Semantized analysis method and system for version difference of three-dimensional assembly

Abstract

The invention discloses a semantic analysis method and system for version differences of three-dimensional assemblies, and relates to the technical field of computer aided design. The method comprises the following steps: acquiring two version data of a three-dimensional assembly design file; carrying out analysis and lightweight sequence processing on the three-dimensional scene graph to generate a structured text only containing a hierarchical relationship and a component identifier; inputting the structured texts of the two versions into a pre-trained large language model for difference reasoning; and obtaining a natural language change abstract output by the model, and displaying the natural language change abstract in the version comparison view. According to the method, the problem that a traditional geometric comparison tool cannot explain structural and semantic changes is solved, a complex BOM structure is converted into sequence data which can be understood by AI, automatic recognition and summarization of advanced change behaviors such as component movement and hierarchical recombination are achieved, and the efficiency of engineering design review is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer-aided design (CAD), product data management (PDM) and artificial intelligence application, and particularly relates to a three-dimensional assembly version difference semantic analysis method and system based on a large language model. BACKGROUND

[0002] In the product development process of complex manufacturing industries such as automobiles, aerospace and consumer electronics, the design of a three-dimensional assembly model often needs to go through dozens or even hundreds of version iterations. Designers, engineering managers and manufacturing engineers need to frequently compare two versions (such as V1.5 and V1.6) to confirm whether the design changes meet expectations or to assess the impact of changes on subsequent processes.

[0003] The existing mainstream CAD software (such as SolidWorks, Catia, NX) usually provides a version comparison function based on "geometric Boolean operation". Its working principle is to spatially superimpose the three-dimensional entities of the two models, calculate the volume difference set, and display the differences through color coding (such as red for removed parts and green for added parts).

[0004] However, this traditional comparison method based on geometric pixels has significant technical limitations:

[0005] First, semantic loss. When a designer moves a sub-assembly (such as "battery module") from the "chassis system" node to the "trunk" node, although its physical position may only be translated by a few millimeters, its BOM (bill of materials) structure has changed fundamentally. The traditional tool will only show a large piece of red and green geometric differences, and the user cannot intuitively understand that this is only a "logical level movement" operation.

[0006] Second, high cognitive load. For a large assembly containing thousands of parts, the geometric difference points may reach hundreds. Engineering management personnel have difficulty extracting key "structural change" information (such as: whether the supplier part is replaced? Whether the sub-assembly is disassembled?) from the massive geometric fragments, resulting in extremely low review efficiency.

[0007] Therefore, the existing technology urgently needs a technical solution that can go beyond simple geometric comparison and automatically analyze and summarize version differences from the assembly logic and semantic level. SUMMARY

[0008] The present application aims to solve the above technical problems, and uses the powerful semantic reasoning and text generation capabilities of the large language model (LLM) to propose an analysis method that can automatically analyze the structural changes of the assembly and generate a natural language summary.

[0009] To achieve the above object, the present application provides the following technical scheme:

[0010] A semantic analysis method for version difference of a three-dimensional assembly body, comprising the following steps:

[0011] Obtaining first version data and second version data of a three-dimensional assembly body design file;

[0012] Respectively analyzing the first version data and the second version data, extracting respective corresponding three-dimensional scene graph structures, and performing lightweight sequence processing on the three-dimensional scene graph structures to generate first structured text and second structured text containing node level relationship and component identification information;

[0013] Constructing a prompt word vector containing difference analysis instructions, inputting the first structured text, the second structured text and the prompt word vector into a pre-trained large language model for semantic reasoning;

[0014] Receiving a difference analysis result output by the large language model, the difference analysis result being a natural language summary describing structural changes of the three-dimensional assembly body;

[0015] Displaying a version comparison view in a graphical user interface and rendering the natural language summary to an associated area of the version comparison view.

[0016] Further, the lightweight sequence processing on the three-dimensional scene graph structure specifically comprises:

[0017] Traversing all nodes of the three-dimensional scene graph, and filtering out auxiliary nodes that do not affect assembly logic;

[0018] Extracting unique identifiers, parent node indexes and component names of the retained nodes;

[0019] Converting the extracted information into a nested tree data structure in JSON format or XML format as the structured text.

[0020] Further, the structural change types covered by the difference analysis result at least include:

[0021] Creation and dissolution of sub-assemblies;

[0022] Hierarchical movement of components between different parent nodes;

[0023] Replacement, addition and deletion of components.

[0024] Further, the construction of the prompt word vector containing difference analysis instructions specifically comprises:

[0025] The role constraint parameter is defined in the prompt word template, indicating the model to ignore pure spatial coordinate transformation data and focus on changes in node topological relations;

[0026] The output format constraint parameter is configured to require the model to output the natural language summary in the form of a bulleted list.

[0027] Further, the version comparison view displayed in the graphical user interface specifically includes:

[0028] The three-dimensional models of the first version data and the second version data are rendered side by side in the screen area;

[0029] In response to a user clicking a specific change description item in the natural language summary, the component corresponding to the description item is highlighted in the three-dimensional model.

[0030] Further, the method further includes:

[0031] The node edit distance between the first structured text and the second structured text is calculated;

[0032] If the node edit distance is less than a preset threshold, a local rule engine is directly called to generate a difference report;

[0033] If the node edit distance is greater than or equal to the preset threshold, the step of inputting the pre-trained large language model for semantic reasoning is performed.

[0034] In addition, the present application also provides a semantic analysis system for version difference of a three-dimensional assembly, comprising:

[0035] A data acquisition module configured to acquire first version data and second version data of a three-dimensional assembly design file;

[0036] A structure extraction module configured to extract respective corresponding three-dimensional scene graph structures and perform lightweight sequence processing to generate first structured text and second structured text;

[0037] A semantic reasoning module configured to input the structured text into a large language model to obtain a natural language summary describing structural changes;

[0038] An interactive display module configured to display the natural language summary in a graphical user interface and display a three-dimensional model in linkage in response to user operation.

[0039] The present application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0040] The present application also provides a computer-readable storage medium having a computer program stored thereon.

[0041] The beneficial effects of this invention are as follows: by "reducing the dimensionality" of three-dimensional data into structured text that AI can understand, a "semantic-level" interpretation of design changes is achieved. Users no longer need to painstakingly search for differences in 3D space; they can quickly grasp the core content of version iterations simply by reading a change report generated by AI and written in a manner similar to that of a human engineer, greatly improving the efficiency and accuracy of collaborative design. Attached Figure Description

[0042] Figure 1 This is a flowchart of the method provided in Embodiment 1 of the present invention;

[0043] Figure 2 This is a schematic diagram of scene graph serialization and data structure conversion in Embodiment 1 of the present invention;

[0044] Figure 3 This is a schematic diagram of the semantic reasoning principle based on a large language model in Embodiment 1 of the present invention;

[0045] Figure 4 This is a system architecture block diagram provided in Embodiment 3 of the present invention;

[0046] Figure 5 This is a schematic diagram of the version comparison interaction interface provided in Embodiment 2 of the present invention. Detailed Implementation

[0047] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0048] Example 1

[0049] like Figure 1 As shown, this embodiment provides a semantic analysis method for version differences in three-dimensional assemblies. The data processing principle and reasoning logic in this embodiment are illustrated in Figures 2 and 3, respectively. The specific execution process is as follows:

[0050] Step S1: Data Analysis

[0051] The system first loads two CAD files to be compared (e.g., Assembly_V1.step and Assembly_V2.step). It then parses the files using a geometry kernel (such as Parasolid or ACIS) or a Web3D engine (such as Three.js) to construct an in-memory scene graph. The scene graph is a directed acyclic graph (DAG), where the root node represents the entire product and the leaf nodes represent specific parts.

[0052] Step S2: Structure Serialization (Feature Dimensionality Reduction)

[0053] This is the core step of the invention. Since commercial large language models (such as Qwen3 or DeepSeek) have limited input windows and cannot directly process GB-level three-dimensional geometric data, "dimensionality reduction" is necessary.

[0054] This embodiment designs a recursive traversal algorithm that, starting from the root node, converts the complex scene graph into a lightweight JSON object. For example... Figure 2 The diagram shown illustrates the scene graph serialization and data structure conversion in an embodiment of the present invention. The original 3D scene graph contains complex geometric information. After lightweight processing, it is converted into a nested JSON structure as shown on the right, retaining only key topological information such as id, name, and children.

[0055] The specific data structure design is as follows: { "version":"V1.0", "assembly_tree":[ { "uuid":"7f8a9c...", "name":"Chassis_System", "type":"Assembly", "children":["uuid_01","uuid_02"] / / Only records the child node ID index }, { "uuid":"uuid_01", "name":"Front_Axle", "type":"Part", "parent":"7f8a9c..." } / / ...Other nodes omitted ] }

[0056] The algorithm automatically filters out invisible auxiliary geometry (such as reference planes and construction lines) and specific 3D mesh data, thereby compressing the file size from hundreds of megabytes to a few kilobytes of text data.

[0057] Step S3: Prompt Engineering and AI Inference

[0058] The system constructs a prompt containing specific instructions, along with JSON data from both V1 and V2 versions, and sends it to the large language model. For example... Figure 3The diagram shown illustrates the semantic reasoning principle based on a large language model in this embodiment of the invention. The system combines the structured texts of V1 and V2 with a preset Prompt template (containing role definitions, tasks, and constraints) into a complete input vector, which is then input into the LLM. After reasoning, the LLM outputs a structured natural language summary.

[0059] The prompt template is as follows:

[0060] "You are a senior CAD data management expert. Please compare the following two JSON data (representing the assembly BOM structure)."

[0061] Task: Analyze what structural changes occurred from V1 to V2.

[0062] Constraints: 1. Ignore minor changes in coordinate values; 2. Focus on component movement (change of parent node), addition, and deletion; 3. Please output a concise Chinese list summary.

[0063] Input data: [JSON_V1], [JSON_V2]

[0064] Step S4: Result Generation

[0065] After inference, the model returns the following natural language summary:

[0066] 1. Restructuring: The component “Battery_Pack” has been moved from the “Root” node to the “Rear_Subframe” sub-assembly.

[0067] 2. New component: Four "M6 Screws" have been added under "Front_Bumper".

[0068] 3. Component removal: The old version of "Radar_Bracket_V1" (radar bracket) has been removed.

[0069] Example 2

[0070] like Figure 5 As shown, this embodiment focuses on describing the interactive display method of the analysis results.

[0071] Traditional text reports are static, while this invention achieves dynamic interaction between "text and model".

[0072] After the AI ​​generates the above summary, the system will bind the keywords in the summary (such as "Battery_Pack") to the UUID in the scene graph.

[0073] In the version comparison view, the right side of the interface displays a list of AI-generated summaries. When the user clicks on the first item in the list, "Structural Reorganization...":

[0074] 1. The system searches for the corresponding model node in the 3D scene based on the UUID.

[0075] 2. Calculate the center point and radius of the world coordinate system bounding box of this node.

[0076] 3. Control the virtual camera to perform smooth interpolation animation (Tweening), fly to the target location, and zoom the viewpoint to the optimal viewing distance (usually 1.5 times the radius).

[0077] 4. Apply a highlighting material (such as a self-illuminating outline) to the component, and set other unrelated components to semi-transparent (Ghost Mode) so that users can clearly see the specific location where the change occurred.

[0078] Example 3

[0079] like Figure 4 As shown, the present invention also provides a system architecture, including:

[0080] Structure extraction module: responsible for reading CAD source files, performing denoising and serialization operations, and converting binary models into text streams.

[0081] Semantic reasoning module: Deployed with a large language model interface (API), responsible for receiving text streams and performing difference comparison reasoning.

[0082] Interactive display module: Built on WebGL technology, it is responsible for rendering dual-screen comparison views and responding to user click events to realize viewpoint jump from text to three-dimensional space.

[0083] In addition, this system includes a rule-based pre-check module. Before invoking expensive AI inference, it calculates the levenshtein distance between two JSON texts. If the distance is 0, it directly displays "no changes"; if the distance is extremely small, it generates a report using only simple regular expression matching, thus saving AI computing power costs.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A semantic analysis method for version differences in three-dimensional assemblies, characterized in that, Includes the following steps: Obtain the first and second version data of the 3D assembly design file; The first version data and the second version data are parsed respectively, and their corresponding three-dimensional scene graph structures are extracted. The three-dimensional scene graph structures are then subjected to lightweight sequence processing to generate a first structured text and a second structured text containing node hierarchy relationships and component identification information. Construct prompt word vectors containing differential analysis instructions, and input the first structured text, the second structured text, and the prompt word vectors into a pre-trained large language model for semantic reasoning; Receive the difference analysis results output by the large language model, wherein the difference analysis results are natural language summaries describing the structural changes of the three-dimensional assembly; Display the version comparison view in the graphical user interface and render the natural language summary to the associated area of ​​the version comparison view.

2. The method according to claim 1, characterized in that, The lightweight sequence processing of the three-dimensional scene graph structure specifically includes: Traverse all nodes of the 3D scene graph and filter out auxiliary nodes that do not affect the assembly logic; Extract the unique identifier, parent node index, and component name of the retained node; The extracted information is converted into a nested tree data structure in JSON or XML format, which serves as the structured text.

3. The method according to claim 1, characterized in that, The structural change types covered by the difference analysis results include at least the following: Creation and dismantling of subassemblies; Components can move hierarchically between different parent nodes; Replacement, addition, and deletion of components.

4. The method according to claim 1, characterized in that, The construction of the prompt word vector containing the differential analysis instruction specifically includes: Define role constraint parameters in the prompt word template to instruct the model to ignore simple spatial coordinate transformation data and focus on changes in node topology; Configure output format constraints to require the model to output the natural language summary in the form of a bulleted list.

5. The method according to claim 1, characterized in that, The graphical user interface displays a version comparison view, specifically including: Render the 3D models of the first and second version data side-by-side in the screen area; In response to a user clicking on a specific change description entry in the natural language summary, the component corresponding to that description entry is highlighted in the 3D model.

6. The method according to claim 1, characterized in that, The method further includes: Calculate the node edit distance between the first structured text and the second structured text; If the node edit distance is less than a preset threshold, the local rule engine is directly invoked to generate a difference report; If the node edit distance is greater than or equal to a preset threshold, then the step of semantic reasoning using the pre-trained large language model is executed.

7. A semantic analysis system for version differences of three-dimensional assemblies, characterized in that, include: The data acquisition module is configured to acquire the first and second version data of the 3D assembly design file. The structure extraction module is configured to extract the corresponding 3D scene graph structure and perform lightweight sequence processing to generate the first structured text and the second structured text. A semantic reasoning module is configured to input the structured text into a large language model to obtain a natural language summary describing the structural changes. An interactive display module is configured to display the natural language summary in a graphical user interface and to display a 3D model in response to user actions.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images