language description information involves converting structured graph data into coherent natural language paragraphs using a predefined template or large language model. For example: "In the area of ​​'Office A', a 'door' with a width of 900mm is identified, and its straight-line distance from the nearest 'evacuation passage' is 15 meters."

[0021] Preferably, in step S4, reviewing the drawings using a knowledge base and a large model involves inputting the natural language description information and the rules in the knowledge base into the large language model, using the large language model as an inference engine to determine whether the described scenario violates relevant rules and to provide an inference chain.

[0022] The beneficial effects of this invention are as follows:

[0023] 1. High Efficiency and Accuracy: The automated process replaces a significant amount of repetitive manual review work, resulting in a review speed far exceeding that of manual review. Furthermore, the combination of large language models and computer vision models enables near-expert-level understanding and judgment, greatly reducing human error and oversight. In testing, the system reviewed an architectural drawing in an average of 5 minutes, a 96% improvement in efficiency compared to manual review (average 2 hours); the accuracy rate for reviewing 100 regulatory clauses reached 98%.

[0024] 2. Powerful semantic understanding capability: This invention breaks through the limitation of traditional tools that can only process geometric information. Through the technical path of "computer vision recognition of primitives + understanding semantics through a large language model", the computer can understand the engineering meaning of drawings, thereby enabling it to review specification clauses involving complex semantics.

[0025] 3. High flexibility and customizability: The rule base and knowledge base are not hard-coded, but dynamically constructed from natural language specifications through a large language model. When the specifications are updated or users have specific review requirements, the system can quickly reconstruct the knowledge base by simply providing the new specification document or customized terms, without modifying the core code, making it highly adaptable.

[0026] 4. The review process is interpretable: The review based on the large language model can generate review reports with reasoning processes, clearly pointing out violations, specific clauses violated, and reasoning logic, which makes it easier for designers to quickly locate and modify problems, and improves the transparency and credibility of the review. Attached Figure Description

[0027] Figure 1 This is an architecture block diagram of an intelligent review system for engineering drawing compliance provided in an embodiment of the present invention.

[0028] Figure 2 This is a flowchart illustrating an intelligent review method for the compliance of engineering drawings provided in an embodiment of the present invention.

[0029] Figure 3 This is a schematic diagram of drawing parsing and natural language description information generation in one embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0031] Example 1:

[0032] refer to Figure 1 The system in this embodiment includes four core modules: a format conversion module, a rule base and knowledge base construction module, a drawing parsing and semantic understanding module, and a review and execution module.

[0033] The rule base and knowledge base construction module segments the PDF document of "Code for Fire Protection Design of Buildings" (GB50016) and inputs it into a large language model (e.g., DeepSeek or GPT-4), with the following prompt: "Please extract all review points from the following code clauses. The output is in JSON format, containing the fields: rule_id (rule number), rule_text (rule text), target_object (review object, such as 'door', 'corridor'), and constraint (constraint conditions, such as 'width ≥ 1.2m', 'distance ≤ 40m')." This method transforms the unstructured code text into a structured rule database.

[0034] When a review is required, the user uploads an architectural floor plan in DWG or PDF format. The format conversion module first converts it into a unified, structured intermediate format (such as a custom JSON or SVG format) that includes information such as layers, elements, and attributes.

[0035] Next, the drawing parsing and semantic understanding module begins its work. This module uses trained computer vision models (such as object detection models based on YOLO and Faster R-CNN, and semantic segmentation models based on U-Net) to process the intermediate format drawings. First, it identifies different functional areas in the drawings, such as "Room 101", "Corridor L1", and "Stairwell S1". Then, it identifies specific primitives within each area, such as "Door D1", "Window W1", and "Wall" within "Room 101", and extracts their geometric attributes (such as the width of the door, the height of the window, and the length of the wall). Finally, it integrates this structured information to generate a natural language description fragment, for example: "Area 'Room 101' is identified as 'Office'. It contains a door 'D1' with a measured width of 0.85 meters. The calculated total length of the evacuation path from the door of this room to the nearest safe exit 'Stairwell S1' is 42 meters."

[0036] Finally, the review and execution module inputs the above natural language description fragment along with the rules in the rule base regarding "evacuation door width" and "evacuation distance" into an LLM instance (such as DeepSeek and ChatGLM). The large language model will perform the following reasoning: "According to rule R001, the width of evacuation doors in office areas should not be less than 0.9 meters. In the description, the width of door D1 is 0.85 meters, which is less than the specified value, therefore it is non-compliant. According to rule R005, the evacuation distance of a room located at the end of a dead-end corridor should not be greater than 22 meters. In the description, the evacuation distance of room 101 is 42 meters, which is much greater than the specified value, therefore it is non-compliant." Ultimately, the module will generate a detailed review report, listing all violations and associating them with specific element IDs and rule clauses. Alternatively, multiple LLM instances (such as DeepSeek and ChatGLM) can be input, and a voting strategy can be used to determine the compliance result. For disputed items, the system generates a highlighted prompt and suggests manual review.

[0037] Example 2:

[0038] In another embodiment, a user wishes to review the process compliance of a set of mechanical assembly drawings. The system flow is similar to that of Embodiment 1. The difference lies in that the rule base construction module analyzes the chapters on tolerances and fits, and geometric tolerances in the "Mechanical Design Handbook." The drawing parsing module focuses on identifying mechanical components such as shafts, holes, and bearing housings, as well as their dimensional tolerances and geometric tolerance symbols. The review execution module compares the identified tolerance information with the rules in the knowledge base to determine whether the fit is reasonable and whether there is any risk of interference.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent review of the compliance of engineering drawings based on a knowledge base and a large model, characterized in that, Includes the following steps: The steps for building the rule base and knowledge base are as follows: Utilize a large language model to intelligently analyze and structure engineering design specification documents in natural language form, extract review items, constraints, and their logical relationships, and build a machine-readable rule base and knowledge base. Drawing format standardization steps: Convert the received engineering drawing files in various original formats into a unified, structured intermediate format file; Drawing semantic parsing steps: Use a computer vision model to process the unified intermediate format drawings, identify and determine each functional area in the drawings and the drawing data within the area, and construct natural language description information describing the drawing elements and their attributes based on the identification results. Intelligent review and report generation steps: The natural language description information is associated with the rules in the rule base and knowledge base. The reasoning ability of the large language model is used to judge the compliance of each element in the drawing and generate a review report containing specific violations and location information.

2. The method according to claim 1, characterized in that, The specific steps for constructing the rule base and knowledge base include: Divide the engineering design specification document into sections; The segmented standard text is input into the large language model, and the large language model is instructed to identify and output the review objects, constraints, compliance standards and reference clauses in a structured manner. The structured output of the large language model is organized and stored as a rule base and a knowledge base.

3. The method according to claim 1, characterized in that, The drawing semantic parsing step involves identifying and determining each functional area in the drawing, as well as the graphic data within those areas. Specifically, this includes: Use object detection models to identify predefined functional areas in drawings; The basic primitives and their geometric attributes within the functional area are identified using a primitive recognition model. The identified functional areas and graphic elements are integrated with the layers and attribute information of the drawing to generate a structured set of graphic data.

4. The method according to claim 3, characterized in that, The construction of natural language description information involves transforming the structured graph data set into coherent natural language paragraphs using predefined natural language templates or large language models.

5. The method according to claim 1, characterized in that, In the intelligent review and report generation step, the natural language description information and the rules in the knowledge base are input into the large language model. The large language model performs logical reasoning to determine whether the drawing scene corresponding to the natural language description information violates the relevant rules, and outputs a review result containing the reasoning chain.

6. A knowledge-based and large-model-based intelligent review system for engineering drawing compliance, used to implement the method described in any one of claims 1 to 5, characterized in that, include: The rule base and knowledge base construction module is configured to use a large language model to intelligently analyze and structure engineering design specification documents in natural language form, extract review items, constraints and their logical relationships, and build a machine-readable rule base and knowledge base. The format conversion module is configured to receive engineering drawing files in various original formats and convert them into a unified, structured intermediate format file. The drawing parsing and semantic understanding module is configured to use a computer vision model to process the drawings in the unified intermediate format, identify and determine the functional areas in the drawings and the drawing data within the areas, and construct natural language description information describing the drawing elements and their attributes based on the identification results. The review execution module is configured to associate the natural language description information with the rules in the rule base and knowledge base, use the reasoning ability of the large language model to make compliance judgments on each element in the drawing, and generate a review report containing specific violations and location information.

7. The system according to claim 6, characterized in that, The rule base and knowledge base construction module is further configured as follows: The engineering design specification document is input into the large language model in segments, and the extracted rules are output by the large language model in a predetermined structured data format through prompt words.

8. The system according to claim 6, characterized in that, The drawing parsing and semantic understanding module includes: The region identification unit is configured to use an object detection model to identify functional regions in the drawing; The primitive recognition unit is configured to use a primitive recognition model to recognize basic primitives within the functional area; The description generation unit is configured to generate the natural language description information based on the results of region recognition and primitive recognition.

9. The system according to claim 6, characterized in that, The review execution module is further configured as follows: The natural language description information and the rules in the knowledge base are input into the large language model; Receive the reasoning results output by the large language model, the results including compliance judgment conclusions and reasoning logic; Based on the reasoning results, a final human-readable review report is generated.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 5.

Images