An industrial CAD drawing method, apparatus, storage medium and device

By performing geometric preprocessing on the 3D boundary representation model and generating visibility facts, the problems of view redundancy and missing data in traditional CAD drawing schemes are solved, and high-precision 2D engineering drawing generation is achieved.

CN123069477APending Publication Date: 2026-11-10SHANGHAI SHEXU TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611578189.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional CAD drawing schemes rely on rule templates, which make it difficult to exhaustively cover all engineering scenarios, resulting in view redundancy or missing views. Furthermore, LLM-assisted schemes lack geometric accuracy guarantees.

Method used

By performing geometric preprocessing on the 3D boundary representation model, visibility facts and expression gap facts are generated as structured inputs for the map completion and cross-sectional decision-making agents. The six-view planner and large language model are used to identify the scene business coordinate system, ensuring the integrity and geometric accuracy of the generated views.

Benefits of technology

The generated two-dimensional engineering drawings can completely and non-redundantly represent the features of the parts, have high geometric accuracy, and meet the traceability requirements of industrial quality inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069477A_ABST
    Figure CN123069477A_ABST
Patent Text Reader

Abstract

This application provides an industrial CAD drawing method, apparatus, storage medium, and device. In this method, a three-dimensional boundary representation model undergoes geometric preprocessing to reduce redundant geometric details. A six-view planner is used to evaluate the visibility of each instance under six standard viewpoints, generating visibility facts. These visibility facts are input into a supplementary drawing decision agent to obtain a set of standard orthogonal views to be supplemented, and a two-dimensional engineering view is generated accordingly. Then, based on the two-dimensional engineering view and the geometric preprocessing results, facts expressing gaps are generated and input into a sectioning decision agent to obtain a sectioning plan, which is then used to generate a sectioning view. In this way, two types of lightweight facts—visibility facts and gap-expressing facts—are constructed and used as structured inputs to the supplementary drawing decision agent and the sectioning decision agent, respectively. This ensures that the final generated two-dimensional engineering drawing can completely and non-redundantly express the part features and has high geometric accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of industrial CAD technology, and more specifically, to an industrial CAD drawing method, apparatus, storage medium, and device. Background Technology

[0002] In the industrial CAD (Computer-Aided Design) workflow, converting a 3D boundary representation (BREP) model into 2D engineering drawings conforming to national or ISO standards is a crucial step in the product's transition from design to manufacturing. This process is commonly referred to as CAD drawing projection. Traditional CAD drawing projection solutions primarily rely on rule templates, which solidify engineering drawing representation specifications into enumerable templates. A rule table is pre-written by the operator for each feature type, specifying "which views should be generated, whether sectional views are needed, and how sectional views should be placed." During runtime, the scheduler executes according to this rule table. Because the rule table cannot exhaustively cover all engineering scenarios, the system cannot make reasonable decisions when encountering feature types not covered by the rule table, easily leading to view redundancy or missing views.

[0003] As LLM (Large Language Model) capabilities have improved, the industry has begun to incorporate LLM into industrial software to assist engineering operations. Users directly input the geometric data of the BREP model into the LLM, which responds to the user's interactive requests. After user confirmation, the response is sent to the industrial software platform to execute the corresponding engineering operations. However, while pure LLM solutions offer semantic flexibility, they lack guarantees of geometric accuracy. Summary of the Invention

[0004] The purpose of this application is to provide an industrial CAD drawing method, apparatus, storage medium and device, which aims to solve the problems of view redundancy or view missingness, or lack of geometric accuracy guarantee in the industrial CAD drawing methods of related technologies.

[0005] Firstly, this application provides an industrial CAD projection method, comprising: preprocessing input data of a three-dimensional boundary representation model to obtain geometric preprocessing results; wherein the geometric preprocessing results include the geometric primitives and topological relationships of the three-dimensional boundary representation model, instances of each data object that can be used for projection, and semantic relationships between instances; determining a scene business coordinate system, generating visibility facts using a six-view planner and the geometric preprocessing results, and inputting the visibility facts into a supplementary mapping decision agent to obtain supplementary mapping decision results; wherein the six-view planner calculates the six coordinates of each instance in the scene business coordinate system. The visibility of each target to be expressed is evaluated in the standard view direction; the visibility facts represent the coverage status of each target to be expressed in the standard view direction and the candidate view direction information of the uncovered features; the supplementary map decision result represents the set of standard orthogonal views to be supplemented; based on the supplementary map decision result, a two-dimensional engineering view is generated, and based on the two-dimensional engineering view and the geometric preprocessing result, expression gap facts are generated, and then the expression gap facts are input into the cross-section decision agent to obtain the cross-section plan, and a cross-section view is generated based on the cross-section plan; wherein, the expression gap facts represent the missing dimension information of each target to be expressed that is not clearly expressed in the six standard view directions.

[0006] In the above implementation process, the 3D boundary representation model undergoes geometric preprocessing to reduce redundant geometric details. A six-view planner is used to evaluate the visibility of each instance under six standard view directions, generating visibility facts. These visibility facts are input into the map completion decision agent to obtain a set of standard orthogonal views to be completed, and a 2D engineering view is generated accordingly. Then, based on the 2D engineering view and the geometric preprocessing results, gap expression facts are generated and input into the sectioning decision agent to obtain a sectioning plan, which is then used to generate a sectioning view. In this way, two types of lightweight facts, visibility facts and gap expression facts, are constructed and used as structured inputs to the map completion decision agent and the sectioning decision agent, respectively. This ensures that the final generated 2D engineering drawing can completely and non-redundantly express the part features and has high geometric accuracy.

[0007] Furthermore, in some examples, determining the scene business coordinate system includes: determining the local coordinate system corresponding to each instance using a file local coordinate agent according to a preset coordinate candidate strategy; and merging the local coordinate systems using the scene business coordinate agent to generate a unified scene business coordinate system.

[0008] In the above implementation process, a specific method for determining the scene business coordinate system is provided.

[0009] Furthermore, in some examples, determining the scene service coordinate system further includes: when the local coordinate system corresponding to any instance cannot be determined by the preset coordinate candidate strategy, or when the fusion of each local coordinate system fails, using a gaze intent agent to determine the scene service coordinate system; wherein, the gaze intent agent calls a geometric kernel and a large language model to identify the gaze intent of the scene service.

[0010] In the above implementation process, the fallback mechanism of the gaze intent agent is used to ensure that a scene business coordinate system that conforms to the engineering semantics can be generated under any assembly form, thereby ensuring that the mapping process is not interrupted.

[0011] Further, in some examples, the visibility facts include: an existing view type field, a candidate view type field, a fully expressed target field, an uncovered target field, and a candidate fully expressed target field divided by view direction; the candidate view type field includes a sequence of candidate standard view directions according to a predetermined priority; generating visibility facts using a six-view planner and the geometric preprocessing results includes: identifying the target to be expressed from the geometric preprocessing results using a six-view planner; projecting each target to be expressed according to six standard view directions, determining whether the key dimension information of the target to be expressed is visible and fully expressed in each standard view direction; recording the fully expressed target field, the uncovered target field, and the candidate fully expressed target field according to the determination results; selecting the view that covers the most uncovered projection instances using a greedy strategy and recording it in the existing view type field; encapsulating the recorded information according to the data structure of the visibility facts to generate the visibility facts.

[0012] In the above implementation process, a specific method for generating visible facts is provided.

[0013] Further, in some examples, the step of inputting the visibility facts into the supplementary graph decision agent to obtain the supplementary graph decision result includes: using the supplementary graph decision agent, sequentially traversing each candidate view according to a predetermined priority in the candidate standard view sequence; if the currently traversed candidate view does not belong to the currently planned standard view set, and the intersection of the coverage contribution record of the candidate view and the uncovered target set is not empty, then the candidate view is added to the new standard view set, and the targets contained in the intersection are removed from the uncovered target set; if the uncovered target set is empty, the traversal is stopped, and the current new standard view set is taken as the standard orthogonal view set to be supplemented; wherein, the currently planned standard view set comes from the existing view type field in the visibility facts, the coverage contribution record comes from the candidate fully expressible target field divided by view in the visibility facts, and the uncovered target set comes from the uncovered target field in the visibility facts.

[0014] In the above implementation process, a decision logic is provided for the supplementary graph decision agent to ensure that each supplementary view can solve at least one uncovered target to be expressed, avoid supplementing invalid views, and avoid repeatedly supplementing existing views.

[0015] Furthermore, in some examples, the expression gap facts include: an existing standard view field, an unresolved feature field, a merge candidate field, and a grouping candidate field; generating expression gap facts based on the two-dimensional engineering view and the geometric preprocessing results includes: determining an existing standard view based on the two-dimensional engineering view and recording it in the existing standard view field; identifying the target to be expressed that is still not clearly expressed under the existing standard view as an unresolved feature, identifying its missing dimension type and recording it in the unresolved feature field; merging and marking multiple unresolved features located on the same cutting plane, generating merge candidate groups and recording them in the merge candidate field; grouping each unresolved feature according to the cutting plane, and recording the grouping results in the grouping candidate field.

[0016] In the above implementation process, a specific method for generating expression gap facts is provided.

[0017] Furthermore, in some examples, the profile plan corresponds to a grouping key; the grouping key is based on the strategy name and missing dimension type of the unresolved feature corresponding to the profile plan and the hash function value of the existing standard view; generating a profile view based on the profile plan includes: obtaining the grouping keys corresponding to all profile plans; merging profile plans with the same grouping key; performing a sectioning operation based on the merged profile plan to generate a profile view; wherein, multiple profile plans with the same grouping key correspond to the same profile view.

[0018] In the above implementation process, a hash operation is performed based on the strategy name of the unresolved feature, the current view set, and the expression gap dimension set to generate a grouping key corresponding to each profile plan. Multiple profile plans sharing the same grouping key will be merged into a single profile view downstream. In this way, the number of profile views is effectively reduced, avoiding redundancy in generating profile views repeatedly for the same type of feature.

[0019] Secondly, this application provides an industrial CAD projection device, comprising: a preprocessing module for preprocessing input data of a three-dimensional boundary representation model to obtain geometric preprocessing results; wherein the geometric preprocessing results include the geometric primitives and topological relationships of the three-dimensional boundary representation model, instances of each data object that can be used for projection, and semantic relationships between instances; and a supplementary projection decision module for determining the scene business coordinate system, generating visibility facts using a six-view planner and the geometric preprocessing results, and inputting the visibility facts into a supplementary projection decision agent to obtain supplementary projection decision results; wherein the six-view planner performs calculations on each instance in the scene business coordinate system. The visibility of each target to be expressed is evaluated in six standard view directions; the visibility facts represent the coverage status of each target to be expressed in the standard view directions and the candidate view direction information of the uncovered features; the supplementary map decision result represents the set of standard orthogonal views to be supplemented; the sectioning decision module is used to generate a two-dimensional engineering view based on the supplementary map decision result, and generate expression gap facts based on the two-dimensional engineering view and the geometric preprocessing result, and then input the expression gap facts into the sectioning decision agent to obtain a sectioning plan, and generate a section view based on the sectioning plan; wherein, the expression gap facts represent the missing dimension information of each target to be expressed that is not clearly expressed in the six standard view directions.

[0020] Thirdly, this application provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method described in any of the first aspects.

[0021] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described in any of the first aspects.

[0022] Fifthly, this application provides a computer program product that, when run on a computer, causes the computer to perform the method described in any of the first aspects.

[0023] Other features and advantages disclosed in this application will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the above-described technology disclosed in this application.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating an industrial CAD drawing method provided in this application embodiment; Figure 2 A schematic diagram of the overall architecture of a view agent decision-making system provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the workflow of a view agent decision-making system provided in this application embodiment; Figure 4 A block diagram of an industrial CAD projection device provided in an embodiment of this application; Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0028] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] In the industrial CAD design process, converting the 3D Boundary Representation (BREP) model into 2D engineering drawings conforming to national or ISO standards is a crucial step in the product's transition from design to manufacturing. This process is commonly referred to as CAD drawing projection. Traditional CAD drawing projection solutions primarily rely on rule templates, which solidify engineering drawing representation specifications into enumerable templates. A rule table is manually pre-written for each feature type, specifying "which views should be generated, whether sectional views are needed, and how sectional views should be placed." During runtime, the scheduler executes according to this rule table. Because the rule table cannot exhaustively cover all engineering scenarios, the system cannot make reasonable decisions when encountering feature types not covered by the rule table, easily leading to view redundancy or missing views. With the increasing capabilities of Large Language Models (LLMs), the industry has begun to introduce LLMs into industrial software to assist engineering operations. Users directly input the geometric data of the BREP model into the LLM, which responds to user interaction requests. After user confirmation, the response is sent to the industrial software platform to execute the corresponding engineering operations. However, LLMs lack geometric accuracy guarantees, making it difficult to meet the traceability requirements of industrial quality inspection.

[0030] To address the aforementioned issues, this application provides an industrial CAD drawing scheme. By performing geometric preprocessing on the 3D boundary representation model, the original geometry is compressed into lightweight facts, retaining only the "feature-view-expression sufficiency" information truly necessary for decision-making. This avoids the "geometric illusion" caused by LLM processing massive amounts of geometric data. Simultaneously, two types of lightweight facts, visibility facts and expression gap facts, are constructed as structured inputs to the map completion decision agent and the sectioning decision agent, respectively. This transforms the semantic question of whether map completion or sectioning is needed into structured facts that can be directly consumed by the agents. As a result, the final generated 2D engineering drawing can completely and non-redundantly express the features of the part, and has guaranteed geometric accuracy.

[0031] The embodiments of this application will be described below: like Figure 1 As shown, Figure 1 This is a flowchart of an industrial CAD projection method provided in an embodiment of this application. The method can be implemented as an industrial projection system; for ease of explanation, the executing entity will be referred to as the system below. The method includes: Step 101: Preprocess the input data of the 3D boundary representation model to obtain geometric preprocessing results; wherein, the geometric preprocessing results include the geometric primitives and topological relationships of the 3D boundary representation model, the instances of each data object that can be used for mapping, and the semantic relationships between instances; The input data for the 3D boundary representation model mentioned in this step can be a STEP (Standard for the Exchange of Product Data) format file, which records the original geometric data of the 3D BREP model. By preprocessing the input data of the 3D boundary representation model, the output geometric preprocessing results are obtained, retaining the information that is truly useful for decision-making, so as to avoid the LLM context being overwhelmed by redundant geometric details.

[0032] The 3D BREP model defines 3D entities using boundary representation, that is, expressing shapes through the topological structure of faces, edges, and vertices, as well as the geometric definitions of curves, surfaces, and points. Therefore, the geometric primitives mentioned in this step refer to the basic geometric elements that constitute the 3D BREP model, mainly faces, edges, and vertices. Topological relationships define how these primitives are connected. In implementation, the system can parse the 3D BREP model and extract the geometric primitives and topological relationships. When the 3D BREP model contains multiple parts, the system assembles these parts into a scene diagram and generates an instance that can be used for projection for each data object, i.e., each part. This instance is a projectible representation of each data object in the scene diagram on the projection link. A part in an assembly, a single sheet metal part, or a single machined part can all be an instance. In addition, the system can also calculate semantic relationships between instances based on the business relationship registry, such as parallel relationships, perpendicular relationships, coplanar relationships, and coaxial relationships. The business relationship registry here can be a set of rules used to define semantic relationships between parts. For example, when the principal plane normal vectors or principal axis directions of two instances are perpendicular, a perpendicular relationship is recorded between the two instances; when two instances have coplanar planes, a coplanar relationship is recorded between the two instances. Through this geometric preprocessing operation, the raw geometric data can be transformed into lightweight structured information.

[0033] Step 102: Determine the scene business coordinate system, generate visibility facts using the six-view planner and the geometric preprocessing results, and input the visibility facts into the supplementary graph decision agent to obtain the supplementary graph decision result; wherein, the six-view planner evaluates the visibility of each instance in the six standard view directions under the scene business coordinate system; the visibility facts represent the coverage status of each target to be expressed under the standard view directions and the candidate view direction information of the uncovered features; the supplementary graph decision result represents the set of standard orthogonal views that need to be supplemented; In this embodiment, the system constructs two types of lightweight facts—visibility facts and expression gap facts—based on the geometric preprocessing results, which serve as structured inputs to the Agent, thereby transforming the semantic question of "whether to fill in / section the image" into structured facts that can be directly consumed by the Agent.

[0034] In 3D CAD assemblies, each part typically corresponds to a separate model file, and each part file usually has its own local coordinate system. The scene business coordinate system mentioned in this step is a unified spatial reference frame established for the entire assembly scene. Subsequent view planning and projection calculations are all performed within this framework. In some embodiments, determining the scene business coordinate system mentioned in this step may include: determining the local coordinate system corresponding to each instance using a file local coordinate agent based on a preset coordinate candidate strategy; and merging the various local coordinate systems using the scene business coordinate agent to generate a unified scene business coordinate system. That is, the system has a built-in strategy library with multiple pre-set coordinate candidate strategies. Each coordinate candidate strategy is used to calculate the local coordinate system for a certain part type. The system can use the file local coordinate agent, with "file-level" granularity, to calculate the local coordinate axes and origin for a single part's model file based on feature directions and the strategy library, forming the local coordinate system corresponding to each instance. Then, the scene business coordinate agent is used to aggregate the local coordinates of multiple files to output the scene business coordinate system. In implementation, the scenario business coordinate agent can determine the scenario business coordinate system through cluster analysis. For example, a reducer assembly contains 10 parts. The Z-axis of the local coordinate system of 8 parts is in the same direction, which is vertically upward. The Z-axis of the local coordinate system of the remaining 2 parts is horizontal. Then, the scenario business coordinate agent can establish the scenario business coordinate system by taking the vertical direction as the Z-axis of the scenario business coordinate system and the center of the bottom surface of the box as the origin through cluster analysis.

[0035] Furthermore, in some embodiments, determining the scene service coordinate system mentioned in this step may further include: when the local coordinate system corresponding to any instance cannot be determined by the preset coordinate candidate strategy, or when the fusion of each local coordinate system fails, the scene service coordinate system is determined using a look-intention agent; wherein, the look-intention agent calls the geometric kernel and the large language model to identify the look-intention of the scene service. That is, if a certain part type of the assembly does not match any of the coordinate candidate strategies in the strategy library, or the local coordinate systems of each part cannot be fused (e.g., the orientations of each part in the assembly are extremely dispersed, there is no main direction, or the number of parts is too small to form a statistical consensus), then the look-intention agent is switched to provide a fallback. The look-intention agent calls the automatic look-in recognition function of geometric kernels such as OpenCascade, generates several candidate looks based on the principal plane and inertial principal axis of the model bounding box, and inputs the geometric preprocessing results into the LLM, which selects a suitable look-in combination based on the engineering semantics. Finally, the automatic recognition results of the geometric kernel and the direction selection results of the LLM are weighted and fused to output the final look-intention of the scene service, thereby generating the scene service coordinate system. In this way, by utilizing the fallback mechanism of the gaze intent agent, we can ensure that a scene business coordinate system that conforms to engineering semantics can be generated under any assembly form, thereby ensuring that the mapping process is not interrupted.

[0036] The target to be represented in this step is a feature of each instance. Each instance of a part typically has multiple features, such as through holes, grooves, bends, chamfers, countersunk holes, etc. The six-view planner can be considered a pure algorithm node in the system. It evaluates the visibility of each instance in six standard view directions (front view, top view, left view, right view, bottom view, and rear view), thereby generating visibility facts F1. Visibility facts F1 indicate in which view directions each target to be represented has been fully represented and which candidate view directions are solvable.

[0037] In some embodiments, the visibility facts mentioned in this step may include existing view type fields (existing_view_types), candidate view type fields (candidate_view_types), existing sufficient target fields (existing_sufficient_targets), uncovered target fields (uncovered_targets), and candidate sufficient target fields (candidate_sufficient_targets_by_view) divided by view direction; wherein, the candidate view type fields include a sequence of candidate standard view directions according to a predetermined priority; accordingly, the visibility facts mentioned in this step... The generation of visibility facts using a six-view planner and the geometric preprocessing results may include: identifying the target to be represented from the geometric preprocessing results using the six-view planner; projecting each target to be represented according to six standard view directions, and determining whether the key dimension information of the target to be represented is visible and sufficiently represented in each standard view direction; recording the fully represented target field, the uncovered target field, and the candidate fully represented target field according to the determination results; selecting the view that covers the most uncovered projection instances using a greedy strategy and recording it in the existing view type field; and encapsulating the recorded information according to the data structure of the visibility facts to generate the visibility facts.

[0038] In other words, the six-view planner's work includes: feature recognition, which identifies the features to be represented from geometric primitives and topological relationships; visibility testing, which performs projection tests on each identified feature v according to six standard viewing directions to determine whether the key information of the feature, such as the outline of a hole, the bottom surface of a groove, and the bevel of a chamfer, can be seen from a certain direction. The test results are recorded in candidate_sufficient_targets_by_view[v] using Boolean values ​​or enumerations; and sufficiency of representation, which determines whether the core geometric information of feature v can be read unambiguously. If so, the feature is considered to be fully represented. For example, if the outline of a hole can be seen in a certain view but the depth cannot be seen, the hole is considered not fully represented, and the result is recorded in e. The `existing_sufficient_targets` and `uncovered_targets` are used for: Greedy visibility evaluation, which greedily selects the view that covers the most hidden instances as the planned standard view and records it in `existing_view_types`. Here, hidden instances are instances with invisible characteristics on a certain view. If, for an assembly, the top view covers 60% of the hidden instances, the front view covers the remaining 30%, and the right view covers the last 10%, then the six-view planner uses these three views as the planned standard views and records them in `existing_view_types`. Structured packaging encapsulates the above information into a structured dictionary based on the visibility facts data structure, which serves as input to the supplementary graph decision agent.

[0039] The supplementary map decision agent mentioned in this step takes the visibility fact F1 as input and outputs a set of standard orthogonal views that still need to be supplemented. In some embodiments, the input of the visibility fact into the supplementary map decision agent to obtain the supplementary map decision result may include: using the supplementary map decision agent to traverse each candidate view that does not belong to the currently planned standard view set in the candidate standard view sequence; if the coverage contribution record of the candidate view has an intersection with the set of uncovered targets, then add the candidate view to the new standard view set; and use the new standard view set as the set of standard orthogonal views that need to be supplemented; wherein, the currently planned standard view set comes from the existing view type field in the visibility fact, the coverage contribution record comes from the candidate fully expressible target field divided by view in the visibility fact, and the uncovered target set comes from the uncovered target field in the visibility fact. That is, the decision logic of the supplementary map decision agent includes: obtaining the planned standard view set according to existing_view_types. Based on `uncovered_targets`, obtain the set of uncovered targets. Based on candidate_view_types, obtain the candidate standard view sequence. Based on candidate_sufficient_targets_by_view, obtain candidate views. Coverage contribution record Coverage contribution records here It should at least include `gained_target_tokens` (which records which target tokens can be added to the view) and `gained_anchor_tokens` (which records which anchor tokens can be added to the view); iterate through... Each viewpoint in ,like And the coverage contribution record of this view direction. and If the intersection of the two sets is not empty, then the view direction will be added to the newly added standard view direction set. The selection order strictly follows Given a fixed position, no reordering, no greedy maximization, once... Stop immediately if the view is empty. This ensures that each supplemented view addresses at least one previously unaddressed target, avoiding the supplementation of invalid views and the duplication of existing views.

[0040] Step 103: Based on the supplementary map decision result, generate a two-dimensional engineering view, and based on the two-dimensional engineering view and the geometric preprocessing result, generate expression gap facts, then input the expression gap facts into the cross-sectional decision agent to obtain a cross-sectional plan, and generate a cross-sectional view based on the cross-sectional plan; wherein, the expression gap facts represent the missing dimension information of each target to be expressed that is not clearly expressed in the six standard viewing directions.

[0041] This step refers to: projecting each standard view according to the supplementary drawing decision results to generate a two-dimensional engineering view; then using the generated two-dimensional engineering view and geometric preprocessing results to generate the expression gap fact F2. The expression gap fact F2 marks the missing dimensions for each feature that is not clearly expressed under the standard six views, such as missing aperture contour, missing hole depth, missing groove bottom, and invisible inner / outer sides of bends. After that, the expression gap fact F2 is input into the sectioning decision agent, which outputs several sectioning plans, and then generates a section view according to the sectioning plan, thereby completing the industrial CAD drawing.

[0042] In some embodiments, the expression gap facts mentioned in this step may include: existing standard view field existing_standard_views, unresolved feature field unresolved_features, merge candidate field merge_candidates, and grouped candidate field grouped_candidates; accordingly, generating expression gap facts based on the two-dimensional engineering view and the geometric preprocessing result mentioned in this step may include: determining existing standard views based on the two-dimensional engineering view and recording them in the existing standard view field; identifying the target to be expressed that is still not clearly expressed under the existing standard view as unresolved features, identifying their missing dimension types and recording them in the unresolved feature field; merging and marking multiple unresolved features located on the same cutting plane, generating merge candidate groups and recording them in the merge candidate field; grouping each unresolved feature according to the cutting plane, and recording the grouping results in the grouped candidate field. In other words, the system determines existing standard views based on the two-dimensional engineering view. For example, if the two-dimensional engineering view includes a front view, a top view, and a right view, the system identifies these three views as existing standard views and records them in `existing_standard_views`. Then, for the target to be represented, the system determines whether it has been adequately represented in the existing standard views. If not, the system identifies it as an unresolved feature, identifies its missing dimension type (e.g., missing aperture contour, missing hole depth, missing groove bottom, invisible inner surface of bends), and records the identification results in `unresolved_features`. If multiple unresolved features are located on the same cutting plane, the system merges these unresolved features and records them in `merge_candidates` to instruct the sectioning decision agent to merge features on the same plane into a single sectioning plan, avoiding duplicate sectioning. Finally, using the cutting plane as an index, `grouped_candidates` is recorded, thus providing a global grouped view for sectioning decisions.

[0043] The profile decision agent mentioned in this step takes the expressed gap fact F2 as input and outputs several profile plans. These profile plans include profile location and expressed content. The profile location defines the position and orientation of the cutting plane in three-dimensional space, and the expressed content defines the unresolved features and their missing dimensions targeted by the profile plan. In some embodiments, the profile plan mentioned in this step corresponds to a grouping key; the grouping key is based on the strategy name and missing dimension type of the unresolved feature corresponding to the profile plan and the hash function value of the existing standard view. Accordingly, generating a profile view based on the profile plan in this step may include: obtaining the grouping keys corresponding to all profile plans; merging profile plans with the same grouping key; performing a sectioning operation based on the merged profile plan to generate a profile view; wherein multiple profile plans with the same grouping key correspond to the same profile view. That is, each profile plan corresponds to one grouping key. The group key This can be achieved by concatenating the strategy names of unresolved features, the current view set, and the set of dimensions representing gaps into a string according to normalization rules, and then generating the string using a hash function. The strategy name here can be a strategy label assigned during feature recognition; for example, when the identified feature is a through hole, the strategy label could be "business_hole_strategy," and when the identified feature is a groove, the strategy label could be "groove_strategy." When generating the sectional view, the same... Multiple sectional plans will be merged into a single sectional view downstream. This effectively reduces the number of sectional views and avoids redundancy by generating sectional views repeatedly for the same type of feature.

[0044] In some embodiments, the file local coordinate agent, scene business coordinate agent, view intention agent, supplementary image decision agent, and sectioning decision agent mentioned above all include a large language model node and a rule path. If any large language model node call fails or returns invalid output, the corresponding agent falls back to the rule path. That is, each of the five agents has two built-in decision paths: a large language model path and a rule path. During runtime, each agent prioritizes the large language model path. If this path call fails (e.g., timeout, network error, service unavailability) or returns invalid output (e.g., format error, incomplete content, logical contradiction), the agent automatically falls back to its own rule path without affecting the normal operation of other agents.

[0045] In some embodiments, before step 101, the method may further include: determining whether the master switch is set to a first state; if the master switch is set to the first state, shutting down all large language model nodes in the file local coordinate agent, the scene business coordinate agent, the look-intention agent, the supplementary image decision agent, and the cross-sectional decision agent. That is, a global degradation control mechanism is set in the system. Specifically, the master switch DO_USE_AGENT configuration item is set, which controls whether LLM capabilities are enabled. When the master switch is set to the first state, such as DO_USE_AGENT=0, all LLM nodes in the system are shut down, and the entire decision chain degenerates into a pure rule path. Thus, in scenarios where LLM call costs are sensitive, users can actively disable LLM by setting the master switch to the first state, allowing the system to run on a pure rule path, thereby effectively controlling costs.

[0046] In this embodiment, the 3D boundary representation model is preprocessed to extract geometric primitives and topological relationships, projectable instances, and semantic relationships between instances. A six-view planner is used to evaluate the visibility of each instance under six standard view directions, generating visibility facts. These visibility facts are input into the map completion decision agent to obtain a set of standard orthogonal views to be completed, and a 2D engineering view is generated accordingly. Then, based on the 2D engineering view and the geometric preprocessing results, gap expression facts are generated. These gap expression facts are input into the sectioning decision agent to obtain a sectioning plan, and a sectioning view is generated accordingly. Thus, by constructing two types of lightweight facts—visibility facts and gap expression facts—as structured inputs to the map completion decision agent and the sectioning decision agent, respectively, the final generated 2D engineering drawing can completely and non-redundantly express the part features while ensuring geometric accuracy.

[0047] To provide a more detailed explanation of the solution in this application, a specific embodiment is described below: This embodiment provides a view agent decision-making system, the overall architecture of which is as follows: Figure 2 As shown, it includes a geometry preprocessing layer 21, a fact layer 22, a six-agent decision layer 23, an engineering drawing generation layer 24, and a configuration and degradation switch 25. Among them: The geometry preprocessing layer 21 includes: a 3D model parsing module 211, used to parse the 3D BREP model and generate geometric primitives and topological relationships; a scene and instance construction module 212, used to assemble multiple parts into a scene graph and generate an instance that can be used for mapping for each data object; and a relationship management module 213, used to calculate the semantic relationships between instances based on the business relationship registry. This geometry preprocessing layer 21 outputs lightweight facts to the upstream fact layer 22, instead of raw geometry, retaining only the "feature-view-expression sufficiency" information that is truly necessary for decision-making, thus preventing the LLM context from being overwhelmed by redundant geometric details.

[0048] The role of Fact Layer 22 is to transform the semantic question of whether additional plotting / sectioning is needed into structured facts that can be directly consumed by the Agent, namely, visibility facts F1 and expression gap facts F2. Visibility facts F1 annotate each target to be expressed with information on which views have been fully expressed and which candidate views can resolve it. Its fields include existing_view_types, candidate_view_types, existing_sufficient_targets, uncovered_targets, and candidate_sufficient_targets_by_view (divided by view). Expression gap facts F2 annotate the missing dimensions for each feature that is not clearly expressed under the standard six views. Its fields include existing_standard_views, unresolved_features, merge_candidates, and grouped_candidates.

[0049] The six-agent decision layer 23 consists of six agents: The file-level local coordinate agent 231 calculates the local coordinate axes and origin at the "file level" granularity, based on the feature directions and the model file (each a single part of the policy library), obtaining the local coordinate system corresponding to each instance. Its output is used by the downstream scene business coordinate agent 232. The scene business coordinate agent 232 aggregates multi-file local coordinates and outputs scene business coordinates. In the event of a double miss (i.e., all candidate methods are unavailable), it switches to the view intent agent 233 as a fallback. The view intent agent 233 combines the OpenCascade automatic recognition results with LLM direction selection to determine the view intent of the scene business. The intelligent six-view planner 234, a pure algorithm node, evaluates the visibility of each instance in the six standard view directions, greedily selecting the view that covers the most hidden instances. The visibility fact F1 is generated by the intelligent six-view planner 234 based on the existing view evaluation. The supplementary map decision agent 235 takes the visibility fact F1 as input and outputs a set of standard orthogonal views that still need to be supplemented. The profiling decision agent236 takes the expressed gap fact F2 as input and outputs several profiling plans.

[0050] The engineering drawing generation layer 24 includes: a standard view projection module 241, which is used to project each standard view according to the supplementary drawing decision result to generate a two-dimensional engineering view; a sectional view generation module 242, which is used to determine the cutting position and expression content according to the sectional plan to generate a sectional view; and a dimension and annotation module 243, which is used to complete the annotation of dimensions, geometric tolerances, technical annotations, etc. on the two-dimensional view.

[0051] The configuration and degradation switches 25 include: a master switch 251, which disables all LLMs when set to 0, degrading the decision chain to a pure rule path; a rule priority switch 252, which prioritizes rule paths when set, with LLMs only supplementing; and a compact LLM switch 253, which narrows LLM inputs to pre-grouped and summarized data when set, restricting output to a structured schema. Additionally, in the six-agent decision layer 23, each agent transmits data between nodes via LangGraph StateGraph in the form of AgentState, i.e., data is transmitted between nodes as shared state objects through the state graph. Furthermore, the master switch 251 is active on all LLM nodes, while the rule priority switch 252 and the compact LLM switch 253 are specifically for profiling the decision agent 236; other LLM nodes also have their own two-level degradation path: "LLM → Rule". Each decision is traced at the fact layer 22, agent nodes, and configuration layer, allowing for node-by-node replay when problems occur.

[0052] The workflow of the system is as follows Figure 3 As shown, it includes: S301. Input the 3D model; S302. Perform geometric preprocessing on the 3D model using geometric preprocessing layer 21; S303. Using file local coordinates, Agent outputs the local coordinate system corresponding to each instance; S304. Aggregate local coordinates of multiple files using the scenario business coordinates Agent and output the scenario business coordinates; S305. Determine if a double miss occurs. If yes, execute S306; otherwise, execute S307. S306. Use the gaze intent agent to determine the gaze intent of the scene service; S307. Use the intelligent six-view planner 234 to perform view planning; S308, Generate visible facts F1; S309. Using the supplementary graph decision agent, output the set of standard orthogonal views that still need to be supplemented based on the visibility facts F1; S310. Perform projection using the standard view projection module 241 to generate a two-dimensional engineering view; S311. Combine the two-dimensional engineering view and geometric preprocessing results to generate the expression gap fact F2; S312. Utilize the profiling decision agent to output several profiling plans based on the expressed gap facts F2; S313. Generate a sectional view using the sectional view generation module 242; S314. Complete the annotation using the dimensioning and annotation module 243; S315, Output engineering drawing.

[0053] This embodiment has at least the following advantages: First, the fact-layer driven approach ensures traceable decisions. Specifically, this embodiment models the question of "whether additional drawings / section views are needed" as a structured query of visible facts / expression gap facts. Each decision can find the corresponding input at the fact layer and be fully replayed through LangGraph StateGraph. Compared to pure rule template solutions and CAD kernel encapsulation solutions, the engineering maintenance costs caused by the decision black box are thus reduced. Second, geometric correctness and semantic reasoning ability are each responsible for their respective strengths. Specifically, this embodiment divides the geometric computation layer (fact generation) and the semantic reasoning layer (LLM decision-making) into two layers. LLM no longer guesses geometry end-to-end, but only makes semantic-level selections based on confirmed facts. Compared with end-to-end LLM schemes and general LLM agent schemes, the risk of LLM having illusions in surface attribution, topological loop closure, and geometric measurement is limited to the "semantic layer". Third, multi-level degradation makes the engineering site controllable. Specifically, the solution in this embodiment is equipped with four levels of degradation for the LLM node: "LLM → Rule → Compact LLM → Master Switch". When the LLM call fails, times out, or the output cannot be parsed, it automatically falls back to the rule path. The rule path retains the same input facts and output schema as the LLM path. The result is controllable and has little difference from the LLM path. Compared with the LLM general intelligent agent solution and the LLM-based domain ReAc solution, the solution in this embodiment can still provide engineering-usable decisions when the production LLM is unavailable. Fourth, the supplementary view and the section view are linked into the same semantic decision chain. Specifically, in this embodiment, the expression gap fact F2 is generated only after the intelligent six-view planner has planned, so that the section view decision can grasp the result after the supplementary view. Compared with the existing solution that treats the supplementary view and the section view as independent modules, it can more effectively eliminate the problems caused by the failure to update the facts after the supplementary view or the section view not being complementary to the existing view. Fifth, it is observable and auditable. Specifically, all agent nodes record input facts, prompts, LLM outputs, tool calls, and termination decisions through the traceable decorator. After the trace is stored as a structured log, it can be used for problem replay, A / B testing, and customer acceptance.

[0054] Corresponding to the embodiments of the aforementioned methods, this application also provides embodiments of industrial CAD projection devices and terminals for their applications: like Figure 4 As shown, Figure 4 This is a block diagram of an industrial CAD projection device provided in an embodiment of this application. The device includes: The preprocessing module 41 is used to preprocess the input data of the three-dimensional boundary representation model to obtain geometric preprocessing results; wherein, the geometric preprocessing results include the geometric primitives and topological relationships of the three-dimensional boundary representation model, the instances that can be used for mapping corresponding to each data object, and the semantic relationships between instances; The supplementary graph decision module 42 is used to determine the scene business coordinate system, generate visibility facts using the six-view planner and the geometric preprocessing results, and input the visibility facts into the supplementary graph decision agent to obtain the supplementary graph decision result; wherein, the six-view planner evaluates the visibility of each instance in the six standard view directions under the scene business coordinate system; the visibility facts represent the coverage status of each target to be expressed under the standard view directions and the candidate view direction information of the uncovered features; the supplementary graph decision result represents the set of standard orthogonal views that need to be supplemented; The sectioning decision module 43 is used to generate a two-dimensional engineering view based on the supplementary map decision result, and generate expression gap facts based on the two-dimensional engineering view and the geometric preprocessing result. The expression gap facts are then input into the sectioning decision agent to obtain a sectioning plan, and a section view is generated based on the sectioning plan. The expression gap facts represent the missing dimension information of each target to be expressed that is not clearly expressed in the six standard viewing directions.

[0055] The specific implementation process of the functions and roles of each module in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0056] This application also provides an electronic device, please refer to [link to application]. Figure 5 , Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of this application. The electronic device may include a processor 510, a communication interface 520, a memory 530, and at least one communication bus 540. The communication bus 540 is used to enable direct communication between these components. In this embodiment, the communication interface 520 of the electronic device is used for signaling or data communication with other node devices. The processor 510 may be an integrated circuit chip with signal processing capabilities.

[0057] The processor 510 described above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor, or the processor 510 can be any conventional processor.

[0058] The memory 530 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc. The memory 530 stores computer-readable instructions. When these computer-readable instructions are executed by the processor 510, the electronic device can perform the aforementioned operations. Figure 1 The various steps involved in the method implementation examples.

[0059] Alternatively, the electronic device may also include a storage controller and an input / output unit.

[0060] The memory 530, storage controller, processor 510, peripheral interface, and input / output unit are electrically connected directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses 540. The processor 510 is used to execute executable modules stored in the memory 530, such as software function modules or computer programs included in electronic devices.

[0061] The input / output unit is used to provide users with the ability to create tasks and to set optional start periods or preset execution times for those tasks, thereby enabling user-server interaction. The input / output unit may be, but is not limited to, a mouse and keyboard.

[0062] Understandable. Figure 5 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 5 The more or fewer components shown, or having the same Figure 5 The different configurations shown. Figure 5 The components shown can be implemented using hardware, software, or a combination thereof.

[0063] This application also provides a storage medium storing instructions. When the instructions are run on a computer, the computer program is executed by a processor to implement the method described in the method embodiment. To avoid repetition, the method will not be described again here.

[0064] This application also provides a computer program product that, when run on a computer, causes the computer to perform the method described in the method embodiment.

[0065] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0066] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0067] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application. 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.

[0068] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0069] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0070] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. An industrial CAD drawing method, characterized in that, include: The input data of the three-dimensional boundary representation model is preprocessed to obtain geometric preprocessing results; wherein, the geometric preprocessing results include the geometric primitives and topological relationships of the three-dimensional boundary representation model, the instances of each data object that can be used for mapping, and the semantic relationships between instances; A scene business coordinate system is determined. Visibility facts are generated using a six-view planner and the geometric preprocessing results. These visibility facts are then input into a supplementary graph decision agent to obtain supplementary graph decision results. The six-view planner evaluates the visibility of each instance in the six standard view directions of the scene business coordinate system. The visibility facts represent the coverage status of each target to be represented and candidate view direction information of uncovered features in the standard view directions. The supplementary graph decision results represent the set of standard orthogonal views that need to be supplemented. Based on the supplementary map decision result, a two-dimensional engineering view is generated, and based on the two-dimensional engineering view and the geometric preprocessing result, expression gap facts are generated. The expression gap facts are then input into the cross-sectional decision agent to obtain a cross-sectional plan, and a cross-sectional view is generated based on the cross-sectional plan. The expression gap facts represent the missing dimensional information of each target to be expressed that is not clearly expressed in the six standard viewing directions.

2. The method according to claim 1, characterized in that, The determination of the scene business coordinate system includes: Based on the preset coordinate candidate strategy, the local coordinate system corresponding to each instance is determined by the file local coordinate agent; By using a scenario-based business coordinate intelligent agent, various local coordinate systems are merged to generate a unified scenario-based business coordinate system.

3. The method according to claim 2, characterized in that, The determination of the scene business coordinate system also includes: If the local coordinate system corresponding to any instance cannot be determined by the preset coordinate candidate strategy, or if the fusion of each local coordinate system fails, the scene business coordinate system is determined by the look-in intent agent. The gaze intent agent invokes the geometric kernel and large language model to identify the gaze intent of the scene service.

4. The method according to claim 1, characterized in that, in, The visibility facts include: existing view type fields, candidate view type fields, fully expressed target fields, uncovered target fields, and candidate fully expressed target fields divided by view direction; the candidate view type fields include a sequence of candidate standard view directions according to a predetermined priority; The generation of visibility facts using the six-view planner and the geometric preprocessing results includes: The target to be represented is identified from the geometric preprocessing results using a six-view planner; For each target to be expressed, projection is performed according to six standard viewing directions to determine whether the key dimension information of the target to be expressed is visible and sufficiently expressed in each standard viewing direction. Record the fully expressed target field, the uncovered target field, and the candidate fully expressed target field based on the judgment result; A greedy strategy is used to select the view that covers the most uncovered projection instances and record it in the existing view type field; The recorded information is encapsulated according to the data structure of the visibility fact to generate the visibility fact.

5. The method according to claim 4, characterized in that, The step of inputting the visibility facts into the graph completion decision agent to obtain the graph completion decision result includes: Using a supplementary graph decision agent, each candidate view is traversed sequentially according to the predetermined priority in the candidate standard view sequence; If the candidate view that is currently being traversed does not belong to the currently planned standard view set, and the intersection of the coverage contribution record of the candidate view and the uncovered target set is not empty, then the candidate view is added to the new standard view set, and the targets contained in the intersection are removed from the uncovered target set. If the set of uncovered targets is empty, stop traversing and take the currently added standard view set as the standard orthogonal view set to be supplemented; Wherein, the currently planned standard view set comes from the existing view type field in the visibility facts, the coverage contribution record comes from the candidate fully expressible target field divided by view in the visibility facts, and the uncovered target set comes from the uncovered target field in the visibility facts.

6. The method according to claim 5, characterized in that, in, The expressed gap facts include: existing standard view fields, unresolved feature fields, merged candidate fields, and grouped candidate fields; The step of generating the expression of the gap fact based on the two-dimensional engineering view and the geometric preprocessing result includes: The existing standard view is determined based on the two-dimensional engineering view and recorded in the existing standard view field; The target to be expressed that is not clearly expressed under the existing standard view is identified as an unresolved feature, its missing dimension type is identified and recorded in the unresolved feature field; Multiple unresolved features located on the same cutting plane are merged and marked to generate merge candidate groups, which are then recorded in the merge candidate field. The unresolved features are grouped according to the cutting plane, and the grouping results are recorded in the grouping candidate field.

7. The method according to claim 6, characterized in that, in, The profile plan corresponds to a grouping key; the grouping key is based on the strategy name and missing dimension type of the unresolved feature corresponding to the profile plan and the hash function value of the existing standard view; The step of generating the sectional view according to the sectional plan includes: Retrieve the grouping keys corresponding to all section plans; Merge the section plans with the same grouping key; A sectioning operation is performed based on the merged sectioning plan to generate a section view; wherein multiple sectioning plans with the same grouping key correspond to the same section view.

8. An industrial CAD projection device, characterized in that, include: The preprocessing module is used to preprocess the input data of the 3D boundary representation model to obtain geometric preprocessing results; wherein, the geometric preprocessing results include the geometric primitives and topological relationships of the 3D boundary representation model, the instances of each data object that can be used for mapping, and the semantic relationships between instances; The map completion decision module is used to determine the scene business coordinate system, generate visibility facts using a six-view planner and the geometric preprocessing results, and input the visibility facts into the map completion decision agent to obtain the map completion decision result. The six-view planner evaluates the visibility of each instance in the six standard view directions under the scene business coordinate system. The visibility facts represent the coverage status of each target to be represented under the standard view directions and candidate view direction information of uncovered features. The map completion decision result represents the set of standard orthogonal views that need to be supplemented. The sectioning decision module is used to generate a two-dimensional engineering view based on the supplementary map decision result, and generate expression gap facts based on the two-dimensional engineering view and the geometric preprocessing result. The expression gap facts are then input into the sectioning decision agent to obtain a sectioning plan, and a section view is generated based on the sectioning plan. The expression gap facts represent the missing dimension information of each target to be expressed that is not clearly expressed in the six standard viewing directions.

9. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method as described in any one of claims 1 to 7.