CN-121982261-A - Industrial three-dimensional model self-adaptive rendering method and system based on semantic driving

CN121982261ACN 121982261 ACN121982261 ACN 121982261ACN-121982261-A

Abstract

The invention provides an industrial three-dimensional model self-adaptive rendering method and system based on semantic driving, and belongs to the technical field of computer graphic processing and industrial digitization. The method comprises the steps of constructing a heterogeneous model data structure containing discrete grid data, parameterized feature data, feature semantic data and a mapping relation between features and discrete grid local areas in an offline stage, dynamically determining a rendering priority according to engineering semantic weights of the features, screen space projection amounts and viewing distances, determining the features meeting conditions as features to be reconstructed in a rendering process, sampling parameter areas and generating corresponding vertex data through parallel calculation based on the parameterized feature data of the features to be reconstructed, shielding, removing or replacing the corresponding local areas in an original discrete grid according to the mapping relation, and fusing and rendering the generated vertex data and the residual discrete grid data. The method realizes the on-demand refined reconstruction of key characteristics of the industrial model.

Inventors

LIU QINGBIAO
GONG XIAOFENG

Assignees

重庆诺源工业软件科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (6)

1. An industrial three-dimensional model self-adaptive rendering method based on semantic driving is characterized by comprising the following steps: the heterogeneous lightweight model data is obtained and comprises discrete grid data, parameterized feature data for representing the geometric shape of an industrial model, feature semantic data and a mapping relation between the parameterized feature data and grid areas in the discrete grid data; dynamically calculating the rendering priority of the feature according to the engineering semantic weight corresponding to the parameterized feature data and the view parameter of the feature corresponding to the parameterized feature data in the current view state; determining the characteristics of the rendering priority meeting the preset conditions as characteristics to be reconstructed; Based on the parameterized feature data of the features to be reconstructed, sampling a parameter domain, generating corresponding vertex data through parallel calculation, shielding, eliminating or replacing a discrete grid local area according to the mapping relation, carrying out fusion rendering on the generated vertex data and the rest discrete grid data, and outputting the vertex data and the rest discrete grid data to display equipment; The parameterized feature data comprise geometric parameters and space positioning information for describing target features, are obtained by analyzing boundary representation data, and further comprise local accurate geometric expression data corresponding to the target features; the engineering semantic weights are preset based on the functional importance of the industrial features in assembly, manufacture or measurement, and different types of industrial features have different semantic weights, wherein the semantic weights of the matching features and the positioning features are higher than those of the nonfunctional decorative features; The view parameters comprise projection areas of features in a screen space and distances between the features and observation points, and the rendering priority is comprehensively determined by the projection areas, the engineering semantic weights and the distances; The calculation model of the rendering priority P is as follows: Wherein S is the projection area of the feature in the screen space, For engineering semantic weights of features, sem is an abbreviation of english word semantic of semantics, D is a distance between a viewpoint and a feature, f () is a calculation function, f () includes a calculation function based on weighted summation and a calculation function based on product; when f () employs a calculation function based on weighted summation, the calculation model of the rendering priority P is: Wherein S max is the maximum projected area of all features in the current view, D max is the preset maximum effective viewing distance, 、、 Is a weight coefficient, satisfies ; When f () employs a product-based computation function, the computation model of the rendering priority P is: Wherein S ref is the reference projection area, D ref is the reference viewing distance, and alpha, beta and gamma are the projection area, semantic weight and index parameters of the viewing distance respectively, and are used for adjusting the nonlinear influence degree of each factor on the priority.
2. The semantic-driven industrial three-dimensional model adaptive rendering method of claim 1, wherein the parallel computing employs at least one of a central processing unit and a graphics processing unit, and generating vertex data by WebGPU when the parallel computing employs WebGPU specifically comprises: Distributing thread groups matched with a preset sampling density in WebGPU computing shaders; each thread samples the parameter domain of the parameterized feature data in parallel, and calculates the three-dimensional coordinates of the sampling points based on the corresponding geometric analysis equation; and directly writing the calculated three-dimensional coordinates into a vertex buffer area in WebGPU video memory to generate vertex data.
3. The semantic-driven industrial three-dimensional model adaptive rendering method according to claim 2, wherein the parameterized feature data comprises cylindrical hole features, and the geometric analysis equation corresponding to the cylindrical hole features performs polar coordinate sampling reconstruction based on axis starting point coordinates, axis vectors, radius and height parameters.
4. The semantic-driven industrial three-dimensional model adaptive rendering method according to claim 1, wherein the method is performed in a browser environment, and the parallel computation is implemented with WebGPU computation shaders and/or vertex shaders.
5. The adaptive rendering method of the industrial three-dimensional model based on semantic driving according to claim 1, wherein the rendering priority has a plurality of levels, the method further comprises dynamically adjusting parameter sampling frequency when reconstruction is performed in parallel calculation according to the level of the rendering priority to generate vertex data with different fine degrees, shielding, eliminating or replacing a discrete grid local area according to the mapping relation, fusing the generated vertex data with the rest discrete grid data to render the vertex data comprises positioning a discrete grid area corresponding to a feature to be reconstructed according to the mapping relation, and rendering the discrete grid area, eliminating triangle or replacing the vertex data instead of the generated vertex data.
6. An industrial three-dimensional model adaptive rendering system based on semantic driving, comprising: the data processing module is used for acquiring and analyzing heterogeneous lightweight model data, and comprises discrete grid data, parameterized characteristic data and a mapping relation between the discrete grid data and the parameterized characteristic data; the scheduling engine module is used for calculating rendering priority according to engineering semantic weights and view parameters of the features and determining the features to be reconstructed based on the rendering priority; the parallel computing module is used for sampling the parameter domain based on the parameterized characteristic data and generating vertex data; the rendering fusion module is used for shielding, eliminating or replacing the local area of the discrete grid according to the mapping relation, and carrying out fusion rendering on the generated vertex data and the residual discrete grid data; The parameterized feature data comprise geometric parameters and space positioning information for describing target features, are obtained by analyzing boundary representation data, and further comprise local accurate geometric expression data corresponding to the target features; the engineering semantic weights are preset based on the functional importance of the industrial features in assembly, manufacture or measurement, and different types of industrial features have different semantic weights, wherein the semantic weights of the matching features and the positioning features are higher than those of the nonfunctional decorative features; The view parameters comprise projection areas of features in a screen space and distances between the features and observation points, and the rendering priority is comprehensively determined by the projection areas, the engineering semantic weights and the distances; The calculation model of the rendering priority P is as follows: Wherein S is the projection area of the feature in the screen space, For engineering semantic weights of features, sem is an abbreviation of english word semantic of semantics, D is a distance between a viewpoint and a feature, f () is a calculation function, f () includes a calculation function based on weighted summation and a calculation function based on product; when f () employs a calculation function based on weighted summation, the calculation model of the rendering priority P is: Wherein S max is the maximum projected area of all features in the current view, D max is the preset maximum effective viewing distance, 、、 Is a weight coefficient, satisfies ; When f () employs a product-based computation function, the computation model of the rendering priority P is: Wherein S ref is the reference projection area, D ref is the reference viewing distance, and alpha, beta and gamma are the projection area, semantic weight and index parameters of the viewing distance respectively, and are used for adjusting the nonlinear influence degree of each factor on the priority.

Description

Industrial three-dimensional model self-adaptive rendering method and system based on semantic driving Technical Field The invention relates to the technical fields of computer graphics and industrial digitization, in particular to an industrial three-dimensional model self-adaptive rendering method and system based on semantic driving. Background With the deep advancement of industrial Internet and digital transformation, a three-dimensional Computer Aided Design (CAD) model of complex industrial products (such as engineering machinery, aerospace equipment, whole automobile and the like) is efficiently displayed and interacted at a Web end, and the three-dimensional Computer Aided Design (CAD) model has become a core requirement of application scenes such as remote operation and maintenance, digital marketing, collaborative design and the like. However, industrial three-dimensional models typically contain tens of thousands or even millions of parts, are data-intensive, complex in structure, and contain large amounts of critical engineering semantic information, which presents a serious challenge for real-time rendering. To address the above challenges, the prior art explores mainly from two levels of model weight reduction and rendering optimization. In one aspect, multi-level of detail (LOD) based model simplification techniques are widely employed. For example, chinese patent application CN110362927a proposes a BIM model lightweight processing method, by separating attribute from grid data, and generating multiple grids (coarse model, middle model, fine model) with different precision for a component, loading a model with corresponding precision according to viewpoint distance. PCT application publication No. WO2023124842A1 further introduces the concept of Screen Space Error (SSE) to determine when to switch finer model levels by computing the projection of geometric errors on the screen, and defines a "delta" and "alternate" node switching pattern to achieve smoother LOD transitions. These approaches alleviate rendering pressure to some extent, but still have the essence of scheduling and switching pre-generated discrete grid data. When a user needs to locally observe key engineering features (such as a positioning hole, a matching surface and a sealing groove) in an amplified manner, the features still generate polygonal distortion due to grid simplification, and the requirements of accurate measurement and assembly analysis in industrial scenes cannot be met. On the other hand, in order to improve the data loading and parsing efficiency, the industry is dedicated to optimize the organization format and transmission protocol of three-dimensional data. For example, chinese patent application publication No. CN112270756a proposes the idea of component templating and instantiation, reducing data redundancy by multiplexing components of the same geometry. The chinese patent application with publication number CN116977523a describes a complete set of Web-side rendering processes, including data parsing, preprocessing, mesh conversion, progressive rendering, and visual field rejection, and discusses the possibility of using distributed rendering to enhance performance. However, these methods still focus on the compression, organization and transmission of the mesh data, and the geometric accuracy of the model after the restoration at the browser end is completely dependent on the quality of the mesh that is pre-generated and stored. Parameterized features and boundary representation (Brep) information in the original computer-aided design (CAD) model are lost, resulting in the Web-side model falling into a "dumb pattern" that cannot support feature-based intelligent interactions and downstream process applications. In addition, in the prior art, a scheme of presetting loading priority according to an application scene or user operation is also available. For example, chinese patent application CN110069733a sets static loading priorities for different components according to the whole-machine-level or component-level application mode. The scheme has thicker decision granularity, can not respond to the dynamically changed observation intention of a user, has no relation with the importance of industrial characteristics, and is difficult to ensure the display precision of a key process surface in a complex scene. In summary, in the prior art, when dealing with rendering of a very large scale industrial 3D model, there are three interrelated drawbacks: The inherent contradiction of accuracy and performance is that whether distance-based LOD switching or SSE-based HLOD refinement, its implementation at the fine level relies on pre-stored high-accuracy grid data. To ensure fluency, the overall data volume has to be cut down drastically, resulting in geometric distortion of the critical engineering features at the time of scaling. The engineering semantics are thoroughly lost, the existing lightweight process gen