Large-scale finite element calculation result post-processing method and system

Abstract

The invention provides a large-scale finite element calculation result post-processing method and system, and the method comprises the steps: S1, analyzing finite element result data through data hierarchical preprocessing, generating a multi-level data structure and a spatial index, and supporting on-demand loading and rapid query, S2, carrying out the self-adaptive real-time rendering through GPU parallel calculation according to a user view angle parameter, the method comprises the following steps: S1, generating a high-frame-rate visualization result by dynamically selecting a data hierarchy, S3, realizing coordinate synchronization of a simulation result and a physical space through VR / AR equipment, and supporting multi-modal interaction and multi-user cooperation, and S4, exporting a rendering result and interaction data, and supporting multiple formats to meet engineering analysis requirements. According to the method, efficient visualization, real-time interaction and multi-user collaborative analysis of large-scale finite element results are finally realized, and the engineering analysis efficiency and the decision support capability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and system for post-processing large-scale finite element calculation results. Background Technology

[0002] Post-processing of large-scale finite element analysis (FEM) results is a crucial step in fields such as engineering analysis, aerospace, and automotive design. Its core lies in transforming complex computational data into intuitive graphical representations through visualization techniques, supporting engineers' analysis and decision-making. The criticality of this field lies in its direct impact on the efficiency and accuracy of engineering designs, as well as the reliability of the final product.

[0003] Large-scale finite element simulation has become a core tool in complex engineering design, but existing post-processing software faces four major bottlenecks in data visualization and interaction: 1. Large-scale data loading and rendering efficiency is extremely low: Traditional post-processing software adopts the "full loading + CPU rendering" mode. When processing 10GB of finite element result data, full loading takes more than 30 minutes and the rendering frame rate is less than 15fps. When rotating to view the stress distribution in a car collision simulation, the delay of a single rotation is more than 2 seconds, making it impossible to smoothly observe global stress changes. 2. High latency in real-time slicing and cross-section analysis: Engineers need to frequently perform slicing analysis on simulation results. Traditional software uses CPU to serially calculate slice data, and it takes 5-10 minutes to generate a stress cloud diagram of an arbitrary angle cross-section. Moreover, recalculation is required when adjusting the slice position, making real-time iterative analysis impossible. 3. Poor VR / AR compatibility and lack of immersive experience: Existing VR / AR post-processing solutions require more than 50% simplification of the original data, such as merging mesh nodes and reducing the resolution of the results, resulting in the loss of key details, such as the simplification of stress concentration areas around bolt holes, and spatial positioning delay of more than 100ms. When engineers perform gesture operations in VR, the simulation results are not synchronized with hand movements, making it impossible to achieve accurate immersive exploration. 4. Insufficient multi-dimensional interactive functions: Traditional software only supports basic zooming and rotation with mouse / keyboard, and cannot realize "multi-user collaborative annotation" and "virtual-real fusion analysis", such as superimposing simulated stress cloud map onto physical parts for comparison in AR, resulting in low efficiency of cross-team collaboration.

[0004] The industry urgently needs a post-processing technology solution that combines "real-time rendering of large-scale data, latency-free interaction, and high-precision VR / AR immersion" to support efficient analysis and collaborative exploration of ultra-large-scale finite element results. Summary of the Invention

[0005] This invention provides a post-processing method for large-scale finite element calculation results, mainly including: Parse the finite element result file, extract mesh topology data and physical field data, and convert the data into a standardized format; Construct a pyramid-shaped hierarchical data structure with multiple precision levels, where each precision level is divided into multiple grid blocks according to spatial coordinates; Generate spatial indices for grid blocks at each level and inter-level association mappings; Calculate user view parameters in real time, dynamically select the rendering level based on user view parameters, and use the GPU for parallel rendering; Responding to user-defined slice planes, the GPU is used to compute slice intersections in parallel and interpolate physical quantity values ​​to generate slice cloud maps in real time. Synchronize simulation results with physical space coordinates using VR / AR devices; Export the rendering results and interactive data.

[0006] This invention also relates to a visualization and immersive interactive system for post-processing large-scale finite element calculation results, the system comprising: The analytical unit is used to parse finite element result files, extract mesh topology data and physical field data, and convert the data into a standardized format; Data building units are used to construct a pyramid-shaped hierarchical data structure with multiple precision levels, where each precision level is divided into multiple grid blocks according to spatial coordinates; The association mapping unit is used to generate spatial indexes of grid blocks at each level and association mappings between levels; The rendering unit is used to calculate user view parameters in real time, dynamically select the rendering level based on the user view parameters, and use the GPU for parallel rendering. The slice cloud map production unit is used to respond to user-defined slice planes, utilize GPU to calculate slice intersections in parallel and interpolate physical quantity values ​​to generate slice cloud maps in real time. The coordinate synchronization unit is used to synchronize the simulation results with the physical space through VR / AR devices; The export unit is used to export the rendering results and interactive data.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a solution for business scenarios where large-scale finite element calculation results post-processing involves massive data volume, low real-time visualization efficiency, and insufficient interactive experience. It proposes an efficient real-time visualization and immersive interaction solution.

[0008] To address the issues of insufficient memory and slow response times caused by massive data volumes, this invention employs layered data preprocessing to construct multi-level data structures based on precision. Combining octree spatial partitioning and R-tree indexing, it enables on-demand loading and fast querying, significantly reducing memory consumption and improving data access efficiency. Simultaneously, it utilizes GPU parallel rendering technology to dynamically select data levels and optimize vertex shading and texture mapping. Combined with occlusion culling and smooth level transitions, it ensures high frame rate visualization effects. Furthermore, it achieves synchronization between simulation results and physical space coordinates through VR / AR devices, supports multimodal interaction such as gestures, voice, and touch, and enables real-time multi-user collaboration via network protocols, improving immersive experience and collaborative efficiency.

[0009] This invention ultimately achieves efficient visualization, real-time interaction, and multi-user collaborative analysis of large-scale finite element results, significantly improving engineering analysis efficiency and decision support capabilities. Attached Figure Description

[0010] Figure 1 This is a flowchart of the visualization and immersive interactive method for post-processing of large-scale finite element calculation results according to the present invention. Detailed Implementation

[0011] The technical solutions of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0012] like Figure 1 The visualization and immersive interactive method for post-processing large-scale finite element calculation results in this embodiment may specifically include: S101. Analyze the finite element result file, extract the mesh topology data and physical field data, and convert the data into a standardized format.

[0013] Data format parsing and standardization It integrates the result formats of mainstream finite element solvers (ANSYS.odb, LS-DYNA.d3plot, Abaqus.fil, Fluent.cas) and extracts core data using a dedicated parser: Mesh topology data: node coordinates (X / Y / Z), cell connection relationships (cell-node mapping table); Physics field results data: physical quantity values ​​of each node / element (such as stress σ_xx / σ_yy / σ_zz, temperature T, flow velocity v), time step sequence (multi-frame data during dynamic simulation); The data is converted into a standardized binary format (.femgdb) that associates "grid blocks" with physical quantities, supporting parallel reading. Parsing 10GB of data takes ≤10 minutes, while traditional software takes 30 minutes.

[0014] First, raw data is obtained from the finite element analysis (FEM) results file. File parsing techniques are used to decompose the file structure, yielding mesh topology data and physical field data. For the mesh topology data, graph theory algorithms are used to analyze node and element connectivity to determine topological integrity. Based on topological integrity, if missing nodes or elements are detected, interpolation methods are used to supplement the data, resulting in repaired mesh topology data. Scalar and vector fields are extracted from the physical field data, and data cleaning techniques are used to remove outliers, resulting in standardized physical field data. For the repaired mesh topology data and standardized physical field data, a data mapping method is used to establish a correspondence between them, generating a unified data structure. Based on this unified data structure, the mesh topology data and physical field data are converted to a unified data format. A data verification algorithm is used to check the converted unified data format, determining data consistency and integrity, resulting in the final standardized data.

[0015] S102. Construct a pyramid-shaped hierarchical data structure with multiple precision levels, where each precision level is divided into multiple grid blocks according to spatial coordinates.

[0016] Specifically, the stratification strategy is as follows: the result data is divided into five levels, L0 (original precision) to L4 (1 / 16 of the original precision), based on data precision. The amount of data in each level decreases exponentially, as shown in Table 1. Table 1 Block partitioning: The grid data of each level is divided into equally sized cubic grid blocks (e.g., 256×256×256 nodes / block) according to spatial coordinates (X / Y / Z axes). Each block is stored independently and supports on-demand loading. For example, 5 million grid nodes at level L0 can be divided into 32 grid blocks, and only 6 blocks within the user's field of view are loaded at a time, reducing the amount of data loaded by 81.25%.

[0017] As another implementation method, multi-scale grid data from finite element analysis is acquired, decomposed into multiple precision levels according to spatial coordinates, and a pyramid-shaped hierarchical data structure is generated to determine the first layer of grid data. For the first layer of grid data, the spatial coordinate range of each grid block is extracted, and the adjacency relationship between grid blocks is determined to obtain the first grid block connection data. Based on the first grid block connection data, if overlaps or gaps are detected between grid blocks, the coordinate range is adjusted to obtain the second grid block data. For the second grid block data, the grid blocks of each precision level are encoded to generate the first compressed hierarchical grid data. Precision level information is extracted from the first compressed hierarchical grid data, and a mapping relationship between grid blocks and precision levels is established to obtain the first indexed hierarchical data. Based on the first indexed hierarchical data, the hierarchical grid data is converted into a standardized storage format to obtain serialized data.

[0018] S103. Generate spatial indexes for each level of grid blocks and inter-level association mappings.

[0019] Establish spatial indexes for grid blocks at each level: R-tree index, which records the spatial bounding box (minimum / maximum X / Y / Z coordinates) of each block, supporting quick query of "grid blocks within the user's view range"; Construct a mapping table between different levels: for example, a certain grid block in level L1 corresponds to 4 grid blocks in level L0, ensuring that when the view is zoomed in, it can smoothly switch from a low-precision level to a high-precision level without any screen tearing.

[0020] As another implementation method, multi-scale grid data is acquired, and multiple precision levels are recursively subdivided according to coordinate range to obtain hierarchical grid data. For the hierarchical grid data, grid blocks are divided according to spatial coordinates, and the spatial range of each grid block is determined to obtain block grid data. The spatial coordinates of the grid blocks are extracted from the block grid data. If the distance between grid blocks is less than a preset threshold, the nearest neighbor is queried to determine the connection relationship, resulting in grid connection data. For the grid connection data, the grid blocks at each precision level are compressed to generate compressed grid data. Hierarchical information is extracted from the compressed grid data, and the data is converted to a standard storage format to obtain serialized grid data.

[0021] S104: Calculate user view parameters in real time, dynamically select the rendering level based on the user view parameters, and use the GPU for parallel rendering.

[0022] Based on the user's perspective and interactive operations, the optimal data level and rendering strategy are dynamically selected, and GPU parallel computing is used to achieve latency-free rendering (frame rate ≥ 30fps): 2.1 Viewpoint-driven adaptive hierarchical selection 11. Real-time calculation of user-perspective parameters: 1. Viewpoint distance D: The spatial distance between the user's viewpoint and the center of the simulation model; 2. Field of view F: The range of model space covered by the current field of view; 12. Automatically select the rendering level based on preset rules: 1. When D > 2 × maximum model size (global overview): Selecting L3 / L4 level results in the smallest amount of rendering data and the fastest loading speed; 2. When 0.5 × maximum model size ≤ D ≤ 2 × maximum model size (view at the component level): select L1 / L2 level to balance accuracy and speed; 3. When D < 0.5 × maximum model size (for viewing local details): Select L0 / L1 level to ensure detail accuracy (e.g., use L0 level for stress concentration areas in bolt holes). 13. Smooth transition between layers: When switching layers, the old and new layers are superimposed and transitioned (transition time ≤ 50ms) to avoid abrupt changes in the screen.

[0023] 2.2 GPU Parallel Rendering Optimization 14. Vertex shader optimization: Utilize NVIDIA CUDA or AMD ROCm to compute the rendering properties of mesh nodes in parallel (such as color mapping - mapping stress values ​​to red-yellow-green gradient colors), the shading computation time for 1 million nodes is ≤1ms; 15. Texture Mapping Acceleration: Pre-generate GPU texture maps (Texture2DArray / Texture3D) from the physical field results of each layer (such as stress cloud map), and directly sample the map during rendering to avoid real-time color calculation, thereby increasing the frame rate to 45-60fps; Occlusion culling: Using the GPU's depth buffer, occluded mesh blocks outside the view (such as suspension components occluded by a car chassis) are culled in real time, reducing the amount of rendering data by 30%-50% and further improving the frame rate.

[0024] In another implementation, first position data and first orientation data are acquired from sensors, and first viewpoint parameters are determined by parsing the first position data and first orientation data. If the change in the first position data exceeds a preset threshold, the second orientation data is recalculated based on the changed second position data to obtain second viewpoint data. Based on the second viewpoint data, the current field of view is analyzed to determine which grid levels are within the field of view, obtaining first-level data. For the first-level data, a priority queue algorithm is used to determine high-priority grid levels, obtaining second-level data. Based on the second-level data, the rendering task is decomposed into parallel computing units to generate first task data. For the first task data, rendering tasks are allocated through the GPU parallel computing interface to obtain first rendering data. Based on the first rendering data, the grid levels are rendered in real time to generate first grid data. For the first grid data, if a change in the first viewpoint parameters is detected, third position data and third orientation data are acquired from sensors and updated to third viewpoint parameters to obtain third viewpoint data.

[0025] S105: Responding to the user-defined slice plane, the GPU is used to calculate the slice intersections in parallel and interpolate the physical quantity values ​​to generate slice cloud maps in real time.

[0026] Slice plane parameterization: Users define slice planes through an interactive interface (mouse / VR gestures) (inputting the plane equation ax+by+cz+d=0), and the system transmits the plane parameters to the GPU in real time; Parallel slice computation: The GPU generates slice results in real time through the following steps (time ≤100ms): 1. Spatial query: Parallel query of the grid blocks traversed by the slice plane, processing only the grid nodes within these blocks; 2. Intersection Calculation: For each grid edge that passes through the slice plane, calculate the coordinates of the intersection point between the edge and the plane; 3. Interpolation assignment: Based on the physical quantity values ​​(such as stress σ) of the nodes at both ends of the intersection, linear interpolation is used to calculate the physical quantity values ​​of the intersection. 4. Slice rendering: Generate slice polygons by connecting all intersection points according to the unit connection relationship, map the physical quantity colors, and display the slice cloud map in real time; Dynamic slicing support: When the user drags the slice plane to adjust its position, the GPU repeats the above calculations in real time, and the slice cloud map is updated without delay (traditional software takes 5 minutes).

[0027] In another implementation, first position and first orientation information are acquired from a sensor. By analyzing the first position and first orientation information, the geometric parameters of the first slice plane are determined, resulting in first slice plane data. Based on the first slice plane data, ray projection is used to emit rays from the pixels of the first slice plane and intersect them with preset volume data to obtain first intersection point coordinate data. For the first intersection point coordinate data, trilinear interpolation is performed in the CUDA kernel to calculate the physical quantity values ​​at the intersection point, resulting in first physical quantity distribution data. Based on the first physical quantity distribution data, a physical quantity distribution texture of the first slice plane is generated, resulting in first texture distribution data. Based on the first texture distribution data, a first slice cloud map is drawn in real time, resulting in first display cloud map data. Second position and second orientation information are acquired from the sensor. If the second position or second orientation information differs from the first position or first orientation information, the position of the second slice plane is recalculated, resulting in second slice plane data. Based on the second slice plane data, ray projection is used to emit rays from the pixels of the second slice plane and intersect them with preset volume data to obtain second intersection point coordinate data. For the second intersection point coordinate data, trilinear interpolation is performed to calculate the physical quantity values ​​at the intersection point, resulting in second physical quantity distribution data. Based on the second physical quantity distribution data, a physical quantity distribution texture of the second slice plane is generated, resulting in second texture distribution data. Then, based on the second texture distribution data, a second slice cloud map is drawn in real time, resulting in second display cloud map data.

[0028] The physical quantity value P is obtained by the formula P=(1-dx)(1-dy)(1-dz)V000+dx(1-dy)(1-dz)V100+(1-dx)dy(1-dz)V010+dxdy(1-dz)V110+(1-dx)(1-dy)dzV001+dx(1-dy)dzV101+(1-dx)dydzV011+dxdydzV111, where dx, dy, and dz are the normalized offsets from the intersection point to the voxel grid, and V000 to V111 are the values ​​of the surrounding 8 voxels.

[0029] Alternatively, the physical quantity value P can be calculated using interpolation methods: When the intersection point is within a tetrahedral element, barycentric coordinate interpolation is used, and the physical quantity value P = Σ(b_i * V_i), where b_i is the barycentric coordinate of the intersection point in the tetrahedron, and V_i is the physical quantity value of the vertex. When the intersection point is within a hexahedral element, isoparametric element shape function interpolation is used, and the physical quantity value is P = Σ(N_i(ξ,η,ζ) * V_i), where N_i is the shape function and (ξ,η,ζ) is the natural coordinate of the intersection point; When the intersection point is within a triangular element, area coordinate interpolation is used.

[0030] Responding to user-defined slicing planes, GPU parallel slicing computation based on unstructured mesh topology can also be performed using the following approach: Slicing plane parameter definition: The user defines the slicing plane equation ax+by+cz+d=0 through the interactive interface, and the system transmits the plane parameters to the GPU; GPU parallel slice computation process: Accelerated Spatial Queries: Utilize octree spatial indexing to quickly locate grid blocks intersecting the slice plane, and load only the data of these blocks into GPU memory; Parallel intersection calculation (CUDA kernel implementation): a. Each GPU thread processes one mesh cell (tetrahedral / hexahedral); b. Determine if the cell intersects the slice plane: Calculate the signed distance from each vertex of the cell to the plane. If the sign changes, then there is an intersection. c. For a tetrahedral element, at most one quadrilateral intersection plane (four intersection points) is generated. The coordinates of the intersection points are obtained by solving the parametric equations of the plane and the edge segments. d. For hexahedral elements, polygonal intersections may be generated, which can be calculated using a variant of the Marching Cubes algorithm; Interpolation of physical quantities based on element type: a. For intersection points within tetrahedral elements, use barycentric coordinate interpolation: Suppose the intersection point P lies in tetrahedron ABCD, calculate the coordinates of the barycenter (b_A, b_B, b_C, b_D). The physical quantity φ_P = b_A·φ_A + b_B·φ_B + b_C·φ_C + b_D·φ_D; b. For the intersection points within the hexahedral elements, solve for the natural coordinates (ξ,η,ζ) using the Newton-Raphson iteration, and then use shape function interpolation; Intersecting polygon generation and rendering: a. Connect the intersection points within the same unit to form a polygon; b. Perform color mapping based on the interpolated physical quantity values; c. Generate texture maps for direct rendering or pass them to the geometry shader; S106. Synchronize the simulation results with the coordinates of the physical space using VR / AR devices.

[0031] Achieve precise alignment between simulation results and VR / AR space, support multimodal interaction (gestures / voice) and multi-user collaboration, and create an immersive post-processing experience: 3.1 Spatial Positioning and Coordinate Synchronization VR spatial synchronization: 1. Device calibration: Obtain the spatial position of the user's head and hands using a VR tracker (such as the Valve Index tracker) (positioning accuracy ≤ 0.1mm). 2. Coordinate mapping: Establish a transformation matrix between the simulation model coordinate system and the VR physical coordinate system (using the 4-point calibration method—when the user clicks on 4 feature points of the simulation model in VR, they are aligned with 4 marker points in the real space), with a spatial positioning error ≤ 2mm; 3. Real-time update: The transformation matrix is ​​updated every frame (1 / 90 second) to ensure that the position of the simulation model is synchronized with the VR space and avoid "screen drift" caused by delay; AR spatial synchronization: 1. Entity Feature Recognition: Identify feature points of physical components (such as flange bolt holes on pressure vessels or weld points on car bodies) using the camera of AR devices. 2. Virtual-to-real alignment: Matching the corresponding feature points of the simulation model with the feature points of the solid component to achieve accurate superposition of simulation results (such as stress cloud diagrams) on the solid component, with a superposition deviation of ≤3mm; 3. Ambient light adaptation: Adjust the brightness and transparency of the simulation results according to the ambient light intensity collected by the AR camera to ensure a natural blend of virtual and real.

[0032] 3.2 Multimodal Immersive Interaction Gesture interaction: 1. Basic operations: Select model with fist gesture, pan with open palm, zoom with two fingers pinch, adjust view by rotating wrist, response latency ≤50ms; 2. Advanced operations: Gestures to mark feature areas (e.g., using the index finger to draw a circle to select the weld area with excessive stress), gestures to generate slicing planes (parallel movement of the palm defines the slicing direction); Voice interaction: 1. Command control: Voice input such as "Display maximum stress point" or "Generate Z=500mm section" will trigger the system to execute the corresponding operation in real time, with a recognition accuracy of ≥95%. 2. Parameter adjustment: Voice input "Adjust the stress threshold to 350MPa", the system automatically updates the color mapping range and highlights the area exceeding the threshold; Haptic feedback (based on VR gloves / force feedback controllers): 1. Risk Area Feedback: When the user's hand approaches a high-stress area (such as σ > 1.2σ_y), the VR glove generates vibration feedback (the vibration intensity increases with the stress value), reminding engineers to pay attention to the risk point; 2. Physical quantity perception: When touching different physical quantity areas (such as the high temperature area T>500℃), the gloves simulate temperature touch (achieved through heating elements) to enhance the immersive experience.

[0033] S107. Export rendering results and interactive data.

[0034] Collaborative Room Creation: Engineers can create collaborative rooms through the system (supporting 2-8 people to join at the same time) and set room permissions (e.g., the host can modify the model, while members can only view the annotations). Real-time data synchronization: The WebSocket protocol is used to synchronize the view, operation and annotation information of all users (such as the stress risk points marked by user A, which are visible to other users in real time), with a data synchronization delay of ≤100ms; Collaborative annotation: Supports multiple users to jointly annotate in VR (different users annotate with different colors), and the annotation content is automatically associated with simulation data (e.g., annotating "stress here is 420MPa, yield strength is 20%"). The annotation results can be exported as JSON format for subsequent report generation.

[0035] This invention also relates to a visualization and immersive interactive system for post-processing large-scale finite element calculation results, the system comprising: The analytical unit is used to parse finite element result files, extract mesh topology data and physical field data, and convert the data into a standardized format; Data building units are used to construct a pyramid-shaped hierarchical data structure with multiple precision levels, where each precision level is divided into multiple grid blocks according to spatial coordinates; The association mapping unit is used to generate spatial indexes of grid blocks at each level and association mappings between levels; The rendering unit is used to calculate user view parameters in real time, dynamically select the rendering level based on the user view parameters, and use the GPU for parallel rendering. The slice cloud map production unit is used to respond to user-defined slice planes, utilize GPU to calculate slice intersections in parallel and interpolate physical quantity values ​​to generate slice cloud maps in real time. The coordinate synchronization unit is used to synchronize the simulation results with the physical space through VR / AR devices; The export unit is used to export the rendering results and interactive data.

[0036] Hardware deployment requirements: 28. Rendering client: Intel i9-13900K CPU, 64GB DDR5 memory, NVIDIA RTX4090 GPU (24GB VRAM), 1TB NVMe SSD (for tiered storage). 29. VR devices: Valve Index headset (90Hz refresh rate), Leap Motion Ultraleap (gesture tracking accuracy 0.1mm), Manus VR gloves (haptic feedback); 30. AR devices: iPad Pro 2024 (LiDAR scanning accuracy 1mm), iPhone 15 Pro (48-megapixel camera); Server-side: Intel Xeon 8375C CPU, 256GB DDR5 memory, 4×NVIDIA A100 GPU, used for collaborative data forwarding and large-scale data preprocessing.

[0037] Taking "post-processing of nuclear power pressure vessel structural integrity simulation" as an example, the implementation process of this invention is explained in detail to verify its technical feasibility and effectiveness: 1. Input Data and Environment Preparation Finite element results data: Abaqus.fil format, simulation results of nuclear power pressure vessel (5 million mesh nodes, 12GB data), including stress (σ_xx / σ_yy / σ_zz), temperature (T), displacement (Ux / Uy / Uz), time step=100 (dynamic thermal stress analysis); Basic data: Pressure vessel geometric model (STEP format, dimensions: diameter 4m, height 10m, wall thickness 0.2m), material parameters (304 stainless steel, σ_y=205MPa, T_max=550℃); Hardware environment: Rendering client (i9-13900K, 64GB DDR5, RTX4090), VR devices (Valve Index, Leap Motion, Manus VR gloves), AR devices (iPad Pro 2024). Software environment: This invention system V1.0, Unity 2023, CUDA 12.2, SteamVR 1.28, ARKit 6.0.

[0038] 2. Pyramid-style hierarchical data preprocessing Data parsing and standardization: Import the .fil file, the system parses and extracts 5 million node coordinates, element connection relationships, and stress / temperature data for each time step, and converts it into .femgdb format, which takes 9 minutes; Layering and Block Partitioning: L0-L4 layers were constructed. The L0 layer (original precision) was divided into 32 grid blocks (each block has 156,000 nodes), and the L4 layer (6.25% precision) was divided into 2 grid blocks. The data volume was reduced from 12GB to 0.047GB (L4). Index building: Generate R-tree spatial indexes for each level of grid blocks, and record the bounding box coordinates of each block (e.g., X range: 0-1m, Y range: 0-1m, Z range: 0-1m for L0 block 1).

[0039] 3. Real-time visualization and interactive operation 3.1 Adaptive Rendering: Initial loading: The default loading level is L4, and a global overview of the pressure vessel is displayed within 2 seconds (temperature distribution: red area T=500℃, green area T=300℃). Zooming in: When the user zooms in with the mouse wheel to the "flange bolt hole area" (D=0.3m<0.5×4m=2m), the system automatically switches to the L0 level, loads 3 mesh blocks around the bolt holes, and the rendering frame rate is stable at 45fps without any stuttering. 3.2 Real-time slice analysis: Users can generate a "Z=5m" slice plane by dragging with the mouse. The GPU calculates the slice intersections and physical quantity values ​​in parallel and displays the slice stress cloud map within 80ms, showing the stress distribution of the flange section (maximum stress 380MPa, located around the bolt holes). Drag the slice plane to "Z=6m" and the slice cloud map will update in real time (delay ≤90ms) to compare the stress changes of different sections; 3.3VR Immersive Interaction: Spatial synchronization: Wearing the ValveIndex headset, a mapping between the simulation coordinate system and VR space was established by 4-point calibration (clicking the 4 flange bolt holes of the pressure vessel), with a positioning error of 1.8mm; Gesture operation: 1. Pinch to zoom: Zoom in on the bolt hole area to view stress concentration details (σ=380MPa>σ_y=205MPa); 2. Gesture circle: Use your index finger to circle the over-stress area, and the system will automatically label it as "Risk area: σ=380MPa, exceeding yield strength by 85%"; 3. Tactile feedback: When the hand approaches the area of ​​high stress, the ManusVR glove generates high-frequency vibration (vibration frequency 50Hz) to alert the user to the risk; Multi-user collaboration: Shanghai engineers create collaborative rooms and invite Beijing engineers to join. Beijing engineers can see the annotations of Shanghai engineers in real time and use voice input to "generate fatigue life analysis for this area". The system automatically associates stress data and triggers subsequent analysis. 3.4AR Virtual-Real Fusion: Scan the flange feature points (bolt holes) of the physical pressure vessel with the iPad Pro camera. After the system recognizes the data, it overlays the simulated stress cloud map onto the physical object with an overlay deviation of 2.5mm. Engineers can intuitively compare the "simulated stress distribution" with the "physical surface inspection results".

[0040] 4. Results Export and Validation Exporting slice contour plot: Export the stress contour plot of the "Z=5m" slice as a PNG format (300dpi resolution), which takes 15 seconds; Export annotation results: Export the risk annotations in VR as a JSON file, including annotation location (X=2m,Y=2m,Z=5m), stress value (380MPa), and annotator (Shanghai Zhang). Performance verification: Compared with traditional software, this invention takes 9 minutes to load 12GB of data (30 minutes for traditional software), generates slices in 80ms (5 minutes for traditional software), and has a VR positioning error of 1.8mm (10mm for traditional software), which meets engineering requirements.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for post-processing results of large-scale finite element calculations, characterized in that, The method comprises: parsing the finite element result file, extracting the grid topology data and the physical field data, and converting the data into a standardized format; constructing a pyramid layered data structure of multiple precision levels, wherein each precision level is divided into multiple grid blocks according to spatial coordinates; generating spatial indexes of grid blocks at each level and inter-level correlation mapping; calculating user perspective parameters in real time, dynamically selecting a rendering level based on the user perspective parameters, and using a GPU for parallel rendering; in response to a user-defined slice plane, using a GPU to calculate slice intersection points and interpolate physical quantity values in parallel, and generating a slice cloud image in real time; synchronizing the simulation results with the coordinates of the physical space through a VR / AR device; exporting rendering results and interaction data.

2. The method of claim 1, wherein, The parsing of the finite element result file, the extraction of the grid topology data and the physical field data, and the conversion of the data into a standardized format comprises: obtaining original data from the finite element result file, decomposing the file structure using file parsing technology to obtain grid topology data and physical field data; for the grid topology data, using graph theory algorithms to analyze the connection relationship between nodes and elements, and determining the integrity of the topology structure; according to the integrity of the topology structure, if a missing node or element is detected, supplementing the data through an interpolation method to obtain repaired grid topology data; extracting scalar fields and vector fields from the physical field data, and using data cleaning technology to remove outliers to obtain standardized physical field data; for the repaired grid topology data and the standardized physical field data, using data mapping methods to establish a correspondence between them, and generating a unified data structure; according to the unified data structure, converting the grid topology data and the physical field data into a unified data format; checking the converted unified data format through a data verification algorithm to judge the consistency and integrity of the data, and obtaining the final standardized data.

3. The method of claim 1, wherein, The construction of a pyramid layered data structure of multiple precision levels, wherein each precision level is divided into multiple grid blocks according to spatial coordinates, comprises: obtaining multi-scale grid data for finite element analysis, decomposing the data into multiple precision levels according to spatial coordinates to generate a pyramid layered data structure, and determining first layered grid data; for the first layered grid data, extracting the spatial coordinate range of each grid block therefrom, determining the adjacency relationship between the grid blocks, and obtaining first grid block connection data; according to the first grid block connection data, if it is detected that there is an overlap or gap between the grid blocks, adjusting the coordinate range to obtain second grid block data; for the second grid block data, encoding the grid blocks at each precision level to generate first compressed layered grid data; extracting precision level information from the first compressed layered grid data, establishing a mapping relationship between the grid blocks and the precision levels, and obtaining first index layered data; according to the first index layered data, converting the layered grid data into a standardized storage format to obtain serialized data.

4. The method of claim 1, wherein, The generation of spatial indexes of grid blocks at each level and inter-level correlation mapping comprises: obtaining multi-scale grid data, recursively subdividing multiple precision levels according to coordinate ranges to obtain layered grid data; The hierarchical grid data is divided into grid blocks according to spatial coordinates, the spatial range of each grid block is determined, and the divided grid data is obtained; The spatial coordinates of the grid blocks are extracted from the divided grid data, and if the distance between the grid blocks is less than a preset threshold, the nearest neighbor is queried to determine the connection relationship, and the grid connection data is obtained; The grid blocks of each precision level are compressed for the grid connection data, and the compressed grid data is generated; The hierarchical information is extracted from the compressed grid data, the data is converted into a standard storage format, and the serialized grid data is obtained.

5. The method of claim 1, wherein, The real-time calculation of the user perspective parameter is based on the user perspective parameter, and the rendering level is dynamically selected, and the GPU is used for parallel rendering, including: The first position data and the first direction data are obtained from the sensor, and the first perspective parameter is determined by analyzing the first position data and the first direction data; If the first position data changes more than a preset threshold, the second direction data is recalculated according to the changed second position data, and the second perspective data is obtained; According to the second perspective data, the current field of view range is analyzed, and it is judged which grid level is in the field of view, and the first level data is obtained; According to the first level data, the high priority grid level is determined by using the priority queue algorithm, and the second level data is obtained; According to the second level data, the rendering task is divided into parallel computing units, and the first task data is generated; The rendering task is distributed to the parallel computing interface for the first task data, and the first rendering data is obtained; According to the first rendering data, the grid level is drawn in real time, and the first grid data is generated; If the first perspective parameter changes are detected, the third position data and the third direction data are obtained from the sensor, and the third perspective parameter is updated, and the third perspective data is obtained.

6. The method of claim 1, wherein, The GPU is used to calculate the slice intersection point and interpolate the physical quantity value in parallel, and the slice cloud image is generated in real time, including: The first position and the first direction information are obtained from the sensor; The geometric parameters of the first slice plane are determined by analyzing the first position and the first direction information, and the first slice plane data is obtained; According to the first slice plane data, the ray projection is used to emit a ray from the first slice plane pixel to intersect with the preset body data, and the first intersection point coordinate data is obtained; According to the first intersection point coordinate data, the trilinear interpolation is performed, the physical quantity value at the intersection point is calculated, and the first physical quantity distribution data is obtained; According to the first physical quantity distribution data, the physical quantity distribution texture of the first slice plane is generated, and the first texture distribution data is obtained; According to the first texture distribution data, the first slice cloud image is drawn in real time, and the first display cloud image data is obtained; The second position and the second direction information are obtained from the sensor; If the second position or the second direction information is different from the first position or the first direction information, the second slice plane position is recalculated, and the second slice plane data is obtained; According to the second slice plane data, the ray projection is used to emit a ray from the second slice plane pixel to intersect with the preset body data, and the second intersection point coordinate data is obtained; According to the second intersection point coordinate data, the trilinear interpolation is performed, the physical quantity value at the intersection point is calculated, and the second physical quantity distribution data is obtained; According to the second physical quantity distribution data, a physical quantity distribution texture of the second slice plane is generated, and second texture distribution data is obtained; For the second texture distribution data, a second slice cloud picture is drawn in real time, and second display cloud picture data is obtained.

7. A system for post-processing results of large-scale finite element calculations, characterized by The system comprises: The analysis unit is used for analyzing the finite element result file, extracting the grid topology data and the physical field data, and converting the data into a standardized format; The data construction unit is used for constructing a pyramid type layered data structure of multiple precision levels, wherein each precision level is divided into multiple grid blocks according to spatial coordinates; The correlation mapping unit is used for generating spatial indexes of the grid blocks of each level and correlation mapping between levels; The rendering unit is used for calculating user visual angle parameters in real time, dynamically selecting a rendering level based on the user visual angle parameters, and performing parallel rendering by using a GPU; The slice cloud picture production unit is used for responding to a slice plane defined by a user, calculating slice intersection points and interpolating physical quantity values by using a GPU, and generating a slice cloud picture in real time; The coordinate synchronization unit is used for realizing coordinate synchronization between a simulation result and a physical space by using a VR / AR device; The export unit is used for exporting rendering results and interaction data.

Images