A three-dimensional visual editing management method and system based on digital twins

Abstract

The present invention discloses a three-dimensional visual editing and management method and system based on digital twins. The method first collects enterprise production parameters and performs structured processing. After constructing the initial three-dimensional model, the method uses a generative adversarial network to optimize the model details through adversarial training between the generator and the discriminator. The parameter weights are analyzed in combination with the optimized multi-head attention mechanism, and the model display and editing priority are dynamically adjusted. After receiving the user's editing instructions, the geometric shape change, material replacement and other visual editing are completed based on the optimized model, and then the edited model is rendered in real time and the relevant data is stored. The system correspondingly sets six modules, including parameter collection and structured processing, model construction, and model optimization, and each module works together. The present invention deeply integrates the generative adversarial network and the optimized multi-head attention mechanism to achieve accurate construction and efficient editing and management of three-dimensional models of enterprise production scenarios.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional visualization of enterprise production, and in particular to a three-dimensional visualization editing and management method and system based on digital twins. Background Art

[0002] As manufacturing undergoes a digital transformation, companies are increasingly demanding more refined and visualized production management. Digital twin technology, with its ability to accurately represent physical entities, has become a key path to efficient production management. This has spawned a 3D visualization editing and management method and system based on digital twins.

[0003] Traditional production management models rely heavily on two-dimensional charts and simple data reports, making it difficult to intuitively present the spatial layout and dynamic interactions of complex production systems. This makes them unable to meet companies' needs for real-time monitoring and in-depth analysis of the entire production process. Some 3D visualization solutions, however, lack deep integration of production parameters during model construction, resulting in a disconnect between the 3D models and the actual production scene, making it difficult to accurately reflect key information such as equipment operation and material flow during the production process.

[0004] Existing technologies also have significant shortcomings in model optimization and editing. Conventional 3D models are difficult to dynamically adjust and optimize based on real-time changes in production parameters, making them unable to adapt to the uncertainties of the production process. Furthermore, during editing operations, there is a lack of effective weight analysis mechanisms, making it difficult to distinguish the degree to which different production parameters affect model editing. This results in a lack of targeted and efficient editing operations, making it impossible to meet the flexible and ever-changing production management needs of enterprises. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a three-dimensional visualization editing and management method and system based on digital twins.

[0006] The technical solution adopted by the present invention is a three-dimensional visual editing and management method based on digital twins, comprising the following steps:

[0007] Step S1: Collect equipment operating parameters, material transportation parameters, and process execution parameters during the enterprise's production process, and perform structured processing on the collected parameters according to a pre-set data organization structure;

[0008] Step S2: For each of the enterprise production parameters after structured processing, an initial 3D model of the enterprise production scene is constructed using 3D modeling technology, and meshing and texture mapping operations are performed on the initial 3D model;

[0009] Step S3: Using the generative adversarial network architecture, the constructed basic digital twin model is used as the input data of the generator. Through the continuous adversarial game training process between the generator and the discriminator, the initial 3D model is refined and optimized in terms of detailed features and key features are strengthened, so that the optimized 3D model can fit the actual production scenario of the enterprise;

[0010] Step S4: Introducing an optimized multi-head attention mechanism, an in-depth analysis of the weight coefficients of various enterprise production parameters is performed to determine the importance of different parameters in the 3D visualization editing management process. Based on the difference in importance, the display of the 3D model and the priority of the editing operation are dynamically adjusted;

[0011] Step S5: Receive editing instructions from the user, and based on the editing instructions and the 3D model optimized by the generative adversarial network and adjusted by the multi-head attention mechanism, perform visual editing operations on the 3D model, such as changing the geometric shape, replacing material attributes, and replanning the spatial layout.

[0012] Step S6: The edited three-dimensional model is rendered in real time, and the rendering results are presented to the user in a visual form. At the same time, the editing operation process and the modified enterprise production parameters are stored and recorded for call query and analysis research.

[0013] Furthermore, in step S3, the generative adversarial network architecture adopts the following model formula:

[0014] ;

[0015] in, represents the adjusted generator, For the generator, Indicates the A discriminator, is the number of discriminators; is the weight adjustment coefficient corresponding to different discriminators; For the generator and The discriminant loss function between the discriminators is used to measure the difference between the 3D model generated by the generator and the real model; is the production constraint loss function based on the enterprise production parameters, Indicates the The discriminator focuses on Production parameters of each enterprise, This function is used to ensure that the three-dimensional model generated by the generator meets the constraints specified by the enterprise production parameters.

[0016] Furthermore, in step S4, the optimized multi-head attention mechanism adopts the following model formula:

[0017] ;

[0018] in, Represents a single attention calculation result, are respectively the query vector, key vector and value vector obtained by nonlinear transformation of various parameters produced by the enterprise, is the value vector dimension, softmax is the activation function, Represents element-wise multiplication; Based on enterprise production parameters The constructed attention adjustment matrix is used to adjust the attention distribution according to the parameter characteristics; Indicates the Attention calculation results, Count the number of attentions, is the final output of the multi-head attention mechanism, Aggregate is the aggregation operation, is the final output weight matrix.

[0019] Furthermore, in step S5, when implementing geometric shape changes on the three-dimensional model, based on the results of the generative adversarial network on enhancing the local details of the three-dimensional model and combined with the optimized multi-head attention mechanism to determine the weight distribution of production process parameters, the geometric structure in the three-dimensional model that is associated with the production process parameters is precisely modified. In the modification process, the modified geometric shape is verified for applicability with reference to the equipment movement space range defined by the equipment operating parameters of the enterprise and the material transmission path requirements specified by the material flow parameters.

[0020] Furthermore, in step S5, when replacing material properties, a generative adversarial network is used to generate a virtual material library that meets the requirements of the light intensity changes, temperature and humidity influence characteristics of the enterprise production environment. Through the optimized multi-head attention mechanism, a comprehensive analysis is performed on the material contact wear conditions reflected by the material flow parameters and the equipment surface temperature change factors reflected by the equipment operation parameters. Materials suitable for the current production scenario are screened out from the virtual material library, and targeted adjustments are made to the reflectivity and transparency optical properties of the material based on the energy utilization efficiency requirements associated with the energy consumption parameters of the enterprise's production.

[0021] Furthermore, in step S5, when performing the visual editing operation of re-planning the spatial layout, the weight evaluation results of various enterprise production parameters are evaluated based on the optimized multi-head attention mechanism, and the importance level ranking of equipment and material storage areas in the three-dimensional model is determined. Combined with the multiple spatial layout candidate schemes generated by the generative adversarial network, the spatial layout schemes are screened and optimized according to the spatial range limited by the site area parameters of the enterprise's production and the shortest transportation distance requirements planned by the logistics transportation path parameters. At the same time, the equipment maintenance space requirements involved in the equipment operation parameters and the material storage and access convenience requirements associated with the material flow parameters are taken into consideration, and the selected spatial layout scheme is further refined.

[0022] Furthermore, in step S6, when the edited three-dimensional model is rendered in real time, based on the richness of details and prominent features presented by the three-dimensional model after generative adversarial network optimization, combined with the model display priority order corresponding to different enterprise production parameters determined by the optimized multi-head attention mechanism, a hierarchical rendering technology strategy is adopted. For the model parts that are more important and closely related to key production parameters, a high-precision rendering algorithm is used to perform detailed rendering processing. For the model parts that are less important and have less impact on the production process, a simplified rendering algorithm is used for fast rendering processing. At the same time, the rendering effect of the equipment model is dynamically updated according to the real-time operating status of the equipment reflected by the equipment operating parameters.

[0023] Furthermore, in step S6, when storing and recording the editing operation process and the modified enterprise production parameters, an index system based on the parameter-operation association relationship is constructed. The index system is based on the feature change information generated by the generative adversarial network in the process of optimizing the three-dimensional model, and the weight evolution data obtained by the optimized multi-head attention mechanism for parameter importance analysis. It records in detail the set of enterprise production parameters involved in each editing operation, the initial value and the modified value of the parameter, and the causal relationship between the editing operation and the production parameters, so as to facilitate subsequent rapid retrieval and in-depth analysis.

[0024] Furthermore, during the entire process of executing steps S1 to S6, real-time data of various production parameters of the enterprise are periodically collected at predetermined time intervals, and the real-time collected data are input into the generative adversarial network. The basic digital twin model is updated and iterated based on the dynamic learning ability of the generative adversarial network. At the same time, the weights of various parameters in the updated model are recalculated and dynamically adjusted through the optimized multi-head attention mechanism. According to the weight adjustment results, the display effect, editing operation process and rendering processing method of the three-dimensional model are correspondingly optimized and improved, and the three-dimensional visualization editing management system is continuously optimized and upgraded.

[0025] A three-dimensional visual editing and management system based on digital twins, the system comprising:

[0026] The enterprise production parameter collection and structured processing module is used to collect various parameters in the enterprise production process and implement structured processing according to the preset data structure;

[0027] A parameter-driven 3D model construction module is connected to the enterprise production parameter acquisition and structured processing module, which uses 3D modeling technology to construct an initial 3D model of the enterprise production scene based on the structured production parameters of the enterprise, and performs meshing and texture mapping;

[0028] A generative adversarial network-driven model optimization module, connected to the parameter-driven 3D model construction module, uses the generative adversarial network to optimize details and enhance features of the initial 3D model;

[0029] A parameter weight analysis module that optimizes the multi-head attention mechanism is connected to the enterprise production parameter acquisition and structured processing module and the generative adversarial network-driven model optimization module, respectively. The optimized multi-head attention mechanism is introduced to perform weight analysis on various enterprise production parameters and adjust the display and editing priority of the three-dimensional model.

[0030] A user-instruction-driven visual editing operation module is connected to the parameter weight analysis module for optimizing the multi-head attention mechanism, receives editing instructions from the user, and performs visual editing operations on the three-dimensional model;

[0031] The real-time rendering and parameter storage management module is connected to the user-instruction-driven visual editing operation module to render the edited three-dimensional model in real time and display it to the user, while storing and managing the editing operations and modified enterprise production parameters.

[0032] Beneficial effects: The present invention proposes a three-dimensional visual editing and management method and system based on digital twins. At the model construction level, the generative adversarial network can perform fine optimization and feature enhancement on the initial three-dimensional model according to the enterprise production parameters through adversarial game training between the generator and the discriminator. Compared with the problem of the three-dimensional model being out of touch with the actual scene in the traditional solution, the model generated by this method can accurately map the real production scene; at the same time, the optimized multi-head attention mechanism performs weight analysis on various production parameters to clarify the importance of different parameters in visual editing and management, changing the lack of pertinence in the editing of traditional solutions, making the display and editing operations of the three-dimensional model more in line with the production management needs of the enterprise. During the visual editing operation, whether it is geometric shape modification, material replacement or spatial layout adjustment, the system can combine the optimization results of the generative adversarial network with the parameter weights determined by the multi-head attention mechanism, and refer to parameters such as equipment operation and material flow to perform precise operations and feasibility verification to achieve efficient editing. In terms of rendering and storage, based on the optimization results and parameter weights of the generative adversarial network, layered rendering technology is used to improve rendering efficiency and quality, and a parameter-operation association index system is constructed to facilitate data retrieval and analysis. In addition, by periodically collecting real-time data to drive dynamic updates and iterations of the model, the 3D visualization editing and management process is continuously optimized, providing enterprises with more efficient, accurate, and intelligent production management tools. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of the method steps of the present invention;

[0034] Figure 2 This is a diagram of the system module composition of the present invention. DETAILED DESCRIPTION

[0035] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown in FIG, a three-dimensional visual editing and management method based on digital twins includes the following steps:

[0037] Step S1: Collect equipment operating parameters, material transportation parameters, process execution parameters, etc. in the enterprise's production process, and perform structured processing on the collected parameters according to a pre-set data organization structure;

[0038] Specifically, this step is the fundamental data acquisition phase of the entire method. Through various sensors and monitoring equipment deployed at the enterprise's production site, real-time data is collected on equipment operating parameters such as temperature, pressure, and rotational speed; parameters such as flow rate, velocity, and path during material transportation; and time points and process parameter values during process execution. These parameters cover every aspect of enterprise production and serve as the raw data foundation for building the digital twin model. After collection, these parameters are classified, coded, and standardized according to a pre-set hierarchical and associative data organization structure to form a structured data set, providing standardized and organized data input for subsequent model construction.

[0039] Different collection technologies are used for different types of parameters. For example, equipment operating parameters are monitored in real time using sensors installed in key locations; material transport parameters are collected using logistics tracking systems and flow monitoring devices. Regarding structured data processing, a data dictionary and metadata model are established based on the logical relationships within the company's production processes and the inherent connections between data. This ensures that the collected raw parameters can be converted into a structured data format that can be recognized and processed by computers, laying a solid data foundation for subsequent steps.

[0040] Step S2: For each of the enterprise production parameters after structured processing, an initial 3D model of the enterprise production scene is constructed using 3D modeling technology, and meshing and texture mapping operations are performed on the initial 3D model;

[0041] Specifically, this step is based on the structured production parameters obtained in the S1 stage, and uses professional 3D modeling software and technology to present the physical entities such as equipment, facilities, and materials in the enterprise production scene in a three-dimensional digital form to construct an initial 3D model. This model not only contains the geometric shape information of the physical entity, but also gives the model dynamic properties through association with production parameters. The subsequent meshing operation is to divide the surface of the 3D model into multiple small mesh modules to facilitate subsequent numerical calculations and rendering processing; texture mapping is to map the surface texture information of the real object to the 3D model to enhance the realism and visualization of the model.

[0042] Select appropriate 3D modeling tools and algorithms based on the scale and complexity of your company's production. For large, complex production scenarios, employ parametric modeling methods to rapidly generate different versions of the model by adjusting model parameters. For meshing, select the appropriate mesh density and type based on the model's accuracy requirements and computational resource constraints. Texture mapping, based on the surface features of objects in the actual production environment, collects or generates corresponding texture images and applies them to the 3D model surface through coordinate mapping. This ensures that the constructed initial 3D model accurately reflects the physical characteristics and spatial relationships of the company's production scenario.

[0043] Step S3: Using the generative adversarial network architecture, the constructed basic digital twin model is used as the input data of the generator. Through the continuous adversarial game training process between the generator and the discriminator, the detailed features of the initial 3D model are refined and the key features are strengthened. This allows the optimized 3D model to more accurately fit the actual production scenario of the enterprise.

[0044] Specifically, this step introduces the Generative Adversarial Network (GAN) architecture, leveraging its powerful generative capabilities and adversarial training mechanism to optimize and upgrade the initial 3D model constructed in the S2 phase. The generator uses the basic digital twin model as input and attempts to generate a more realistic and detailed 3D model; the discriminator is responsible for determining whether the generated model conforms to the characteristics of the company's actual production scenario. Through the continuous adversarial game between the two, the generator gradually learns the characteristic distribution of real production scenarios, thereby making fine adjustments to the initial 3D model, strengthening key features in the model, such as the key structure of the equipment and the flow path of the material, while optimizing the detailed performance of the model, so that the final generated 3D model can be highly close to the company's actual production scenario in terms of geometry, texture details, and dynamic characteristics.

[0045] First, a suitable GAN network structure and loss function are designed based on the characteristics and needs of the enterprise's production scenarios. To ensure the effectiveness and stability of model optimization, a multi-scale training strategy and progressive generation method are adopted. During training, the parameters of the generator and discriminator are continuously adjusted to achieve a balance between the two, thereby achieving optimal model optimization results. This adversarial training mechanism effectively addresses the problems of insufficient model accuracy and missing details that exist in traditional 3D modeling methods, providing a high-quality 3D model foundation for subsequent visual editing and management.

[0046] Step S4: Introducing an optimized multi-head attention mechanism, an in-depth analysis of the weight coefficients of various enterprise production parameters is performed to clarify the importance of different parameters in the 3D visualization editing management process. Based on the difference in importance, the display of the 3D model and the priority of the editing operation are dynamically adjusted;

[0047] Specifically, this step uses an optimized and improved multi-head attention mechanism to conduct in-depth analysis of various parameters in the company's production process. Different production parameters have different degrees of influence on 3D visual editing and management. Through the multi-head attention mechanism, the weight coefficient of each parameter can be automatically learned and calculated to determine its level of importance in the entire process. Based on these weight coefficients, the system can dynamically adjust the display of the 3D model, presenting the model parts corresponding to important parameters to the user in a more prominent manner; at the same time, during the editing process, the editing priority will be sorted according to the importance of the parameters, ensuring that users can prioritize the parameters and model parts that are most critical to production management.

[0048] Enterprise production parameters are vectorized and fed into a multi-head attention network for processing. The network uses multiple attention heads to perform parallel computations, capturing the correlations and importance between parameters from different perspectives. To improve the accuracy and adaptability of the attention mechanism, the traditional attention calculation method has been optimized, introducing a parameter adaptive adjustment strategy and context-aware mechanisms. This optimized multi-head attention mechanism enables the system to more accurately identify key production parameters, enabling dynamic adjustment of 3D model display and editing priorities, thereby enhancing the system's intelligence and management efficiency.

[0049] Step S5: Receive editing instructions from the user, and based on the editing instructions and the 3D model optimized by the generative adversarial network and adjusted by the multi-head attention mechanism, perform visual editing operations on the 3D model, such as changing the geometric shape, replacing material attributes, and replanning the spatial layout;

[0050] Specifically, this step is the core link for the system to realize user interaction and model editing. When the user issues an editing instruction, the system first parses the instruction content to determine the type of editing operation and specific requirements to be performed. Then, combined with the high-precision three-dimensional model optimized by the generative adversarial network in the S3 stage and the parameter importance determined by the multi-head attention mechanism in the S4 stage, the corresponding visual editing operations are performed on the three-dimensional model. In terms of geometric shape changes, the system will accurately modify the geometric structure of the model according to the requirements of production process parameters and equipment operating parameters; when replacing material properties, it will consider the influence of material flow parameters and environmental parameters, select appropriate materials and adjust their properties; when re-planning the spatial layout, it will comprehensively consider factors such as site area, logistics routes, equipment maintenance, etc. to ensure that the new layout plan meets the actual needs of enterprise production.

[0051] To ensure the accuracy and feasibility of editing operations, the system verifies in real time during the editing process whether the results meet the constraints of the company's production parameters. Operation preview and simulation evaluation functions have been introduced, allowing users to preview the effects and potential impacts of operations before officially implementing them. Furthermore, to improve editing efficiency, a variety of editing tools and shortcuts are provided, supporting advanced features such as batch editing and parametric editing. This allows users to more conveniently complete visual editing of 3D models, meeting the diverse needs of enterprise production management.

[0052] Step S6: The edited three-dimensional model is rendered in real time, and the rendering results are presented to the user in a visual form. At the same time, the editing operation process and the modified enterprise production parameters are stored and recorded for call query and analysis research.

[0053] Specifically, this step is responsible for rendering the edited 3D model in real time and presenting the rendering results to the user in an intuitive and visual manner. During the real-time rendering process, the system calculates the display effect of the model on the screen based on factors such as the model's geometric structure, material properties, and lighting conditions, generating high-quality images. In order to improve rendering efficiency and quality, a hierarchical rendering technology strategy is adopted. Different rendering accuracies and algorithms are used for different parts of the model based on the parameter importance and model display priority determined in the S4 stage. For important model parts, a high-precision rendering algorithm is used to ensure clear details; for less important parts, a simplified algorithm is used to increase rendering speed.

[0054] In terms of presentation, the system provides multiple viewing modes and interaction methods, allowing users to observe 3D models from different angles and perform operations such as zooming, rotating, and panning. Furthermore, the editing process and modified enterprise production parameters are recorded and stored in detail, establishing a parameter-operation correlation index system. These records not only include the specific content and results of the operation, but also the changes in production parameters before and after the operation, providing a rich data resource for subsequent production analysis, process optimization, and decision support. This approach enables traceability of 3D model editing operations and dynamic management of production parameters, improving the informationization level of enterprise production management and the scientific nature of decision-making.

[0055] Preferably, in step S3, the generative adversarial network architecture adopts the following model formula:

[0056] ;

[0057] in, represents the adjusted generator, For the generator, Indicates the A discriminator, is the number of discriminators; is the weight adjustment coefficient corresponding to different discriminators; For the generator and The discriminant loss function between the discriminators is used to measure the difference between the 3D model generated by the generator and the real model; is the production constraint loss function based on the enterprise production parameters, Indicates the The discriminator focuses on Production parameters of each enterprise, This function is used to ensure that the three-dimensional model generated by the generator meets the constraints specified by the enterprise production parameters.

[0058] Specifically, a more complex training mechanism was constructed by introducing a multi-discriminator architecture and a production constraint loss function. In this mechanism, multiple discriminators focus on different types of enterprise production parameters, each with a specific weight adjustment coefficient, and the training direction can be dynamically adjusted based on the importance of the parameter. The discriminant loss function is used to measure the difference between the generated model and the real scene, while the production constraint loss function ensures that the generated 3D model meets the physical constraints of the enterprise production parameters. Through this multi-objective optimization approach, the generator can learn a more accurate representation of the production scene features, improve the model's fit to the actual production environment, and ensure the feasibility of the generated results in engineering practice.

[0059] Preferably, in step S4, the optimized multi-head attention mechanism adopts the following model formula:

[0060] ;

[0061] in, Represents a single attention calculation result, are respectively the query vector, key vector and value vector obtained by nonlinear transformation of various parameters produced by the enterprise, is the value vector dimension, softmax is the activation function, Represents element-wise multiplication; Based on enterprise production parameters The constructed attention adjustment matrix is used to adjust the attention distribution according to the parameter characteristics; Indicates the Attention calculation results, Count the number of attentions, is the final output of the multi-head attention mechanism, Aggregate is the aggregation operation, is the final output weight matrix.

[0062] Specifically, the introduction of an attention adjustment matrix and multi-round attention calculation enhances the dynamic analysis capabilities of enterprise production parameters. The attention adjustment matrix, constructed based on the characteristics of production parameters, adaptively adjusts the distribution of attention across different parameters, allowing the model to focus more on key production indicators. Multi-round attention calculation, through multiple iterations, gradually refines the assessment of parameter importance, improving the accuracy of weighted analysis. Finally, the results of multiple rounds of calculations are integrated through aggregation and converted into a final attention output using a weight matrix. This provides a more precise basis for decision-making in 3D model display and editing priority adjustment, enabling in-depth exploration and efficient utilization of production parameters.

[0063] Preferably, in step S5, when implementing geometric shape changes on the three-dimensional model, based on the results of local detail enhancement of the three-dimensional model by the generative adversarial network and the weight distribution of production process parameters determined by the optimized multi-head attention mechanism, precise modification operations are implemented for the geometric structures in the three-dimensional model that are associated with the production process parameters. In the modification process, the applicability verification operation is performed on the modified geometric shape with reference to the equipment motion space range defined by the equipment operating parameters of the enterprise and the material transmission path requirements specified by the material flow parameters.

[0064] Specifically, the local detail enhancement capabilities of the generative adversarial network are combined with the parameter weight analysis of the multi-head attention mechanism to form a closed-loop process for precise modification and applicability verification. When performing modifications, the system first determines the key geometric structures associated with the production process parameters based on the parameter weights, and then uses the detailed features learned by the generative adversarial network to perform targeted optimization of these structures. During the modification process, parameter constraints such as the equipment motion space and material transfer path are referenced in real time to ensure that the modified geometry meets actual production needs, avoid problems such as equipment interference or poor logistics caused by design changes, and improve the engineering practicality of 3D model editing.

[0065] Preferably, in step S5, when replacing material properties, a generative adversarial network is used to generate a virtual material library that meets the requirements of the characteristics of the enterprise's production environment, such as changes in light intensity, temperature and humidity, and the like. Through an optimized multi-head attention mechanism, a comprehensive analysis is performed on factors such as the material contact wear reflected by the material flow parameters and the equipment surface temperature changes reflected by the equipment operation parameters. Materials suitable for the current production scenario are screened out from the virtual material library, and targeted adjustments are made to the optical properties of the material, such as reflectivity and transparency, based on the energy utilization efficiency requirements associated with the energy consumption parameters of the enterprise's production.

[0066] Specifically, by constructing a virtual material library driven by a generative adversarial network and a material screening system driven by a multi-head attention mechanism, a deep correlation between material properties and production parameters is achieved. The generation of the virtual material library takes into account the impact of environmental factors such as lighting, temperature, and humidity on material performance. The material screening process comprehensively analyzes parameters such as material wear characteristics and equipment temperature fluctuations to ensure that the selected materials are suitable for actual production environments. Further adjustments to the material's optical properties based on energy consumption parameters optimize energy efficiency while ensuring product quality, demonstrating the role of digital twin technology in optimizing all production factors.

[0067] Preferably, in step S5, when performing the visual editing operation of re-planning the spatial layout, the weight evaluation results of various parameters of enterprise production are evaluated based on the optimized multi-head attention mechanism, and the importance ranking of equipment, material storage areas, etc. in the three-dimensional model is determined. Combined with the multiple spatial layout candidate schemes generated by the generative adversarial network, the spatial layout schemes are screened and optimized according to the spatial range limited by the site area parameters of the enterprise production and the shortest transportation distance requirements planned by the logistics transportation path parameters. At the same time, the equipment maintenance space requirements involved in the equipment operation parameters and the material storage and access convenience requirements associated with the material flow parameters are taken into consideration, and the selected spatial layout scheme is further refined.

[0068] Specifically, a two-level optimization strategy based on parameter weight evaluation and generative adversarial networks was proposed. First, a multi-head attention mechanism was used to determine the importance of factors such as equipment and storage areas, providing priority ranking for layout planning. A generative adversarial network was then used to generate multiple candidate solutions, which were initially screened based on hard constraints such as site area and logistics routes. Furthermore, soft requirements such as equipment maintenance space and material accessibility were further considered, and the selected solutions were refined to achieve a balanced optimization between space utilization efficiency and production operation convenience, providing scientific layout decision support for enterprises.

[0069] Preferably, in step S6, when the edited three-dimensional model is rendered in real time, according to the richness of details and prominent features presented by the three-dimensional model after generative adversarial network optimization, combined with the model display priority order corresponding to different enterprise production parameters determined by the optimized multi-head attention mechanism, a hierarchical rendering technology strategy is adopted. For the model parts that are more important and closely related to key production parameters, a high-precision rendering algorithm is used to perform detailed rendering processing. For the model parts that are less important and have less impact on the production process, a simplified rendering algorithm is used for fast rendering processing. At the same time, according to the real-time operating status of the equipment reflected by the equipment operating parameters, the rendering effect of the equipment model is dynamically updated.

[0070] Specifically, the system employs a layered rendering strategy based on model detail and parameter importance. Based on the richness of model detail after generative adversarial network optimization and the parameter priorities determined by the multi-head attention mechanism, the system divides the model into different levels of importance. For model sections corresponding to key production parameters, a high-precision rendering algorithm is used to ensure clear details; for less important sections, a simplified algorithm is used to improve rendering efficiency. Furthermore, the rendering effect is dynamically updated based on the real-time operating status of the equipment, ensuring that the visualization results can promptly reflect changes in the production site and provide users with an accurate and intuitive means of monitoring production status.

[0071] Preferably, in step S6, when storing and recording the editing operation process and the modified enterprise production parameters, an index system based on the parameter-operation association relationship is constructed. The index system is based on the feature change information generated by the generative adversarial network in the process of optimizing the three-dimensional model, and the weight evolution data obtained by the optimized multi-head attention mechanism for parameter importance analysis. It records in detail the set of enterprise production parameters involved in each editing operation, the initial value and the modified value of the parameter, and the causal relationship between the editing operation and the production parameters, so as to facilitate subsequent rapid retrieval and in-depth analysis.

[0072] Specifically, an index system based on parameter-operation relationships is constructed to enable bidirectional traceability between editing operations and production parameter changes. This index system records the parameter sets involved in each editing operation, the parameter value evolution process, and the causal relationship between the operation and the parameters. This information is derived from feature changes during the generative adversarial network optimization process and the weight evolution analysis of the multi-head attention mechanism. Through this structured storage method, companies can quickly retrieve historical editing records and deeply analyze the impact of parameter adjustments on the production process. This provides a data foundation for process optimization and decision support, and enhances the system's knowledge management capabilities.

[0073] Preferably, during the entire process of executing steps S1 to S6, real-time data of various parameters of enterprise production are periodically collected at predetermined time intervals, and the real-time collected data are input into the generative adversarial network. The basic digital twin model is updated and iterated based on the dynamic learning ability of the generative adversarial network. At the same time, the weights of various parameters in the updated model are recalculated and dynamically adjusted through the optimized multi-head attention mechanism. According to the weight adjustment results, the display effect, editing operation process and rendering processing method of the three-dimensional model are correspondingly optimized and improved, and the three-dimensional visualization editing management system is continuously optimized and upgraded.

[0074] Specifically, a continuous optimization mechanism is proposed. By periodically collecting real-time production data, it drives a generative adversarial network and a multi-head attention mechanism to perform model updates and parameter weight adjustments. In this mechanism, the generative adversarial network iteratively optimizes the underlying digital twin model based on new data, capturing dynamic changes in the production process; the multi-head attention mechanism recalculates parameter weights to adapt to shifts in production focus. This dynamic adjustment not only affects the model's display quality but also optimizes the editing process and rendering processing, ensuring that the system remains highly aligned with actual production practices. This enables the continuous evolution of the 3D visual editing and management system, providing enterprises with long-term, effective production management tools.

[0075] like Figure 2 As shown, a three-dimensional visual editing and management system based on digital twins includes:

[0076] The enterprise production parameter collection and structured processing module is used to collect various parameters in the enterprise production process and implement structured processing according to the preset data structure;

[0077] A parameter-driven 3D model construction module is connected to the enterprise production parameter acquisition and structured processing module, which uses 3D modeling technology to construct an initial 3D model of the enterprise production scene based on the structured production parameters of the enterprise, and performs meshing and texture mapping;

[0078] A generative adversarial network-driven model optimization module, connected to the parameter-driven 3D model construction module, uses the generative adversarial network to optimize details and enhance features of the initial 3D model;

[0079] A parameter weight analysis module that optimizes the multi-head attention mechanism is connected to the enterprise production parameter acquisition and structured processing module and the generative adversarial network-driven model optimization module, respectively. The optimized multi-head attention mechanism is introduced to perform weight analysis on various enterprise production parameters and adjust the display and editing priority of the three-dimensional model.

[0080] A user-instruction-driven visual editing operation module is connected to the parameter weight analysis module for optimizing the multi-head attention mechanism, receives editing instructions from the user, and performs visual editing operations on the three-dimensional model;

[0081] The real-time rendering and parameter storage management module is connected to the user-instruction-driven visual editing operation module to render the edited three-dimensional model in real time and display it to the user, while storing and managing the editing operations and modified enterprise production parameters.

[0082] A three-dimensional visualization editing and management method and system based on digital twins deeply integrates generative adversarial networks and optimized multi-head attention mechanisms, accurately solving the shortcomings of traditional technologies in model construction and visualization presentation. In the model construction stage, traditional solutions have the problem of disconnection between three-dimensional models and actual production scenarios. This method, through adversarial game training of the generator and discriminator in the generative adversarial network, can fine-tune the detailed features and strengthen the key features of the initial three-dimensional model based on enterprise production parameters such as equipment operation and material transportation, so that the final three-dimensional model is highly consistent with the actual production scenario and intuitively presents the spatial layout and dynamic interaction relationship of complex production systems. At the same time, the optimized multi-head attention mechanism performs weight analysis on various production parameters, breaking the limitation of traditional two-dimensional charts and simple data reports that are difficult to intuitively present information on the entire production process, allowing enterprises to clearly grasp the key factors of production.

[0083] In terms of visual editing and operation, existing technologies lack an effective weight analysis mechanism, resulting in blind and inefficient editing operations. This method closely combines the optimization results of the generative adversarial network with the parameter weights determined by the multi-head attention mechanism when users execute editing commands such as geometric shape modification, material replacement, and spatial layout adjustment. It then performs precise operations and feasibility verification based on parameters such as equipment operation and material flow. For example, when replacing materials, a virtual material library is generated based on the characteristics of the enterprise's production environment. Parameters are then used to select and adapt materials and adjust properties to ensure that editing operations meet actual production needs, greatly improving the pertinence and efficiency of editing.

[0084] In terms of data management and system updates, traditional technologies struggle to adapt to the uncertainties of the production process, and models cannot dynamically adjust to parameter changes. This system builds a parameter-operation association index system. Based on feature changes and parameter weight evolution data during model optimization, it records the causal relationship between editing operations and production parameters in detail, facilitating rapid retrieval and analysis for enterprises. It also collects real-time data at predetermined intervals, driving a generative adversarial network to dynamically update and iterate the model. Simultaneously, it leverages a multi-head attention mechanism to recalculate and adjust parameter weights, thereby optimizing the display, editing, and rendering processes of 3D models. This enables continuous system optimization and upgrades, providing enterprises with a stable, intelligent, and flexible production management solution.

[0085] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0086] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional visual editing and management method based on digital twins, characterized in that: The steps include: Step S1: Collect equipment operating parameters, material transportation parameters, and process execution parameters during the enterprise's production process, and perform structured processing on the collected parameters according to a pre-set data organization structure; Step S2: For each of the enterprise production parameters after structured processing, an initial 3D model of the enterprise production scene is constructed using 3D modeling technology, and meshing and texture mapping operations are performed on the initial 3D model; Step S3: Using the generative adversarial network architecture, the constructed basic digital twin model is used as the input data of the generator. Through the continuous adversarial game training process between the generator and the discriminator, the initial 3D model is refined and optimized in terms of detailed features and key features are strengthened, so that the optimized 3D model can fit the actual production scenario of the enterprise; Step S4: Introducing an optimized multi-head attention mechanism, an in-depth analysis of the weight coefficients of various enterprise production parameters is performed to determine the importance of different parameters in the 3D visualization editing management process. Based on the difference in importance, the display of the 3D model and the priority of the editing operation are dynamically adjusted; Step S5: Receive editing instructions from the user, and based on the editing instructions and the 3D model optimized by the generative adversarial network and adjusted by the multi-head attention mechanism, perform visual editing operations on the 3D model, such as changing the geometric shape, replacing material attributes, and replanning the spatial layout. Step S6: Render the edited 3D model in real time, present the rendering result to the user in a visual form, and store and record the editing process and the modified enterprise production parameters for query and analysis; In step S3, the generative adversarial network architecture adopts the following model formula: ， in, represents the adjusted generator, For the generator, Indicates the A discriminator, is the number of discriminators; is the weight adjustment coefficient corresponding to different discriminators; For the generator and The discriminant loss function between the discriminators is used to measure the difference between the 3D model generated by the generator and the real model; is the production constraint loss function based on the enterprise production parameters, Indicates the The discriminator focuses on Production parameters of each enterprise, The number of production parameters that each discriminator pays attention to. This function is used to force the 3D model generated by the generator to meet the constraints specified by the enterprise production parameters. In step S4, the optimized multi-head attention mechanism adopts the following model formula: ， in, Represents a single attention calculation result, are respectively the query vector, key vector and value vector obtained by nonlinear transformation of various parameters produced by the enterprise, is the value vector dimension, softmax is the activation function, Represents element-wise multiplication; Based on enterprise production parameters The constructed attention adjustment matrix is used to adjust the attention distribution according to the parameter characteristics; Indicates the Attention calculation results, Count the number of attentions, is the final output of the multi-head attention mechanism, Aggregate is the aggregation operation, is the final output weight matrix.

2. A three-dimensional visual editing and management method based on digital twins according to claim 1, characterized in that: In step S5, when implementing geometric shape changes on the three-dimensional model, based on the results of local detail enhancement of the three-dimensional model by the generative adversarial network and the weight distribution of production process parameters determined by the optimized multi-head attention mechanism, precise modification operations are implemented on the geometric structures in the three-dimensional model that are associated with the production process parameters. In the modification process, the applicability of the modified geometric shape is verified by referring to the equipment movement space range defined by the equipment operating parameters of the enterprise and the material transmission path requirements specified by the material flow parameters.

3. A three-dimensional visual editing and management method based on digital twins according to claim 1, characterized in that: In step S5, when replacing material properties, a generative adversarial network is used to generate a virtual material library that meets the requirements of the light intensity changes, temperature and humidity influence characteristics of the enterprise production environment. Through an optimized multi-head attention mechanism, a comprehensive analysis is performed on the material contact wear conditions reflected by the material flow parameters and the equipment surface temperature change factors reflected by the equipment operation parameters. Materials suitable for the current production scenario are screened out from the virtual material library, and targeted adjustments are made to the reflectivity and transparency optical properties of the material based on the energy utilization efficiency requirements associated with the energy consumption parameters of the enterprise's production.

4. A three-dimensional visual editing and management method based on digital twins according to claim 1, characterized in that: In step S5, when performing the visual editing operation of re-planning the spatial layout, the weight evaluation results of various enterprise production parameters are evaluated based on the optimized multi-head attention mechanism, and the importance ranking of equipment and material storage areas in the three-dimensional model is determined. Combined with the multiple spatial layout candidate schemes generated by the generative adversarial network, the spatial layout schemes are screened and optimized according to the spatial range limited by the site area parameters of the enterprise's production and the shortest transportation distance requirements planned by the logistics transportation path parameters. At the same time, the equipment maintenance space requirements involved in the equipment operation parameters and the material access convenience requirements associated with the material flow parameters are taken into consideration, and the selected spatial layout scheme is further refined.

5. The three-dimensional visual editing and management method based on digital twins according to claim 1 is characterized in that: In step S6, when performing real-time rendering on the edited three-dimensional model, based on the richness of details and prominent features presented by the three-dimensional model after generative adversarial network optimization, combined with the model display priority order corresponding to different enterprise production parameters determined by the optimized multi-head attention mechanism, a hierarchical rendering technology strategy is adopted. For the model parts that are more important and closely related to key production parameters, a high-precision rendering algorithm is used to perform detailed rendering processing. For the model parts that are less important and have less impact on the production process, a simplified rendering algorithm is used for fast rendering processing. At the same time, the rendering effect of the equipment model is dynamically updated according to the real-time operating status of the equipment reflected by the equipment operating parameters.

6. A three-dimensional visual editing and management method based on digital twins according to claim 1, characterized in that: In step S6, when storing and recording the editing operation process and the modified enterprise production parameters, an index system based on the parameter-operation association relationship is constructed. The index system is based on the feature change information generated during the three-dimensional model optimization process of the generative adversarial network and the weight evolution data obtained from the parameter importance analysis of the optimized multi-head attention mechanism. It records in detail the enterprise production parameter set involved in each editing operation, the initial value and the modified value of the parameter, and the causal relationship between the editing operation and the production parameters, so as to facilitate subsequent rapid retrieval and in-depth analysis.

7. The three-dimensional visual editing and management method based on digital twins according to claim 1 is characterized in that: During the entire process of executing steps S1 to S6, real-time data of various production parameters of the enterprise are periodically collected at predetermined time intervals, and the real-time collected data are input into the generative adversarial network. The basic digital twin model is updated and iterated based on the dynamic learning ability of the generative adversarial network. At the same time, the weights of various parameters in the updated model are recalculated and dynamically adjusted through the optimized multi-head attention mechanism. According to the weight adjustment results, the display effect, editing operation process and rendering processing method of the three-dimensional model are optimized and improved accordingly, and the three-dimensional visualization editing management system is continuously optimized and upgraded.

8. A three-dimensional visual editing and management method based on digital twins according to any one of claims 1 to 7, characterized in that: The method is implemented through a three-dimensional visual editing and management system based on digital twins, which includes: The enterprise production parameter collection and structured processing module is used to collect various parameters in the enterprise production process and implement structured processing according to the preset data structure; A parameter-driven 3D model construction module is connected to the enterprise production parameter acquisition and structured processing module, which uses 3D modeling technology to construct an initial 3D model of the enterprise production scene based on the structured production parameters of the enterprise, and performs meshing and texture mapping; A generative adversarial network-driven model optimization module, connected to the parameter-driven 3D model construction module, uses the generative adversarial network to optimize details and enhance features of the initial 3D model; A parameter weight analysis module that optimizes the multi-head attention mechanism is connected to the enterprise production parameter acquisition and structured processing module and the generative adversarial network-driven model optimization module, respectively. The optimized multi-head attention mechanism is introduced to perform weight analysis on various enterprise production parameters and adjust the display and editing priority of the three-dimensional model. A user-instruction-driven visual editing operation module is connected to the parameter weight analysis module for optimizing the multi-head attention mechanism, receives editing instructions from the user, and performs visual editing operations on the three-dimensional model; The real-time rendering and parameter storage management module is connected to the user-instruction-driven visual editing operation module to render the edited three-dimensional model in real time and display it to the user, while storing and managing the editing operations and modified enterprise production parameters.

Images