Design drawing auditing method and device and electronic equipment
By combining neural network models and vertical domain large models with knowledge bases to automate the processing of design drawings, the problems of low efficiency and low accuracy in existing technologies have been solved, achieving efficient and accurate drawing review.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-24
- Publication Date
- 2026-03-13
AI Technical Summary
Existing methods for reviewing design drawings are inefficient and inaccurate, rely on manual review, and have large errors in information extraction.
By acquiring design drawings, preprocessing them, and then inputting them into a neural network model to extract structured BOM data, a vertical domain large model combined with a knowledge base is used for review to generate a structured review report.
It improves the efficiency and accuracy of design drawing review, and can automatically process drawing information and generate detailed review reports.
Smart Images
Figure CN121661665A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, and electronic device for reviewing design drawings. Background Technology
[0002] In the core chain of engineering practice, design drawings serve as a crucial carrier connecting project planning, construction, and final acceptance. The quality of their review directly determines the project's safety redundancy, cost control accuracy, and overall implementation efficiency. Especially in scenarios such as communication base station deployment, large-scale building structure design, and precision mechanical assembly, a single drawing often contains the topological relationships and parameter annotations of thousands of heterogeneous components. Currently, existing design drawing review methods employ a fragmented review model, relying on manual labor, resulting in low efficiency, large information extraction errors, and low accuracy. Summary of the Invention
[0003] This invention provides a design drawing review method, apparatus, and electronic device to address the shortcomings of existing design drawing review methods, namely low efficiency and low accuracy.
[0004] This invention provides a method for reviewing design drawings, comprising: Obtain the design drawings pending review; The design drawings are processed to obtain the structured bill of materials (BOM) data corresponding to the design drawings; The structured BOM data corresponding to the design drawings is input into a pre-built vertical domain model. The vertical domain model, based on a pre-built knowledge base, reviews the structured BOM data and outputs the review result of the design drawings. The vertical domain model is trained based on sample structured BOM data corresponding to sample design drawings and the review result labels of the sample design drawings. Based on the audit results, a structured audit report is generated.
[0005] In some embodiments, processing the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings includes: The design drawings are preprocessed to obtain the design image; The design image is input into a pre-built neural network model to obtain the structured BOM data corresponding to the design image output by the neural network model. The structured BOM data includes component codes, technical parameters, and assembly relationships between components. The neural network model is trained based on the sample design image and the corresponding structured BOM data label.
[0006] In some embodiments, before inputting the structured BOM data corresponding to the design drawings into the pre-built vertical domain large model, the method further includes: Determine the target audit dimensions for the structured BOM data; Based on the target review dimension, a target sub-review agent is determined from multiple sub-review agents, and the vertical domain large model is invoked based on the target sub-review agent.
[0007] In some embodiments, determining the target audit dimension of the structured BOM data includes: The structured BOM data is matched with multiple preset regular expressions to obtain matching results; Semantic recognition is performed on the structured BOM data to obtain the contextual semantic information of the structured BOM data; Based on the matching results and the contextual semantic information, the target audit dimension of the structured BOM data is determined.
[0008] In some embodiments, the knowledge base includes at least one of a specification knowledge base, a component knowledge base, and a case knowledge base; the specification knowledge base contains industry standard clauses, the component knowledge base contains component compatibility information, and the case knowledge base contains historical design defect cases.
[0009] In some embodiments, the training process of the vertical domain large model includes: Obtain sample design drawings; The sample design drawings are processed to obtain the sample structured BOM data corresponding to the sample design drawings; Determine the review result label for the sample design drawings; Using the sample structured BOM data corresponding to the sample design drawings as training samples and the review result tags of the sample design drawings as sample tags, an initial vertical domain large model is trained. After training, the vertical domain large model is obtained.
[0010] In some embodiments, training the initial vertical domain large model includes: Freeze the basic parameters of the initial vertical domain large model; Based on the knowledge base, a low-rank adapter for the initial vertical domain large model is trained, and the parameters of the low-rank adapter are iteratively optimized.
[0011] In some embodiments, the training process of the neural network model includes: Obtain sample design drawings; The sample design drawings are preprocessed to obtain sample design images; Determine the structured BOM data label corresponding to the sample design image; The sample design image is input into the initial neural network model to obtain the structured BOM prediction data output by the initial neural network model; Based on the structured BOM prediction data and the structured BOM data labels, a loss function value is calculated. Based on the loss function value, the parameters of the initial neural network model are iteratively optimized to obtain the neural network model.
[0012] The present invention also provides a design drawing review device, comprising: The acquisition unit is used to acquire the design drawings to be reviewed. The processing unit is used to process the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings; The review unit is used to input the structured BOM data corresponding to the design drawings into a pre-built vertical domain model. The vertical domain model reviews the structured BOM data based on a pre-built knowledge base and outputs the review result of the design drawings. The vertical domain model is trained based on sample structured BOM data corresponding to sample design drawings and the review result labels of the sample design drawings. The generation unit is used to generate a structured audit report based on the audit results.
[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the design drawing review method described above.
[0014] The design drawing review method, apparatus, and electronic device provided by this invention acquire the design drawings to be reviewed; process the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings; input the structured BOM data corresponding to the design drawings into a pre-built vertical domain model; the vertical domain model, based on a pre-built knowledge base, reviews the structured BOM data and outputs the review results of the design drawings; based on the review results, a structured review report is generated, thereby improving the efficiency and accuracy of drawing review. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0016] Figure 1This is a flowchart illustrating the design drawing review method provided in this embodiment of the invention.
[0017] Figure 2 This is a schematic diagram of the process for processing design drawings provided in an embodiment of the present invention.
[0018] Figure 3 This is a flowchart illustrating the training process of a large vertical model provided in an embodiment of the present invention.
[0019] Figure 4 This is a flowchart illustrating the training process of the neural network model provided in this embodiment of the invention.
[0020] Figure 5 This is a schematic diagram of the design drawing review device provided in an embodiment of the present invention.
[0021] Figure 6 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0023] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, the first object can be one or more. Furthermore, in this invention, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0024] Figure 1 This is a flowchart illustrating the design drawing review method provided in an embodiment of the present invention. Figure 1 As shown, a method for reviewing design drawings is provided, including the following steps: step 110, step 120, step 130, and step 140. This method's steps are merely one possible implementation of the invention.
[0025] Step 110: Obtain the design drawings to be reviewed.
[0026] The design drawings to be reviewed refer to original digital files created by designers using professional computer-aided design (CAD) software, containing complete technical information about the product or engineering project. The design drawings can be in the industry-standard vector graphics format. They contain various elements describing the design scheme.
[0027] For example, communication design drawings contain graphic symbols for equipment such as base stations, antennas, and feeders, as well as key textual parameters and dimensional annotations such as antenna gain, feeder loss, and material specifications.
[0028] Step 120: Process the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings.
[0029] The structured BOM data includes multiple fields and their confidence scores, with information for each component being broken down into fixed fields.
[0030] Optionally, the graphic and textual information on the design drawings can be extracted and organized into a database record with a uniform format, clear fields, and direct computer analysis and review capability.
[0031] Figure 2 This is a schematic diagram illustrating the process of processing design drawings according to an embodiment of the present invention. Figure 2 As shown, in some embodiments, step 120 processes the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings, including: Step 121: Preprocess the design drawings to obtain the design image; Optionally, a CAD file parsing engine can be invoked to convert the design drawings and obtain the design image.
[0032] Optionally, the design image can be preprocessed, such as resizing and normalizing.
[0033] Step 122: Input the design image into the pre-built neural network model to obtain the structured BOM data corresponding to the design image output by the neural network model. The structured BOM data includes component codes, technical parameters and assembly relationships between components. The neural network model is trained based on the sample design image and the corresponding structured BOM data label.
[0034] The neural network models include Residual Network (ResNet) and Convolutional Recurrent Neural Network (CRNN); ResNet is used to extract features from the design image, and CRNN is used to recognize text in the design image.
[0035] In some embodiments, before inputting the structured BOM data corresponding to the design drawings into the pre-built vertical domain large model, the method further includes: Determine the target audit dimensions for structured BOM data.
[0036] Optionally, the target audit dimension can be determined from a number of preset audit dimensions, such as electrical safety dimension and material cost dimension.
[0037] In some embodiments, determining the target audit dimensions for structured BOM data includes: The structured BOM data is matched against multiple preset regular expressions to obtain the matching results; Semantic recognition is performed on structured BOM data to obtain the contextual semantic information of the structured BOM data; Based on the matching results and contextual semantic information, the target audit dimensions of the structured BOM data are determined.
[0038] Optionally, the structured BOM data is parsed to obtain the parsed field data; the target field data is matched with multiple regular expressions to determine the matching target regular expression; and based on the target regular expression, the component information, such as the component model, is determined.
[0039] Optionally, semantic recognition can be performed on structured BOM data based on existing semantic recognition models.
[0040] Based on the target review dimension, the target sub-review agent is determined from multiple sub-review agents, and the vertical domain large model is invoked based on the target sub-review agent.
[0041] Optionally, each auditing agent focuses on a specific auditing dimension. For example, the electrical safety auditing agent is specifically responsible for checking fields related to electrical codes and electrical safety; the material cost auditing agent is specifically responsible for analyzing the rules related to component selection, quantity, and cost control in the BOM; the structural compliance auditing agent is specifically responsible for auditing whether the design complies with structural specifications such as physical installation and spatial layout; and the signal performance auditing agent is specifically responsible for analyzing whether communication parameters meet performance indicators. When a new auditing dimension needs to be added, only a new sub-auditing agent and corresponding routing rules need to be added, making it easy to expand.
[0042] Among them, the vertical domain big model refers to an AI big language model that has been deeply optimized with knowledge of a specific industry or professional field; the vertical domain big model can understand context, semantics and implicit logical relationships.
[0043] Step 130: Input the structured BOM data corresponding to the design drawings into the pre-built vertical domain model. The vertical domain model, based on the pre-built knowledge base, reviews the structured BOM data and outputs the review results of the design drawings. The vertical domain model is trained based on the sample structured BOM data corresponding to the sample design drawings and the review result labels of the sample design drawings.
[0044] Among them, the knowledge base is an external reference library that the large vertical model relies on when making inferences and judgments.
[0045] In some embodiments, the knowledge base includes at least one of a specification knowledge base, a component knowledge base, and a case knowledge base; the specification knowledge base contains industry standard clauses, the component knowledge base contains component compatibility information, and the case knowledge base contains historical cases of design defects.
[0046] Optionally, the specification knowledge base stores a massive amount of industry specifications, national standards, safety regulations, and other clauses in a structured manner. For example, the horizontal distance between the base station antenna and surrounding buildings should be greater than or equal to 5m. The component knowledge base records detailed parameters, performance indicators, prices, supplier information, and compatibility matrices of various communication components. The case knowledge base contains historical cases with design defects. For each case, the problem description, error cause, rectification plan, and comparison data before and after rectification are structured and labeled.
[0047] Optionally, the audit results may include, but are not limited to, the following: The issue list clearly identifies non-compliant, unreasonable, or risky items in the design; Problem Level: Determines the severity level of each problem; According to the explanation: For each question, the basis for judgment is clearly cited, such as a clause in the specification knowledge base, a compatibility rule in the component knowledge base, or a similar historical error in the case knowledge base; Suggested modifications: Specific and feasible modification solutions are provided for the identified issues.
[0048] It should be noted that the vertical domain large model can utilize its internal multi-head attention mechanism to simultaneously focus on and associate knowledge from different sources to perform complex logical reasoning; it can not only check whether individual components are compliant, but also analyze whether the combination of components is reasonable and whether it is similar to historical error patterns.
[0049] Step 140: Based on the audit results, generate a structured audit report.
[0050] Optionally, the audit report can be encrypted to generate a encrypted audit report.
[0051] In this embodiment of the invention, the following steps are taken: First, the design drawings to be reviewed are obtained. Then, the design drawings are processed to obtain the structured Bill of Materials (BOM) data corresponding to the design drawings. Next, the structured BOM data corresponding to the design drawings is input into a pre-built vertical domain model. Based on a pre-built knowledge base, the vertical domain model reviews the structured BOM data and outputs the review results for the design drawings. Finally, a structured review report is generated based on the review results, thereby improving the efficiency and accuracy of drawing review.
[0052] Figure 3 This is a flowchart illustrating the training process of a large vertical domain model provided in an embodiment of the present invention. Figure 3 As shown, in some embodiments, the training process of the large vertical model includes: Step 310: Obtain sample design drawings; Step 320: Process the sample design drawings to obtain the sample structured BOM data corresponding to the sample design drawings; Step 330: Determine the review result label for the sample design drawings; Step 340: Using the sample structured BOM data corresponding to the sample design drawings as training samples and the review result labels of the sample design drawings as sample labels, train the initial vertical domain large model. After training, the vertical domain large model is obtained.
[0053] Optionally, training the initial large vertical model includes: The sample structured BOM data corresponding to the sample design drawings is input into the initial vertical domain large model. The initial vertical domain large model, based on the pre-built knowledge base, reviews the sample structured BOM data and outputs the predicted review results of the sample design drawings.
[0054] Based on the predicted review results and the review result labels, the reward function value is calculated. Based on the reward function value, the parameters of the initial vertical domain large model are iteratively optimized to obtain the vertical domain large model.
[0055] The knowledge base includes at least one of the following: a specification knowledge base, a component knowledge base, and a case knowledge base.
[0056] In some embodiments, training an initial large vertical model includes: Freeze the basic parameters of the initial vertical domain large model; Based on the knowledge base, a low-rank adapter for the initial large vertical model is trained, and the parameters of the low-rank adapter are iteratively optimized.
[0057] Low-rank adapters are a technique for fine-tuning large pre-trained models. The core idea is to adapt the model to new tasks by injecting and training only small, trainable adapter modules into the model without changing any parameters of the original large model.
[0058] Optionally, the basic parameters of the initial vertical domain large model are set to an untrainable state, and the values of these parameters will not be updated by the gradient descent algorithm during the training of the initial vertical domain large model; trainable low-rank adapters are dynamically injected into specific layers of the initial vertical domain large model.
[0059] In this embodiment of the invention, by freezing the basic parameters of the initial vertical domain large model, training a low-rank adapter of the initial vertical domain large model based on a knowledge base, and iteratively optimizing the parameters of the low-rank adapter, the training efficiency is improved, the training cost is reduced, and the reusability and deployment flexibility of the vertical domain large model are greatly improved.
[0060] Figure 4 This is a flowchart illustrating the training process of the neural network model provided in an embodiment of the present invention. Figure 4 As shown, in some embodiments, the training process of the neural network model includes: Step 410: Obtain sample design drawings; Step 420: Preprocess the sample design drawings to obtain the sample design image; Step 430: Determine the structured BOM data label corresponding to the sample design image; Step 440: Input the sample design image into the initial neural network model to obtain the structured BOM prediction data output by the initial neural network model; Step 450: Based on the structured BOM prediction data and structured BOM data labels, calculate the loss function value. Based on the loss function value, iteratively optimize the parameters of the initial neural network model to obtain the neural network model.
[0061] The sample design drawings cover different design styles, different project types, and different CAD software versions.
[0062] Optionally, the sample design drawings are preprocessed by format conversion, normalization, data augmentation, etc., to obtain sample design images; the sample design images are then subjected to data augmentation processing such as inversion, blurring, and mirroring; and the amplified sample design images are divided to obtain training sets, validation sets, and test sets.
[0063] Optionally, a semi-supervised learning method can be used to annotate the sample design images. First, some sample design images are manually annotated, and then the remaining sample design images are automatically annotated using pseudo-labeling technology.
[0064] The initial neural network models include ResNet and CRNN; ResNet is used to extract features from the sample design image, and CRNN is used to recognize text in the sample design image.
[0065] Optionally, based on the structured BOM prediction data and the structured BOM data labels, the cross-entropy loss function value is calculated, and the connectionist temporal classification (CTC) loss function value is calculated; the weights of the cross-entropy loss function value and the CTC loss function value are determined; based on the cross-entropy loss function value and the CTC loss function value, as well as the weights of the cross-entropy loss function value and the CTC loss function value, the joint loss function value is calculated; based on the joint loss function value, the parameters of the initial neural network model are iteratively optimized to obtain the neural network model.
[0066] Optionally, the performance of the neural network model can be evaluated using metrics such as precision and recall on the validation and test sets during and after training. If the performance of the neural network model is unsatisfactory, the data augmentation strategy, model structure, or hyperparameters can be adjusted, and the neural network model can be retrained.
[0067] Optionally, a transfer learning strategy can be adopted to freeze the first N convolutional layers of the initial neural network model, where N is a natural number greater than or equal to 1.
[0068] In this embodiment of the invention, by inputting the sample design image into the initial neural network model, the structured BOM prediction data output by the initial neural network model is obtained; based on the structured BOM prediction data and the structured BOM data labels, the loss function value is calculated; based on the loss function value, the parameters of the initial neural network model are iteratively optimized to obtain the neural network model, thereby improving the data extraction capability of the neural network model.
[0069] The design drawing review device provided in the embodiments of the present invention will be described below. The design drawing review device described below can be referred to in correspondence with the design drawing review method described above.
[0070] Figure 5 This is a schematic diagram of the design drawing review device provided in an embodiment of the present invention, such as... Figure 5 As shown, the design drawing review device 500 includes: Acquisition unit 510 is used to acquire design drawings to be reviewed; The processing unit 520 is used to process the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings. The review unit 530 is used to input the structured BOM data corresponding to the design drawings into the pre-built vertical domain big model. The vertical domain big model reviews the structured BOM data based on the pre-built knowledge base and outputs the review results of the design drawings. The vertical domain big model is trained based on the sample structured BOM data corresponding to the sample design drawings and the review result labels of the sample design drawings. Generation unit 540 is used to generate a structured audit report based on the audit results.
[0071] Optionally, the design drawings are processed to obtain the structured bill of materials (BOM) data corresponding to the design drawings, including: Preprocess the design drawings to obtain the design image; The design image is input into a pre-built neural network model to obtain the structured BOM data corresponding to the design image output by the neural network model. The structured BOM data includes component codes, technical parameters and assembly relationships between components. The neural network model is trained based on the sample design image and the corresponding structured BOM data label.
[0072] Optionally, the design drawing review device 500 also includes: The first determining unit is used to determine the target audit dimensions of the structured BOM data; The second determining unit is used to determine the target sub-review agent from multiple sub-review agents based on the target review dimension, and to call the vertical domain large model based on the target sub-review agent.
[0073] Optionally, the target audit dimensions for the structured BOM data can be determined, including: The structured BOM data is matched against multiple preset regular expressions to obtain the matching results; Semantic recognition is performed on structured BOM data to obtain the contextual semantic information of the structured BOM data; Based on the matching results and contextual semantic information, the target audit dimensions of the structured BOM data are determined.
[0074] Optionally, the knowledge base includes at least one of a specification knowledge base, a component knowledge base, and a case knowledge base; the specification knowledge base contains industry standard clauses, the component knowledge base contains component compatibility information, and the case knowledge base contains historical design defect cases.
[0075] Optionally, the training process for a large vertical model includes: Obtain sample design drawings; The sample design drawings are processed to obtain the sample structured BOM data corresponding to the sample design drawings; Determine the review result label for the sample design drawings; Using the sample structured BOM data corresponding to the sample design drawings as training samples and the review result tags of the sample design drawings as sample tags, an initial vertical domain large model is trained. After training, the vertical domain large model is obtained.
[0076] Optionally, train an initial large vertical model, including: Freeze the basic parameters of the initial vertical domain large model; Based on the knowledge base, a low-rank adapter for the initial large vertical model is trained, and the parameters of the low-rank adapter are iteratively optimized.
[0077] Optionally, the training process of a neural network model includes: Obtain sample design drawings; Preprocess the sample design drawings to obtain the sample design image; Determine the structured BOM data label corresponding to the sample design image; The sample design image is input into the initial neural network model to obtain the structured BOM prediction data output by the initial neural network model; Based on the structured BOM prediction data and structured BOM data labels, the loss function value is calculated. Based on the loss function value, the parameters of the initial neural network model are iteratively optimized to obtain the neural network model.
[0078] It should be noted that the design drawing review device provided in this embodiment of the invention can implement all the method steps implemented in the above-mentioned design drawing review method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.
[0079] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 6As shown, the electronic device may include a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a design drawing review method, which includes: acquiring the design drawing to be reviewed; processing the design drawing to obtain the structured bill of materials (BOM) data corresponding to the design drawing; inputting the structured BOM data corresponding to the design drawing into a pre-built vertical domain model; the vertical domain model, based on a pre-built knowledge base, reviews the structured BOM data and outputs the review result of the design drawing; the vertical domain model is trained based on sample structured BOM data corresponding to sample design drawings and the review result labels of sample design drawings; and generating a structured review report based on the review result.
[0080] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for reviewing design drawings, characterized in that, include: Obtain the design drawings pending review; The design drawings are processed to obtain the structured bill of materials (BOM) data corresponding to the design drawings; The structured BOM data corresponding to the design drawings is input into a pre-built vertical domain model. The vertical domain model reviews the structured BOM data based on a pre-built knowledge base and outputs the review result of the design drawings. The vertical domain large model is trained based on the sample structured BOM data corresponding to the sample design drawings and the review result labels of the sample design drawings. Based on the audit results, a structured audit report is generated.
2. The design drawing review method according to claim 1, characterized in that, The process of processing the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings includes: The design drawings are preprocessed to obtain the design image; The design image is input into a pre-built neural network model to obtain the structured BOM data corresponding to the design image output by the neural network model. The structured BOM data includes component codes, technical parameters, and assembly relationships between components. The neural network model is trained based on the sample design image and the corresponding structured BOM data label.
3. The design drawing review method according to claim 1, characterized in that, Before inputting the structured BOM data corresponding to the design drawings into the pre-built vertical domain large model, the process also includes: Determine the target audit dimensions for the structured BOM data; Based on the target review dimension, a target sub-review agent is determined from multiple sub-review agents, and the vertical domain large model is invoked based on the target sub-review agent.
4. The design drawing review method according to claim 3, characterized in that, Determining the target audit dimensions for the structured BOM data includes: The structured BOM data is matched with multiple preset regular expressions to obtain matching results; Semantic recognition is performed on the structured BOM data to obtain the contextual semantic information of the structured BOM data; Based on the matching results and the contextual semantic information, the target audit dimension of the structured BOM data is determined.
5. The design drawing review method according to claim 1, characterized in that, The knowledge base includes at least one of a specification knowledge base, a component knowledge base, and a case knowledge base; the specification knowledge base contains industry standard clauses, the component knowledge base contains component compatibility information, and the case knowledge base contains historical design defect cases.
6. The design drawing review method according to claim 1, characterized in that, The training process of the large vertical model includes: Obtain sample design drawings; The sample design drawings are processed to obtain the sample structured BOM data corresponding to the sample design drawings; Determine the review result label for the sample design drawings; Using the sample structured BOM data corresponding to the sample design drawings as training samples and the review result tags of the sample design drawings as sample tags, an initial vertical domain large model is trained. After training, the vertical domain large model is obtained.
7. The design drawing review method according to claim 6, characterized in that, The initial vertical domain large model for training includes: Freeze the basic parameters of the initial vertical domain large model; Based on the knowledge base, a low-rank adapter for the initial vertical domain large model is trained, and the parameters of the low-rank adapter are iteratively optimized.
8. The design drawing review method according to claim 2, characterized in that, The training process of the neural network model includes: Obtain sample design drawings; The sample design drawings are preprocessed to obtain sample design images; Determine the structured BOM data label corresponding to the sample design image; The sample design image is input into the initial neural network model to obtain the structured BOM prediction data output by the initial neural network model; Based on the structured BOM prediction data and the structured BOM data labels, a loss function value is calculated. Based on the loss function value, the parameters of the initial neural network model are iteratively optimized to obtain the neural network model.
9. A design drawing review device, characterized in that, include: The acquisition unit is used to acquire the design drawings to be reviewed. The processing unit is used to process the design drawings to obtain the structured bill of materials (BOM) data corresponding to the design drawings; The review unit is used to input the structured BOM data corresponding to the design drawings into a pre-built vertical domain model. The vertical domain model reviews the structured BOM data based on a pre-built knowledge base and outputs the review result of the design drawings. The vertical domain large model is trained based on the sample structured BOM data corresponding to the sample design drawings and the review result labels of the sample design drawings. The generation unit is used to generate a structured audit report based on the audit results.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the design drawing review method as described in any one of claims 1 to 8.