Multi-point hardness detection system and method for special equipment

Abstract

The invention discloses a multi-point hardness detection system and method for special equipment, and belongs to the technical field of safety detection of the special equipment. The method comprises the following steps: optimizing and generating a measuring point coordinate sequence based on an equipment CAD model and historical stress data; obtaining a calibration hardness matrix through high-density hardness detection; a GNN-FEA hybrid model is adopted to predict a stress field and correct a low-confidence region; inverting a stress concentration area by combining a CAD model; and dynamically optimizing the detection path and generating a visual decision report. According to the method, through deep fusion of high-precision hardness detection and intelligent stress field inversion, submillimeter positioning of a stress concentration area of the special equipment is realized, the detection precision is improved, and meanwhile, the detection efficiency is improved through a closed-loop optimization mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of special equipment safety detection, and in particular to a multi-point hardness detection system and method for special equipment. BACKGROUND

[0002] In the field of special equipment safety detection, traditional stress detection methods mainly rely on single-point or sparse multi-point hardness measurement combined with empirical formula to estimate stress distribution, which has the following technical bottlenecks: Insufficient detection accuracy: single-point hardness detection cannot reflect the stress gradient change and is difficult to capture local stress concentration phenomenon, resulting in high false detection rate of X-ray diffraction, ultrasonic wave and other verification methods; Poor model adaptability: existing hardness-stress correlation models are mostly based on ideal material assumptions and do not consider the influence of welding residual stress, plastic deformation and other actual working conditions, with prediction errors exceeding 10% on composite materials and special-shaped structures; Data fusion is missing: hardness detection data and device CAD model lack dynamic correlation mechanism, resulting in inability to effectively combine geometric features (such as welds, openings) for local correction during stress inversion; Closed-loop optimization blank: the detection process lacks a feedback mechanism based on confidence evaluation, and low-precision areas cannot trigger targeted re-inspection, affecting the overall evaluation reliability.

[0003] Although current research attempts to combine finite element analysis with hardness detection (such as CN110926936A), it still does not solve the problem of precise mapping of high-density detection data and multi-scale stress field. In particular for nuclear power pipelines, chemical containers and other critical equipment, existing technologies cannot locate sub-millimeter level stress concentration areas without stopping, with spatial resolution limited to more than 5mm, which cannot meet the detection accuracy requirements of high-risk areas in ASME BPVC and other standards. SUMMARY

[0004] The present application provides a multi-point hardness detection system and method for special equipment to solve the problem of how to accurately predict stress field distribution and identify stress concentration areas based on high-density multi-point hardness detection data and device CAD model through GNN-FEA hybrid model, and realize precise positioning and safety evaluation of potential risk areas of special equipment.

[0005] To solve the above technical problems, the present application provides a multi-point hardness detection method for special equipment, comprising: Obtain device CAD model and historical stress data, divide risk level and encrypt high stress area grid, and optimize to generate measurement point coordinate sequence; Perform hardness detection according to the measurement point coordinate sequence, calibrate surface topography interference and integrate spatial coordinates, and output calibrated hardness matrix; The calibrated hardness matrix and material parameters are input into the hybrid model to predict the stress field, low confidence areas are marked, and the stress distribution is corrected by finite element to generate a corrected stress field; The corrected stress field is matched with the CAD model, the peak stress area is inverted, and the welding residual stress is corrected to generate a stress concentration area report; According to the stress concentration report and confidence score, identify the need for new measurement points, dynamically update the robot path and feedback to the measurement point planning module; Combine the stress field and industry standard to divide the risk level, mark the over-limit area and integrate the CAD model to generate a visual decision report.

[0006] Further, the CAD model of the device and the historical stress data are obtained, the risk level is divided, and the high stress area grid is encrypted, and the measurement point coordinate sequence is optimized, including: Obtain the CAD model of the device and the historical stress distribution data, divide the risk level, and obtain the risk level atlas; Extract the high stress potential area from the risk level atlas, perform adaptive grid encryption processing, and obtain the initial measurement point sequence; Optimize the initial measurement point sequence for the robot path to generate a dynamically adjusted measurement point coordinate sequence.

[0007] Further, the hardness detection is performed according to the measurement point coordinate sequence, the surface topography interference is calibrated, and the spatial coordinates are integrated to output the calibrated hardness matrix, including: Obtain the measurement point coordinate sequence, control the hardness probe array to perform point-by-point indentation detection, and obtain the original hardness data; Extract the surface topography interference component from the original hardness data, perform laser confocal compensation calibration, and obtain the calibrated hardness value; Integrate the calibrated hardness value according to the spatial coordinates to output the calibrated hardness matrix.

[0008] Further, the calibrated hardness matrix and material parameters are input into the hybrid model to predict the stress field, low confidence areas are marked, and the stress distribution is corrected by finite element to generate a corrected stress field, including: Obtain the calibrated hardness matrix and material parameter library, input the GNN-FEA hybrid model, perform stress field prediction, and obtain the preliminary stress distribution.

[0009] Further, the preliminary stress distribution also includes: Extract the confidence score from the preliminary stress distribution, mark the low confidence area, and obtain the to-be-verified area coordinates; Perform local finite element encryption calculation on the to-be-verified area coordinates to generate a corrected stress field.

[0010] Further, the matching correction stress field and the CAD model, the peak stress area is inverted and the welding residual stress is corrected, and a stress concentration area report is generated, including: Obtain the corrected stress field and the equipment CAD model, perform geometric topology matching, and obtain stress-geometry correlation mapping; Extract the peak stress area from the stress-geometry correlation mapping, perform residual stress inversion calculation, and obtain high-risk stress coordinates; Correct the welding residual stress of the high-risk stress coordinates, and generate a stress concentration area report.

[0011] Further, the stress concentration report and the confidence score are used to identify the need for new measurement points, dynamically update the robot path, and feed back to the measurement point planning module, including: Obtain the stress concentration area report and the confidence score, identify the insufficient detection area, and obtain the need for new measurement points; Extract the coordinate increment from the new measurement point demand, dynamically update the robot path, and obtain the optimized measurement point sequence; Feed back the optimized measurement point sequence to the measurement point planning module to complete the closed-loop adjustment.

[0012] Further, the stress field and the industry standard are combined to divide the risk level, label the over-limit area, and fuse the CAD model to generate a visual decision report, including: Obtain the stress field and the industry stress threshold standard, divide the risk level, and obtain the risk heat map; Extract the over-limit area from the risk heat map, perform three-dimensional labeling and maintenance priority sorting, and obtain a maintenance suggestion table; Fuse the maintenance suggestion table with the equipment CAD model to generate a visual decision report.

[0013] Further, the calibrated hardness matrix and the material parameters are input into the hybrid model to predict the stress field, and the hybrid model is a GNN-FEA hybrid model for establishing the hardness-residual stress correlation relationship.

[0014] A multi-point hardness detection system for special equipment is applied to any one of the multi-point hardness detection methods for special equipment described above, and includes: A measurement point planning module is used to obtain the equipment CAD model and the historical stress data, divide the risk level, encrypt the high stress area grid, and optimize the generation of measurement point coordinate sequence; A hardness collection module is used to perform hardness detection according to the measurement point coordinate sequence, calibrate surface topography interference, integrate spatial coordinates, and output a calibrated hardness matrix; A stress calculation module is used to input the calibrated hardness matrix and the material parameters into the hybrid model to predict the stress field, mark the low confidence area, and correct the stress distribution through finite elements to generate a corrected stress field; The stress inversion module is used to match and correct the stress field with the CAD model, invert the peak stress region and correct the welding residual stress, and generate a stress concentration region report. The closed-loop optimization module is used to identify new measurement point requirements based on stress concentration reports and confidence scores, dynamically update the robotic arm path, and feed it back to the measurement point planning module. The decision output module is used to combine stress field and industry standards to classify risk levels, mark out-of-limit areas, integrate CAD models, and generate a visual decision report.

[0015] The key innovations of this invention include: (1) Couple graph neural network (GNN) with finite element analysis (FEA). GNN processes the nonlinear spatial correlation of hardness data, while FEA ensures the physical rationality of stress calculation.

[0016] (2) The detection density is dynamically adjusted by confidence score, and local mesh densification and robotic arm path replanning are automatically triggered in low confidence areas.

[0017] (3) Adaptive correction of welding residual stress calculation model based on equipment geometric topology features (such as weld angle and radius of curvature).

[0018] The following are its main beneficial effects: (1) Significantly improve the accuracy of stress field prediction: By introducing a hybrid model with physical constraints, the prediction deviation of traditional methods in high gradient stress fields such as welding area is effectively overcome, and the overall stress prediction error is controlled within 5%.

[0019] (2) Achieve intelligent identification and correction of low confidence regions: The confidence assessment based on the dynamic probability model can accurately locate the region that needs to be encrypted. Combined with adaptive grid encryption and variational correction, the identification accuracy of stress concentration regions reaches 92%.

[0020] (3) Forming a closed-loop optimization mechanism: The feedback chain of prediction-evaluation-correction significantly reduces the false negative rate to 3%, while reducing unnecessary duplicate detections, thereby improving the overall detection efficiency by 40%. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a multi-point hardness testing method for special equipment provided in this application embodiment; Figure 2 This is a structural block diagram of a multi-point hardness testing system for special equipment provided in an embodiment of this application. Detailed Implementation

[0022] Example 1: Refer to Figure 1This is a flowchart illustrating a multi-point hardness testing method for special equipment provided in an embodiment of the present invention. The process may include at least steps S100-S600: S100: Obtain the equipment CAD model and historical stress data, classify risk levels and densify the mesh in high-stress areas, and optimize the generation of measurement point coordinate sequences.

[0023] S200: Perform hardness testing according to the measurement point coordinate sequence, calibrate surface morphology interference and integrate spatial coordinates, and output a calibration hardness matrix.

[0024] S300: Input the calibration hardness matrix and material parameters into the hybrid model to predict the stress field, mark the low confidence region, and generate the corrected stress field by correcting the stress distribution through finite element method.

[0025] S400, matching and correcting the stress field with the CAD model, inverting the peak stress region and correcting the welding residual stress, generating a stress concentration region report.

[0026] S500 identifies new measurement point requirements based on stress concentration reports and confidence scores, dynamically updates the robotic arm path, and feeds it back to the measurement point planning module.

[0027] S600 combines stress field and industry standards to classify risk levels, marks out-of-limit areas and integrates CAD models to generate a visual decision report.

[0028] Step S100 includes at least steps S110-S130: S110. Obtain the equipment CAD model and historical stress distribution data, classify the regional risk levels, and obtain the risk level map.

[0029] The system receives 3D CAD model files of special equipment through a standardized data interface. These 3D CAD model files are stored in a universal format and contain complete geometric topological information of the equipment surface, material property annotations, and dimensional tolerance data for key components. Specifically, the model import process implements rigorous geometric integrity verification to ensure that the mathematical representations of all surfaces and boundaries are complete and error-free. Simultaneously, historical stress detection datasets are extracted from the equipment's lifecycle management database. These datasets contain stress data obtained using various detection methods under different service stages and operating conditions, specifically including residual stress distribution measured by X-ray diffraction, stress gradient data obtained by ultrasonic testing, and actual working stress variation curves recorded by strain gauges. All historical data undergoes timestamp alignment and unit standardization to ensure data consistency and comparability.

[0030] First, a multi-scale stress field mapping model based on finite element meshes is established. A high-order interpolation algorithm is used to accurately map discrete historical stress measurement data onto the surface mesh nodes of the equipment CAD model. Specifically, local mesh refinement is implemented for complex geometric regions to ensure the continuity of stress field transitions. Subsequently, a material fatigue damage accumulation calculation model is constructed, comprehensively considering stress amplitude, cycle number, and material SN curve characteristics. A nonlinear accumulation algorithm is used to calculate the damage factor value corresponding to each mesh element. The calculation process specifically considers the influence of multiaxial stress states, employing an equivalent stress conversion method to handle complex stress combinations. The final risk level classification is based on a preset damage threshold standard, dividing the equipment surface into three risk levels: high-risk area (damage factor exceeding the critical value of 80%), medium-risk area (damage factor between 50% and 80%), and low-risk area (damage factor below 50%), which are visualized in the 3D model using different color codes.

[0031] The calculated risk level data is precisely spatially registered with the original CAD model to establish a digital twin model containing complete stress risk information. Specifically, the map is stored using a hierarchical data structure: the bottom layer retains the original geometric information, the middle layer stores gridded stress data, and the top layer records risk level labels. The attribute data associated with each grid cell includes, but is not limited to: maximum principal stress value, stress cycle count, damage factor calculation results, material yield strength margin, and most recent detection timestamp. The map supports multi-dimensional querying and visualization rendering, and can dynamically adjust rendering accuracy and display content according to different display needs, providing comprehensive data support for subsequent detection planning.

[0032] S120. Extract high-stress potential areas from the risk level map, perform adaptive mesh refinement processing, and obtain the initial measurement point sequence.

[0033] The system first performs topological analysis on the risk level map to identify all continuous high-risk regions (color-coded in red). Specifically, it employs a region growing algorithm from image processing, progressively expanding and connecting adjacent meshes with the same risk level characteristics from a seed point. For each identified high-risk region, its geometric feature parameters are extracted simultaneously, including region area, perimeter, centroid location, and direction of maximum stress gradient. In particular, for regions containing geometric discontinuities (such as welds, openings, etc.), edge enhancement processing is implemented. The Canny operator is used to accurately identify feature boundaries, eliminating the jagged edge effect caused by mesh discretization.

[0034] Within high-risk areas, a quadtree-based spatial subdivision algorithm is used to implement multi-level mesh refinement, with the specific refinement level positively correlated with the area's risk level. For Level 1 high-risk areas (damage factor > 90%), a three-level mesh subdivision is implemented; for Level 2 high-risk areas (80%-90%), a two-level subdivision is implemented. During each subdivision, the system automatically calculates the spatial coordinates and normal vector information of the newly generated mesh nodes. In medium- and low-risk areas, a uniform mesh of fixed size is used, with the mesh size dynamically calculated and determined based on the overall equipment size and detection accuracy requirements. Throughout the generation of all mesh nodes, the system detects and eliminates any potential mesh distortions in real time to ensure that the mesh quality meets the requirements of subsequent calculations. Specifically, in areas with complex geometric features (such as areas with drastic changes in surface curvature), an adaptive curvature algorithm is automatically introduced to adjust the mesh density, ensuring geometric approximation accuracy.

[0035] The encrypted mesh nodes are optimized and sorted according to their spatial location to generate an efficient detection path sequence. Specifically, an improved Hilbert space-filling curve algorithm is used to sort the detection points in 3D space, ensuring that adjacent detection points remain spatially continuous. Each detection point records complete attribute information, including: absolute coordinates (X, Y, Z), surface normal vector (NX, NY, NZ), risk level, local mesh size, topological relationship of adjacent points, and estimated detection time. The sequence is initially optimized using the time-optimal principle, considering the kinematic parameters of the robotic arm to calculate the theoretical movement time between points. The final generated initial measurement point sequence not only contains basic detection position information but also pre-sets various parameter interfaces required for subsequent dynamic adjustments.

[0036] S130. Optimize the robot arm path for the initial measurement point sequence to generate a dynamically adjusted measurement point coordinate sequence.

[0037] The system first establishes a path planning model with multiple constraints. This model integrates the following key parameters: spatial coordinates and normal vector information of the detection points, DH kinematic parameters of the robotic arm, velocity / acceleration limits of each joint, accessibility constraints of the tool center point (TCP), and obstacle data of the surrounding environment. In specific implementation, a sampling-based fast random tree (RRT*) algorithm is used to perform a global path search in the configuration space to ensure that the path satisfies the robotic arm's kinematic constraints. For adjacent detection points where the normal vector changes by more than 45 degrees, transition path points are automatically inserted, and a smooth attitude transition trajectory is calculated through fifth-order polynomial interpolation. In particular, the system detects potential singular configurations in real time and avoids kinematic singularities in advance through Jacobian matrix analysis.

[0038] The system monitors the actual data acquisition quality during the detection process in real time. When the confidence level of the measurement data in a certain area falls below a preset threshold, it automatically triggers local path adjustments. Specifically, a data quality assessment model based on a sliding window is established to calculate the data dispersion and noise level of the current detection area in real time. For areas with substandard data quality, redundant detection points are automatically added and the local path is replanned. Simultaneously, the system continuously receives feedback data from the stress calculation module (S300). When unexpected stress concentration phenomena are identified, supplementary detection points are immediately inserted in the corresponding areas. All dynamic adjustment operations are logged in detail, including the reason for the adjustment, location coordinates, timestamps, and other information.

[0039] The optimized path sequence is converted into a set of motion instructions executable by the robotic arm, and a corresponding detection parameter configuration file is generated. Specifically, a piecewise cubic Hermitian interpolation algorithm is used to smooth the path points, ensuring the continuity of the robotic arm's speed and acceleration. Each motion instruction includes complete parameters such as target pose, motion speed, and contact force threshold. The final output measurement point coordinate sequence is stored using a hierarchical data structure: the base layer contains the core set of mandatory detection points; the optimization layer records all dynamically adjusted and newly added detection points; and the metadata layer stores various parameters and log information during the path optimization process. The sequence is transmitted to the robotic arm control system through a standard interface, while retaining complete version management information to support subsequent traceability and analysis of the detection process.

[0040] Step S200 includes at least steps S210-S230: S210. Obtain the coordinate sequence of the measuring points, control the hardness probe array to perform point-by-point indentation detection, and obtain the original hardness data.

[0041] The system receives a dynamically adjusted sequence of measurement point coordinates from the measurement point planning module. This sequence includes the three-dimensional spatial coordinates, surface normal vector information, and risk level markers for each detection point. Specifically, the system first performs format verification and integrity checks on the measurement point coordinate sequence to ensure all necessary parameters are complete and valid. Then, a measurement point priority queue is established, and the detection points are sorted according to the risk level markers, with higher-risk measurement points being prioritized for detection. For measurement points in areas with special geometric features, the system automatically loads preset detection parameter templates, including but not limited to key parameters such as indentation force, holding time, and data acquisition frequency.

[0042] The system drives a multi-degree-of-freedom robotic arm carrying a hardness probe array to move to the target measurement point along an optimized path. In practice, the probe array employs a modular design, comprising a main probe and an auxiliary probe, capable of simultaneously performing hardness tests of varying precision. The main probe performs standard indentation testing, applying a preset test force and measuring the diagonal length of the indentation after a specified holding time. The auxiliary probe performs rapid micro-destructive testing for data verification and anomaly detection. During the testing process, the system monitors the indentation morphology in real time, capturing indentation images through a built-in optical system and calculating the indentation size using digital image correlation algorithms. Specifically, for high-risk measurement points, the system automatically increases the sampling frequency and implements redundant testing to ensure data reliability.

[0043] The system collects complete test data from each measuring point, including raw information such as indentation size, applied force, loading curve, and surface morphology images. Specifically, a raw data quality assessment model is established to score the confidence level of the test data for each measuring point and remove obviously abnormal data points. All raw data is stored according to the measuring point coordinate index, forming a structured raw hardness dataset. This dataset not only contains hardness measurements but also records complete testing environment parameters and equipment status information, providing comprehensive basic data for subsequent data calibration. The system automatically generates a testing process log, recording detailed metadata such as the actual testing time, operator information, and equipment parameter settings for each measuring point.

[0044] S220. Extract the surface morphology interference component from the original hardness data, perform laser confocal compensation calibration, and obtain the calibrated hardness value.

[0045] The system first preprocesses the original hardness dataset to establish a surface morphology interference analysis model. In practice, it analyzes the indentation image features at each measuring point to extract surface roughness-related parameters, including key indicators such as the arithmetic mean deviation of the profile, the maximum height of the profile, and the spacing of micro-irregularities in the profile. Specifically, for areas with obvious machining textures or surface defects, the system employs a directional filtering algorithm to separate anisotropic interference components. Simultaneously, based on the probe contact mechanics model, it calculates the influence coefficient of surface morphology on indentation size measurement, quantifying the intensity level of morphology interference. All interference component extraction processes consider the local geometric features of the measuring points, including the influence of parameters such as surface curvature and the rate of change of the normal vector.

[0046] The system activates the laser confocal scanning module integrated in the probe array to perform high-precision three-dimensional topography measurement of the detection area. Specifically, the laser confocal system performs a spiral scan along the periphery of the indentation to acquire surface height distribution data with micron-level resolution. Based on the scanning results, a digital elevation model of the local surface topography is established, accurately reconstructing the microscopic geometric features of the area surrounding the indentation. The system employs a finite element-assisted analysis method to simulate the influence of different topographic features on hardness measurement and establish a topography-hardness correction matrix. In particular, for special material surfaces (such as areas that have undergone shot peening), the system calls upon a preset material property database and loads the corresponding correction coefficient algorithm. All calibration calculations consider the influence of environmental factors such as temperature drift and mechanical vibration, and implement real-time compensation.

[0047] The system fuses morphological interference components with laser confocal measurement results to calculate the comprehensive calibration coefficient for each measurement point. Specifically, a weighted average algorithm is used to process multi-source calibration data, with weight allocation based on the confidence scores of each data source. The calibration calculation process is iteratively optimized until the standard deviation of the hardness value reaches a preset accuracy threshold. The final output calibrated hardness value includes two parts: a base hardness value and calibration metadata. The metadata records complete calibration process parameters and confidence assessment results. The system automatically establishes a calibration traceability chain to ensure that each hardness value can be traced back to the original measurement data and calibration parameters. In particular, for measurement points that still exhibit significant dispersion after calibration, the system automatically marks them as suspicious points, triggering a review and inspection process.

[0048] S230: Integrate the calibrated hardness values ​​according to spatial coordinates and output the calibrated hardness matrix.

[0049] The system first establishes a three-dimensional spatial index structure, mapping discrete calibration hardness values ​​to grid nodes on the equipment surface. In practice, an improved octree spatial partitioning algorithm is used to organize the hardness data, optimizing query efficiency while maintaining spatial topological relationships. For cases where hardness data is missing from grid nodes, the system performs spatial interpolation calculations based on radial basis functions to ensure matrix integrity. Specifically, in high-risk and high-stress gradient regions, the system automatically improves interpolation accuracy, employing a kriging spatial prediction algorithm to calculate the hardness values ​​at missing points. All interpolation processes consider the anisotropic characteristics of the material, applying corresponding directional correction coefficients.

[0050] The system calculates the spatial correlation coefficient of hardness values ​​between adjacent measuring points in real time to detect potential outliers. Specifically, a spatial consistency assessment model based on a sliding window is established to identify outliers that do not conform to local variation patterns. For detected outliers, the system automatically triggers a local verification process, implementing verification testing through an auxiliary probe. Simultaneously, the system continuously monitors the spatial gradient changes of hardness values, automatically increasing data sampling density in areas of abrupt gradient changes to ensure the matrix's detailed representation capabilities. All quality control operations are logged in detail, including outlier handling records and verification test results, among other key information.

[0051] The system converts the integrated hardness data into a standard matrix format, containing complete spatial coordinate information and data quality markers. Specifically, the matrix data structure adopts a layered design: the base layer stores the hardness values ​​and coordinate information of the grid nodes; the intermediate layer records spatial interpolation parameters and quality control markers; and the metadata layer contains various parameters and operation logs from the matrix generation process. The matrix supports multiple query interfaces, allowing for efficient retrieval by region, risk level, or data quality. The system automatically generates a matrix integrity report, detailing key indicators such as data coverage and spatial resolution. The final output calibration hardness matrix maintains strict spatial alignment with the equipment CAD model, providing accurate input data for subsequent stress field calculations. In particular, a final consistency check is performed before matrix output to ensure that the error with the original test data is controlled within acceptable limits.

[0052] Step S300 includes at least steps S310-S330: S310. Obtain the calibration hardness matrix and material parameter library, input them into the GNN-FEA hybrid model, perform stress field prediction, and obtain the preliminary stress distribution.

[0053] The system receives a calibration hardness matrix from the hardness detection module. This matrix contains the hardness values ​​and spatial coordinates of each grid node on the device surface. Specifically, the system first performs a data integrity check on the calibration hardness matrix, verifying whether the hardness values ​​of each node are within the reasonable range for the material and confirming the correctness of the spatial topological relationships. Simultaneously, the system extracts a complete set of material parameters from the material database, including key mechanical properties such as elastic modulus, Poisson's ratio, and yield strength, as well as process parameters such as heat treatment state and work hardening coefficient. Particularly for composite materials or welded areas, the system automatically loads special material data such as interface bonding strength and anisotropy parameters.

[0054] The system fuses the calibration hardness matrix with the material parameter set to construct the input feature vector of the hybrid model. Specifically, a graph neural network (GNN) module is first used to process the spatial correlation of the hardness data and extract multi-scale features of the hardness distribution. Then, the feature vector is input into a finite element analysis (FEA) module to calculate the stress field distribution based on the principle of virtual work and constitutive relations. The hybrid model employs an iterative solution strategy. In each iteration, the GNN module adjusts the feature extraction weights based on the stress gradient fed back by the FEA, while the FEA module optimizes the stress calculation results based on the updated features. In particular, for regions with residual stress, the model automatically activates a plastic deformation correction algorithm to improve the accuracy of stress prediction.

[0055] The system performs post-processing on the raw stress field output from the hybrid model, including data smoothing, outlier removal, and boundary condition correction. Specifically, a physical constraint-based data filtering algorithm is used to eliminate high-frequency noise introduced by numerical calculations while preserving realistic stress concentration characteristics. All stress components (including normal and shear stresses) are stored in a standard format and strictly aligned with the coordinate system of the original CAD model. The system automatically generates a preliminary stress distribution quality assessment report, recording the calculation convergence status and error estimates for each region. This preliminary stress distribution not only includes stress amplitude information but also stores complete metadata of the calculation process, providing a foundation for subsequent confidence assessments.

[0056] S320. Extract confidence scores from the preliminary stress distribution, mark low-confidence regions, and obtain the coordinates of the region to be verified.

[0057] The system establishes a multi-index fusion confidence assessment model, comprehensively considering evaluation parameters across three dimensions: computational convergence, data consistency, and physical rationality. In practice, it first analyzes the stress gradient change rate of each grid node to calculate the local numerical oscillation index; then, it assesses the stress continuity between adjacent nodes to obtain a spatial consistency score; finally, it checks whether the stress values ​​conform to the material yield criterion to obtain a physical rationality judgment. Specifically, for boundary conditions and load application areas, the system additionally considers the degree of constraint satisfaction and adjusts the confidence weights for the corresponding regions. After normalization, all evaluation parameters are weighted and summed to obtain a comprehensive confidence score.

[0058] The system automatically identifies areas requiring focused verification based on a preset confidence threshold. Specifically, it employs a region growing algorithm, starting from a seed point, to merge adjacent low-confidence nodes into continuous regions, while simultaneously recording the geometric center and spatial extent of each region. For identified low-confidence regions, the system further analyzes the main reasons for the decrease in confidence, including but not limited to: missing hardness data, uncertain material parameters, and complex geometric features. Based on different cause types, the system automatically matches corresponding verification strategies and records the suggested verification methods in the region labeling. In particular, for large, continuous low-confidence regions, the system implements hierarchical labeling to distinguish between core issues and peripheral influence areas.

[0059] The system converts the labeled results into a set of spatial coordinates and attaches a detailed verification requirement description. Specifically, each region to be verified records its bounding rectangle, the coordinates of core problem nodes, and the recommended verification density level. The system automatically optimizes the spatial distribution of verification points, appropriately increasing the density of verification points in high-gradient areas and maintaining economical verification in flat areas. All regions to be verified are prioritized, taking into account factors such as regional risk level, confidence score, and geometric feature complexity. The final output set of coordinates of the regions to be verified maintains a strict spatial correspondence with the equipment CAD model, ensuring accurate positioning of the verification operation. The system simultaneously generates a verification task specification, clarifying the specific verification requirements and expected goals for each region.

[0060] S330. Perform local finite element refinement calculations on the coordinates of the area to be verified to generate a corrected stress field.

[0061] The system first establishes a multi-scale finite element model and then implements adaptive mesh refinement in the region to be verified. Specifically, an h-refinement method based on error estimation is used, dynamically adjusting the mesh size according to the stress gradient change rate, and automatically doubling the mesh density in stress concentration regions. For particularly complex stress states, the system activates a p-refinement strategy to increase the order of the element shape functions and enhance local computational accuracy. In particular, during the refinement process, the system strictly maintains the displacement compatibility between the old and new meshes to ensure the continuity of stress results. All refinement operations are logged in detail, including key information such as refinement location, refinement level, and computational resource consumption.

[0062] The system employs a multi-grid method to accelerate convergence, performing high-precision calculations in refined local regions while maintaining the original mesh accuracy in other regions. Specifically, a local-global iterative calculation framework is established. In each iteration, the high-precision local results influence the global calculation through interface conditions, while the global results provide boundary constraints for the local calculations. For regions with material nonlinearity, the system automatically activates an incremental iterative algorithm, gradually loading boundary conditions to improve computational stability. All local calculations consider the influence of the initial stress state, especially for regions that have undergone plastic deformation, accurately accounting for complex material behaviors such as the Bauschinger effect. In particular, the system monitors the computational convergence in real time, dynamically adjusting the iteration step size and convergence tolerance to balance computational accuracy and efficiency.

[0063] The system seamlessly integrates localized density calculation results with the global stress field, eliminating interface inconsistencies. Specifically, a least-squares-based field variable splicing algorithm is employed to achieve a smooth transition of stress components in transition regions. For interfaces where significant jumps still exist, the system performs additional stress smoothing processing, reconstructing a continuous stress field through polynomial fitting. The final generated corrected stress field retains complete calculation process information, including metadata such as mesh density level, calculation method, and convergence history for each region. The system automatically compares the stress field differences before and after correction, identifies areas of significant change, and marks potential risk points. The corrected stress field maintains a strict spatial registration relationship with the equipment CAD model, providing high-precision input data for subsequent stress inversion. In particular, the system synchronously updates the confidence score, reflecting the improvement in the reliability of the corrected stress field.

[0064] In another embodiment, S310: Obtain the calibration hardness matrix and material parameter library, input them into the GNN-FEA hybrid model, perform stress field prediction, and obtain the preliminary stress distribution.

[0065] In step S310, the system performs a crucial operation for stress field prediction. Specifically, the system first receives the calibration hardness matrix directly from the hardness detection module. (This matrix contains the hardness values ​​of each grid node on the device surface) and its spatial coordinates ,in For node indexing, (For grid point indexing). The process of obtaining the calibration hardness matrix includes: collecting multi-point hardness data through hardness sensors (such as Rockwell hardness testers) mounted on a robotic arm, and generating a spatial distribution matrix after processing surface morphology interference by the calibration module. The system also extracts a material parameter library from a material database, including elastic modulus. Poisson's ratio Yield strength Key parameters, and heat treatment state coefficient and work hardening coefficient In particular, for welded areas or composite materials, the system automatically applies the interfacial bonding strength. and anisotropic tensor These input data (hardness values ​​and material parameters) need to undergo data integrity verification before fusion: the system verifies whether the hardness values ​​are within a reasonable range (e.g., HV for steel materials), and check spatial topological relationships (mesh adjacency matrix). This is to eliminate abnormal topology errors. During data fusion, the system constructs feature vectors. As input to the hybrid model, where each feature vector contains a hardness value Spatial coordinates Combination of material parameters (e.g., ).

[0066] Furthermore, the system inputs the fused feature vector into the GNN-FEA hybrid model for stress field prediction. This model combines the advantages of Graph Neural Networks (GNN) and Finite Element Analysis (FEA), and employs an improved mathematical model to enhance accuracy. The GNN module handles the spatial correlation of hardness data and extracts multi-scale features: the system uses an improved graph convolution operator (based on functional analysis concepts), which innovatively incorporates the material yield criterion as a constraint, avoiding prediction bias caused by local fluctuations in hardness data. The FEA module calculates stress distribution based on physical principles, specifically activating plastic deformation correction for residual stress regions. The hybrid model employs an iterative solution strategy: the GNN adjusts its weights based on the stress gradient feedback from the FEA, while the FEA optimizes the stress field calculation. The output is the raw stress field, which requires post-processing (such as data smoothing and outlier removal) to generate a preliminary stress distribution. This distribution includes stress amplitude and computational metadata (convergence status, error estimation), which serve as input to S320.

[0067] The system employs an improved multiple regression equation (combined with functional analysis) to model the GNN-FEA hybrid process. This formula quantifies stress distribution based on equipment input data (hardness matrix and material parameters), as follows:

[0068] ① in: Preliminary stress field distribution function, unit: MPa, representing a spatial point. The predicted stress value.

[0069] : Regression coefficients, determined through optimization using training data, describe the feature weights.

[0070] Improved feature basis functions, based on GNN output , It is the feature vector extracted by the GNN module (dimension: (This is derived from graph convolution processing of the hardness matrix).

[0071] Physical constraint weighting factor, value range From the material parameter library Dynamic adjustment.

[0072] Physical loss function, definition Unit: MPa², ensure stress prediction conforms to material mechanics criteria ( The hardness-stress conversion factor is from a material library; (where is the gradient regularization coefficient).

[0073] Number of feature basis functions, default value 10, based on the number of input grid nodes. Adaptive adjustment.

[0074] Calibrate the hardness matrix, input data source, elements This represents the node hardness value.

[0075] : A set of material parameters, including .

[0076] The originality of this formula lies in extending multiple regression into a functional form under physical constraints, which solves the bias of traditional models in the hardness oversaturation region (such as the welded area).

[0077] enter How it is generated: The system first performs graph convolution operations. The adjacency matrix Calculation based on spatial coordinates. Formula ① utilizes equipment data (hardness value). and material parameters Dynamically optimize coefficients to ensure output Consistent with the actual stress field. During post-processing, the system applies a least-squares-based smoothing algorithm to remove high-frequency noise (preserving stress concentration features), ultimately outputting a value aligned with the CAD model. The matrix. The quality assessment report records the convergence error, providing a basis for the S320.

[0078] The output of S310 is It contains stress values ​​and metadata for each region, which can be directly input into the confidence assessment module of S320.

[0079] In step S310, the system first receives a calibration hardness matrix from the hardness detection module. This matrix contains the precise hardness values ​​and spatial coordinates of all grid nodes on the device surface. Simultaneously, the system extracts a complete set of material parameters from the material database, including key mechanical properties such as elastic modulus, Poisson's ratio, and yield strength, as well as process parameters such as heat treatment state coefficient and work hardening coefficient. For welded and composite material areas, the system specifically loads interface bonding strength and anisotropic characteristic parameters. During the data processing phase, the system performs integrity checks on the input data, verifying that the hardness values ​​are within a reasonable range and checking the accuracy of the spatial topological relationships. During data fusion, the system constructs feature vectors containing a combination of hardness values, spatial coordinates, and material parameters.

[0080] The system inputs the fused feature vectors into the GNN-FEA hybrid model to perform stress field prediction. This model employs an innovative iterative solution strategy: the graph neural network module handles the spatial correlation features of the hardness data, extracting multi-scale distribution patterns; the finite element analysis module calculates the stress distribution based on physical constitutive relations. Specifically for areas with residual stress, the model automatically activates a plastic deformation correction algorithm to improve prediction accuracy. After the prediction process, the system performs post-processing operations on the original stress field, including data smoothing, outlier removal, and boundary condition correction, ultimately generating preliminary stress distribution data that is precisely aligned with the equipment's CAD model. This distribution not only contains stress amplitude information but also stores complete computational metadata (such as convergence status and error estimates), serving as a direct input source for the S320 module.

[0081] Connection with S320: The preliminary stress distribution data output by S310 will be completely reused by S320. Specifically, the core input data required by the confidence assessment model in the S320 module—including stress values ​​at each node, stress gradient vectors, and metadata such as convergence residuals—are directly derived from the calculation results of S310. This direct data transfer ensures the continuity of the technical path, allowing any prediction errors that may exist in the S310 stage to be quantified and evaluated in the S320 stage.

[0082] S320: Extract confidence scores from the preliminary stress distribution, mark low-confidence regions, and obtain the coordinates of the region to be verified.

[0083] In step S320, the system analyzes the preliminary stress distribution. To ensure reliability and identify low-confidence regions, the following implementation is performed: The input data is generated by S310. (Includes stress amplitude, gradient, and convergence metadata). The system first establishes a multi-index confidence evaluation model, which comprehensively considers computational convergence, data consistency, and physical plausibility. Evaluation parameters include: computational convergence dimension, and the numerical oscillation index calculated by the system for each grid node. (Based on stress gradient change rate) (and iterative residuals); data consistency dimension, scored by stress continuity between adjacent nodes. (Calculate the ratio of the stress difference between adjacent nodes to the mean); Physical rationality dimension: check whether the stress value violates the material yield criterion (e.g., von Mises criterion: if...). (If the labeling is incorrect, then the labeling is unreasonable). Specifically, for boundary regions, the system evaluates the degree to which constraints are met (e.g., displacement boundary deviations). ), and dynamically adjust the weights. Input data source (stress distribution) After normalization, the system extracts features for confidence calculation.

[0084] Furthermore, confidence assessment is achieved through an improved Shannon entropy model: this model quantifies information uncertainty and innovatively combines stress gradient physical quantities to avoid misjudging true stress concentration areas. After calculating the local confidence score for each node, the system marks low-confidence regions according to a preset threshold (default 0.7, configurable). The marking process uses a region growing algorithm: starting from a low-confidence seed point, adjacent nodes are merged to form a continuous region, and the geometric center and spatial extent are recorded. The algorithm analyzes the reasons for the confidence decline (e.g., missing hardness data, uncertain material parameters, etc.) and matches a verification strategy (such as high-density hardness testing). The output is a set of coordinates of the region to be verified, including the coordinates of the circumscribed rectangle and a priority list.

[0085] Furthermore, an innovative formula is used for confidence assessment: the system employs a modified Shannon entropy formula (combined with a probability density function) to calculate the overall confidence score. This formula outputs a score based on the input stress data (gradient, range), as follows:

[0086] ② in: :node Overall confidence score, unit: no unit, range This indicates reliability (the higher the value, the higher the confidence level).

[0087] : Probability distribution elements, calculated based on input stress data, defined as follows ,in Cluster centers and standard deviations (using K-means cluster stress values) and gradient get).

[0088] Number of clusters, default value 5, determined based on stress field complexity.

[0089] Dynamic attenuation factor, formula ,in It calculates the convergence residual (from S310 metadata).

[0090] Numerical oscillation rate of change, unit: s⁻¹, representing instability in the time dimension ( It is an oscillation indicator, calculated as follows: ).

[0091] :node The stress value is input from the S310 output.

[0092] : Stress gradient vector, input from S310 calculation data.

[0093] The innovation of this formula lies in extending information entropy into a dynamic probability model, which solves the problem of misjudgment in high-gradient regions (such as welding zones) (for example, traditional entropy models tend to overlook true stress concentration). Input data and From Formula ② output Then, the system was normalized. Low-confidence regions are labeled using a region growing algorithm (mathematical details are not assigned separate numbers), and the result of formula ② is ( Directly driven labeling process: The system filters nodes based on a threshold and calculates region coordinates. Output ( (Assigning coordinates to the region center), with added priority. S320 and S330 interface: Includes the coordinates of the area to be verified; input S330 for local encryption calculation.

[0094] In step S320, the system performs a confidence assessment analysis based on the preliminary stress distribution data provided in S310. The system establishes a multi-dimensional comprehensive evaluation model, analyzing data reliability from three key dimensions: computational convergence (calculating numerical oscillation indices by analyzing the stress gradient change rate and iteration residuals); data consistency (assessing stress continuity between adjacent nodes and detecting potential outliers); and physical rationality (checking whether stress values ​​conform to material yield criteria and marking areas violating mechanical principles). For boundary regions, the system also additionally assesses the degree to which constraints are satisfied, serving as the basis for weighted adjustments.

[0095] After confidence assessment, the system generates a comprehensive confidence score for each grid node. Based on a preset confidence threshold (default 0.7), the system automatically identifies low-confidence regions requiring focused verification. The labeling process employs an intelligent region growing algorithm: starting from a low-confidence seed point, it gradually merges adjacent nodes to form continuous regions, while accurately recording the geometric center and spatial extent of each region. The algorithm deeply analyzes the root causes of confidence decline (such as missing hardness data, material parameter uncertainties, or complex geometric features) and matches the most suitable verification strategy accordingly. The final output is a structured set of coordinates for the regions to be verified, including the bounding rectangle of each region, coordinates of core problem nodes, recommended verification density level, and processing priority.

[0096] Connection with S330: The key outputs of the S320 phase—including the set of coordinates of the areas to be verified and the confidence score data for each area—will be directly input into the S330 module for subsequent processing. Specifically, the adaptive mesh refinement process in S330 relies entirely on these area coordinates to determine the spatial locations that require priority refinement; simultaneously, the confidence score data will determine the priority of the correction weights for each area, ensuring that resources are allocated preferentially to the least reliable areas.

[0097] S330: Perform local finite element refinement calculations on the coordinates of the area to be verified to generate a corrected stress field.

[0098] In step S330, the system is based on the coordinates of the region to be verified. Perform high-precision stress correction. The specific implementation is as follows: the input data is the output of the S320. (Including region coordinates and priorities). The system first establishes a multi-scale finite element model and implements adaptive mesh refinement in the target region: an improved h-refinement method (based on error estimation) is used to dynamically adjust the mesh size (increasing density in regions with large stress gradients). The refinement process prioritizes high-priority regions (such as core problem points), and the system records mesh change logs. Finite element calculations are accelerated using a multi-mesh method: a local-global iterative framework is constructed, performing high-precision calculations in the refined region while maintaining the original mesh in other regions. The system activates an incremental algorithm for material nonlinearities (e.g., plastic deformation) and considers the initial stress state (e.g., Bauschinger effect). The output is a local stress field, which needs to be fused with the global field: the system applies a least-squares-based stitching algorithm to eliminate interface discontinuities.

[0099] An innovative formula is used for mesh refinement and correction: the system employs an improved adaptive mesh formula (based on variational principles) to refine the target region. This formula takes the region coordinates as input. And the original stress field, output encryption parameters:

[0100] ③ in: : The size of the encrypted grid, in mm.

[0101] : Original mesh size, input from the system mesh library (based on CAD model by default).

[0102] Adaptive coefficient, formula ,in The coordinates of the region center (from) ), These are the coordinates of the boundary point.

[0103] : Stress gradient vector, input from the initial stress distribution of S310.

[0104] : Unit normal vector, calculated based on coordinates.

[0105] : The set of coordinates of the region to be verified, input from S320 output.

[0106] The innovation of this formula lies in introducing an exponential decay function to dynamically adjust the mesh size, solving the problems of over- or under-refinement in traditional methods (e.g., precise location of stress concentration points). Formula ③ output. Afterwards, the system underwent mesh reconstruction. Subsequently, the stress correction formally used the finite element equations, combined with the confidence data from formula ②, for modification. The modified formula adopted an improved functional analysis form:

[0107] ④ in: : Corrected stress field, unit: MPa.

[0108] Preliminary stress distribution, input from S310 output.

[0109] High-precision stress calculated locally and solved using FEA (the equations do not have independent numbering).

[0110] : Fusion weights, formula (in This is the area confidence score output by the S320. (Normalization factor).

[0111] : Area to be verified, input from S320.

[0112] : Global computation domain.

[0113] : Boundary of the region to be verified.

[0114] Formula ④ ensures priority correction in low-confidence regions, achieving seamless fusion using input data (region coordinates and confidence levels). Output Afterwards, the system updates the confidence score (reflecting improved reliability) and aligns it with the CAD model. The revised comparison report identifies risk points and outputs them to the subsequent stress inversion module. The entire S330 forms a closed loop: formulas ③ and ④ are input into the output data of S320 (…). Generates high-precision output.

[0115] Explanation of the connection between steps and the functional loop: Transition from S310 to S320: Preliminary stress distribution output by S310 (Generated by formula ①) is directly used as the input to S320. Specifically, the input data in formula ② ( and All of these are derived from the calculation results of S310 (from...) (Extracted from the matrix). For example, the core probability distribution of the confidence model. Depends on stress value This ensures a consistent technical path: the prediction error in S310 is quantified in S320.

[0116] Connection between S320 and S330: Coordinates of the area to be verified output by S320 and confidence score Enter S330 directly. The variable in formula ③ From The weights in formula ④ Depends on This achieves a closed-loop function: low-confidence regions are corrected through local encryption of S330 (Formula ④ uses region boundaries). The output corrected stress field improves overall reliability (e.g., in the welding area, the error is reduced by 30%).

[0117] Functional closed loop and empirical evidence: All formulas form an input-output chain: Formula ① (S310) output → Input and output Formula ② (S320) and → Input and output formulas ③ and ④ (S330) System feedback mechanism: The confidence score updated in S330 can be used to retrain the GNN-FEA model (in S310), forming a dynamic optimization closed loop.

[0118] In step S330, the system performs high-precision stress correction based on the coordinate data of the region to be verified provided in S320. The system first performs adaptive mesh refinement on the target region in a multi-scale finite element model: based on the stress gradient characteristics and priorities of the region, a dynamic size adjustment strategy is adopted—automatically increasing the mesh density in the core region where stress gradient changes drastically, while maintaining an economical mesh distribution in stable regions. During the refinement process, the system strictly maintains the displacement coordination relationship between the old and new meshes to ensure the continuity of stress results, while recording all mesh change information in detail.

[0119] The finite element method employs an innovative multi-mesh acceleration strategy: high-precision local calculations are performed in the refined region, while the original mesh accuracy is maintained in the unrefined region, achieving computational efficiency optimization through a local-global iterative framework. Specifically for regions with material nonlinearity (such as plastic deformation regions), the system activates an incremental iterative algorithm, progressively loading boundary conditions to improve computational stability. The influence of the initial stress state is fully considered during the calculation process, particularly for the accurate modeling of complex material behaviors such as the Bauschinger effect.

[0120] Finally, the system seamlessly integrates the localized calculation results with the global stress field. An advanced field variable stitching algorithm ensures smooth transitions of stress components in transition regions, eliminating interface discontinuities. For boundary regions where significant jumps still exist, the system performs additional stress smoothing optimization. After correction, the system automatically updates the confidence score and generates a correction comparison report, outputting high-precision stress field data that is strictly aligned with the equipment's CAD model.

[0121] Closed-loop feedback explanation: The entire S310-S330 process forms a complete technical closed loop: the initial stress distribution generated by S310 is used by S320 for confidence assessment; the low-confidence regions identified by S320 drive S330 to perform precise orientation correction; the updated confidence data from S330 can then be fed back into the model optimization process of S310. This closed-loop mechanism enables the system to continuously self-optimize, maintaining high-precision stress prediction capabilities under complex working conditions (such as achieving a 30% error reduction in the welding area), providing reliable technical support for the safety assessment of special equipment.

[0122] Step S400 includes at least steps S410-S430: S410. Obtain the modified stress field and the equipment CAD model, perform geometric topology matching, and obtain the stress-geometric correlation mapping.

[0123] The system receives corrected stress field data from the stress calculation module. This data includes stress components and their spatial distribution information at each node on the equipment surface. Specifically, the system first performs a data integrity check on the corrected stress field, verifying the validity of each stress component and the rationality of the boundary conditions. Simultaneously, the system loads the equipment's 3D CAD model file, parsing the model's geometric topology and material partitioning information. Particularly for large and complex equipment, the system employs a block-based loading strategy, processing model data region by region to ensure efficient memory usage.

[0124] The system establishes a precise mapping relationship between stress field nodes and the surface of the CAD model. In practice, a spatial indexing algorithm is first used to quickly locate the CAD model surface region corresponding to each stress node. Then, the nearest-point projection method is used to map the discrete stress nodes to the continuous CAD surface. For regions with abrupt changes in geometric features, the system automatically activates a local refinement matching algorithm to improve mapping accuracy. The matching process specifically considers the identification and processing of special geometric features such as welds and openings to ensure accurate correspondence between stress distribution and geometric features.

[0125] The system organizes the matching results into a structured association dataset, containing multi-level correspondences between stress nodes and CAD geometric elements. Specifically, a graph-based association network is established to record the connection relationships between each stress node and CAD faces, edges, and vertices. All association data includes matching quality assessment parameters, such as projection error and curvature matching degree. The system automatically generates a quality report of the association mapping, marking areas where matching deviations may exist, providing a reference for subsequent analysis. The stress-geometry association mapping supports efficient spatial querying and visualization, achieving deep integration of stress distribution and equipment geometry.

[0126] S420. Extract the peak stress region from the stress-geometric correlation mapping, perform residual stress inversion calculation, and obtain the high-risk stress coordinates.

[0127] The system establishes a multi-scale peak stress detection algorithm, comprehensively analyzing stress amplitude and gradient variation characteristics. In practice, it first identifies stress nodes exceeding the material's yield strength threshold globally, then aggregates them using a region growing algorithm to form continuous peak stress regions. Specifically, the system incorporates the geometric features of the CAD model to conduct specialized analysis of stress distribution at key locations such as weld fusion lines and structural corners, ensuring no local stress concentration points are missed. All detected peak regions are recorded, including their spatial extent, maximum stress value, and gradient change rate.

[0128] The system establishes a residual stress calculation model based on the principles of plasticity mechanics. In practice, it first separates the external load stress and residual stress components through constitutive relations, and then uses an iterative algorithm to solve for the self-equilibrium residual stress field. The calculation process specifically considers the effects of material work hardening and heat treatment history, applying corresponding material parameter correction coefficients. For welded areas, the system activates a dedicated welding residual stress calculation module, considering the effects of temperature field history and phase transformation strain to improve inversion accuracy. All calculations undergo rigorous convergence monitoring to ensure the physical reasonableness of the results.

[0129] The system converts the inversion calculation results into a structured coordinate dataset. Specifically, it records complete information for each high-risk stress point, including its three-dimensional coordinates, stress components, risk level, and calculated confidence level. The system automatically performs spatial clustering analysis on high-risk points, identifies key risk concentration areas, and calculates risk indicators for each area. All coordinate data maintains a strict spatial correspondence with the CAD model, supporting precise positioning and visualization. The system simultaneously generates a heatmap of the high-risk point distribution, intuitively displaying the spatial clustering characteristics of risk points and providing support for subsequent decision-making.

[0130] S430. Correct welding residual stress for high-risk stress coordinates and generate a stress concentration area report.

[0131] The system establishes a welding process characteristic database, containing information such as typical joint types, welding methods, and process parameters. In practice, it first identifies the welding areas where high-risk stress points are located, and then matches the corresponding welding process characteristic parameters. The system employs a neural network-based correction model to predict the distribution pattern of welding residual stress and compensates for and corrects the original stress calculation results. In particular, for multi-layer, multi-pass weld areas, the system considers the influence of weld sequence and interpass temperature, performing more refined stress correction calculations.

[0132] The system monitors stress changes before and after correction in real time and evaluates the rationality of the correction effect. Specifically, a correction limit check mechanism based on material mechanical properties is established to prevent over-correction. For points where the correction amount exceeds a preset threshold, the system automatically triggers a review process, verifying the reliability of the correction results through additional local calculations. All correction operations are logged with detailed information, including correction parameters, correction amounts, and review results, supporting subsequent traceability and analysis.

[0133] The system integrates all analysis results to generate structured report documents. Specifically, the report includes complete information such as the coordinate distribution of high-risk areas, stress levels, risk grades, and correction records. The system automatically classifies and sorts risk areas, prioritizing them according to stress level and hazard severity. All data is linked to the CAD model, supporting interactive 3D viewing and analysis. The report is output in a standardized format, including text descriptions, data tables, and visualizations to meet the viewing needs of different users. The system simultaneously generates a report summary, extracting key risk points and recommended measures to improve the efficiency of report use.

[0134] Step 500 includes at least steps S510-S530: S510: Obtain stress concentration area reports and confidence scores, identify areas with insufficient detection, and obtain the requirements for new measurement points.

[0135] The system receives a stress concentration area report from the stress inversion module. This report contains complete information such as high-risk stress coordinates, stress amplitudes, and their distribution characteristics. Specifically, the system first performs format validation and integrity checks on the report data, verifying the validity and logical consistency of each field. Simultaneously, the system loads confidence score data from the stress calculation module, which records the reliability of the stress calculation results for each area. In particular, the system establishes a mapping between the report data and the confidence score, ensuring that each high-risk point corresponds to a specific confidence index.

[0136] The system establishes a detection requirement analysis model based on multi-dimensional assessment. In practice, it first analyzes the spatial distribution characteristics of stress concentration areas to identify regions where existing measurement points are insufficiently covered; then, it assesses the confidence level of each region, marking suspicious areas with confidence levels below a preset threshold; finally, it combines equipment geometry and material properties to comprehensively determine key locations requiring supplementary detection. This analysis process specifically considers the detection needs of special areas such as weld heat-affected zones and structural discontinuities to ensure no critical risk points are overlooked.

[0137] The system converts the recognition results into a structured dataset of new measurement point requirements. Specifically, for each region requiring supplementary detection, it records complete information such as its center coordinates, the suggested number of measurement points, detection priority, and recommended detection parameters. The system automatically optimizes the spatial distribution of new measurement points, appropriately increasing the density in high-gradient regions and maintaining an economical arrangement in flat regions. All new measurement point requirements are accompanied by detailed documentation, including the recognition basis, risk assessment, and detection recommendations, providing ample support for subsequent path planning.

[0138] S520. Extract coordinate increments from the new measurement point requirements, dynamically update the robotic arm path, and obtain an optimized measurement point sequence.

[0139] The system first analyzes the data on the required new measurement points, extracting the coordinate information of all necessary additional detection points. In practice, a coordinate increment extraction algorithm is established to identify the spatial relationship between the new points and the existing measurement point sequence, calculating the optimal insertion position. Specifically, the system considers equipment geometric constraints and probe accessibility, automatically filtering out invalid coordinate points that the robotic arm cannot reach. All extracted coordinate increments are accompanied by priority markers and suggested detection parameters to ensure the targeted nature of subsequent path optimization.

[0140] The system establishes a real-time path optimization model based on kinematics and dynamics. In practice, the incremental coordinates of newly added measuring points are first fused with the existing measuring point sequence to construct a complete set of points to be detected. Then, considering the range of motion, speed, and acceleration limitations of each joint of the robotic arm, the optimal motion trajectory is calculated. Finally, collision detection and obstacle avoidance planning are implemented to ensure the safety and feasibility of the path. The update process pays particular attention to the transition between newly added measuring points and surrounding existing measuring points to ensure the smoothness and accuracy of the detection motion.

[0141] The system converts the updated path into an executable set of measurement point sequence instructions. Specifically, it records complete information for each measurement point, including its spatial coordinates, normal vector, detection order, and motion parameters. The system automatically generates a path optimization report, detailing key data such as the insertion position of newly added measurement points, path adjustments, and estimated detection time. All measurement point data is numbered and sorted according to the execution order, forming a structured optimized measurement point sequence to provide precise guidance for subsequent detection operations.

[0142] S530: Feed the optimized measurement point sequence back to the measurement point planning module to complete the closed-loop adjustment.

[0143] The system establishes a standardized data feedback interface to ensure the complete transmission of the optimized measurement point sequence. In practice, the optimized measurement point sequence is first converted and compressed to reduce data transmission volume; then, the data is transmitted to the measurement point planning module via a secure communication protocol; finally, data reception confirmation and integrity verification are implemented to ensure the accuracy of the feedback data. Specifically, the system records a complete feedback log, including feedback time, data volume, and reception status, supporting process traceability.

[0144] The system implements data fusion and version management in the measurement point planning module. Specifically, it compares and analyzes the optimized measurement point sequence with the original planning data to identify differences and optimization effects; then, it updates the measurement point planning database, recording the detailed process and results of this closed-loop adjustment; finally, it generates a closed-loop adjustment report, summarizing key indicators such as optimization content, performance improvements, and resource consumption. The system automatically triggers the version control mechanism of the planning module, saving historical version data and supporting backtracking and comparative analysis. All adjustment operations undergo a strict approval process to ensure the stability and reliability of the system operation.

[0145] Step S600 includes at least steps S610-S630: S610. Obtain the stress field and industry stress threshold standards, classify the risk levels, and obtain a risk heat map.

[0146] The system receives corrected stress field data from the stress calculation module. This data includes stress components and their spatial distribution information at each node on the equipment surface. Specifically, the system first performs a data integrity check on the corrected stress field, verifying the validity of each stress component and the rationality of the boundary conditions. Simultaneously, the system loads stress threshold standards matching the current equipment type from an industry standard database. These standards include key parameters such as allowable stress values ​​and fatigue limits for different materials and operating conditions. Particularly for special areas such as composite materials and welded joints, the system automatically matches relevant specific standard clauses to ensure the assessment is targeted.

[0147] The system establishes a risk assessment model based on multi-parameter fusion. In practice, the stress values ​​at each node are first compared with industry standard thresholds to calculate the stress exceedance coefficient; then, the stress gradient and its changing trend are analyzed to assess the degree of stress concentration; finally, historical equipment testing data and remaining life prediction results are combined to comprehensively determine the risk level of each area. The assessment process adopts a three-level classification system: Level 1 risk (immediate action), Level 2 risk (planned maintenance), and Level 3 risk (monitored operation), with clearly defined judgment criteria and color codes for each level.

[0148] The system spatially integrates the risk classification results with the equipment CAD model to generate a visualized risk heatmap. Specifically, gradient color rendering technology is used to represent areas with different risk levels: red indicates high risk, yellow indicates medium risk, and green indicates low risk. Each risk area in the heatmap is associated with detailed attribute data, including maximum stress value, exceedance percentage, risk level, and recommended measures. The system automatically generates heatmap legends and explanatory documents to ensure accurate communication of risk information. The heatmap supports multi-scale viewing and interactive querying; users can click on any area to obtain detailed stress analysis and risk assessment data.

[0149] S620. Extract the over-limit areas from the risk heat map, perform three-dimensional annotation and maintenance priority ranking, and obtain a maintenance recommendation table.

[0150] The system first performs image analysis on the risk heatmap to identify all stress regions exceeding the standard threshold. Specifically, a region growing algorithm is used to merge adjacent out-of-limit nodes into continuous out-of-limit regions, starting from a seed point. For each identified out-of-limit region, the system extracts its geometric feature parameters, including key indicators such as region area, centroid location, maximum stress value, and stress gradient. In particular, the system considers the special stress distribution patterns at structural discontinuities such as welds and openings, implementing targeted region boundary optimization to ensure the accuracy of the extraction results.

[0151] The system implements interactive 3D annotation on the equipment CAD model. Specifically, annotation elements are created for each out-of-limit area, including visual markers such as leader lines, text descriptions, and 3D arrows. The annotation content includes key information such as stress value, out-of-limit percentage, risk level, and discovery date. Maintenance priority ranking is based on a multi-factor evaluation model, considering factors such as stress out-of-limit degree, area criticality, accessibility, and maintenance cost, to calculate the comprehensive priority score for each out-of-limit area. The ranking process uses the analytic hierarchy process (AHP) to determine the weight of each factor, ensuring the scientific and reasonable nature of the ranking results.

[0152] The system converts the annotation and sorting results into a structured maintenance recommendation table. Specifically, the table includes complete fields such as area number, location description, stress data, risk level, priority, and recommended measures. The system automatically generates detailed maintenance plan suggestions for each out-of-limit area, including practical information such as maintenance methods, required materials, and estimated timeframe. All suggestions are based on maintenance specifications and a historical case library to ensure feasibility and effectiveness. The maintenance recommendation table is output in a standard format, supporting filtering, sorting, and export operations, facilitating the development and execution of maintenance plans.

[0153] S630: Integrate the maintenance recommendation form with the equipment CAD model to generate a visual decision report.

[0154] The system establishes a two-way association mechanism between maintenance data and CAD models. Specifically, spatial indexing technology links each maintenance suggestion item in the table to the corresponding area in the CAD model, achieving precise matching between data and geometry. The fusion process considers model simplification needs, implementing appropriate lightweight processing for complex geometry to ensure smooth interaction. In particular, the system generates a 3D marker symbol for each maintenance suggestion item, visually displaying its actual location and spatial relationship on the equipment.

[0155] The system integrates all analysis results to generate an interactive 3D decision report. Specifically, the main body of the report uses a CAD model display framework, overlaid with risk heatmaps and maintenance annotations; the sidebar displays a structured maintenance recommendation table, supporting click-to-search and linked highlighting; the bottom area contains comprehensive information such as a report summary, key risk points, and overall recommendations. The system automatically generates report version information and a digital signature to ensure the document's authority and traceability. The decision report supports multiple output formats, including web version, PDF version, and 3D interactive version, meeting the needs of different usage scenarios. In particular, the system provides a report customization function, allowing users to customize the display content and format style, enhancing the report's usability.

[0156] Example 2: Figure 2 A structural block diagram of a multi-point hardness testing system for special equipment according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include: The measuring point planning module 10 is used to acquire the equipment's CAD model and historical stress data, classify risk levels, refine the mesh in high-stress areas, and optimize the generation of measuring point coordinate sequences. Specifically, this module analyzes the equipment's 3D CAD model and combines it with historical stress detection data to establish a risk assessment model based on finite element meshes, achieving intelligent optimization of the measuring point distribution. The module outputs a measuring point sequence containing spatial coordinates, normal vectors, and detection parameters, providing precise guidance for subsequent inspections.

[0157] The hardness acquisition module 20 is used to perform hardness testing according to the coordinate sequence of measurement points, calibrate surface morphology interference, integrate spatial coordinates, and output a calibration hardness matrix. This module integrates a high-precision hardness probe array and a laser confocal measurement system, performs point-by-point indentation detection, and eliminates the influence of surface roughness through a morphology compensation algorithm. The calibration hardness matrix output by the module includes the spatial coordinates, hardness values, and data quality indicators for each measurement point.

[0158] The stress calculation module 30 is used to input the calibration hardness matrix and material parameters into the GNN-FEA hybrid model to predict the stress field, mark low-confidence regions, and correct the stress distribution through finite element analysis to generate the corrected stress field. This module uses a graph neural network to process the spatial correlation of hardness data, combines it with finite element analysis for stress field calculation, and improves the calculation accuracy of key areas through local mesh refinement. The corrected stress field output by the module includes complete stress components and confidence scores.

[0159] The stress inversion module 40 is used to match and correct the stress field with the CAD model, invert peak stress regions, correct welding residual stress, and generate a stress concentration region report. This module establishes a mapping between the stress field and the equipment model using a geometric topology matching algorithm and calculates the residual stress distribution using the principles of plasticity. The module's output report includes high-risk stress coordinates, stress amplitudes, and correction records.

[0160] The closed-loop optimization module 50 identifies new measurement point requirements based on stress concentration reports and confidence scores, dynamically updates the robotic arm path, and feeds it back to the measurement point planning module. This module implements real-time kinematic path optimization, considering robotic arm accessibility and detection efficiency, forming a closed-loop feedback mechanism of "detection-calculation-optimization." The optimized measurement point sequence output by the module ensures detection coverage and data quality in key areas.

[0161] The decision output module 60 combines stress field analysis with industry standards to classify risk levels, mark areas exceeding limits, and integrate with CAD models to generate a visual decision report. This module establishes a multi-parameter risk assessment model and uses 3D annotation technology to intuitively display risk information. The interactive report output by the module supports maintenance priority ranking and maintenance plan recommendations, providing a basis for equipment safety management decisions.

[0162] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A multi-point hardness testing method for special equipment, characterized in that, include: Acquire the equipment CAD model and historical stress data, classify risk levels and densify the mesh in high-stress areas, and optimize the generation of measurement point coordinate sequences; Hardness testing is performed according to the measurement point coordinate sequence, surface morphology interference is calibrated and spatial coordinates are integrated, and a calibration hardness matrix is ​​output. The calibration hardness matrix and material parameters are input into the hybrid model to predict the stress field. Low confidence regions are marked and the stress distribution is corrected by finite element method to generate the corrected stress field. Match and correct the stress field with the CAD model, invert the peak stress region and correct the welding residual stress, and generate a stress concentration region report; Based on the stress concentration report and confidence score, identify the need for new measurement points, dynamically update the robotic arm path and feed it back to the measurement point planning module; By combining stress field and industry standards to classify risk levels, marking out-of-limit areas and integrating CAD models, a visual decision report is generated.

2. The multi-point hardness testing method according to claim 1, characterized in that, The process of acquiring the equipment CAD model and historical stress data, classifying risk levels and densifying the mesh in high-stress areas, and optimizing the generation of measurement point coordinate sequences includes: Obtain the equipment CAD model and historical stress distribution data, classify the regional risk levels, and obtain a risk level map; High-stress potential areas are extracted from the risk level map, and adaptive mesh refinement is performed to obtain the initial measurement point sequence. The robot arm path is optimized based on the initial measurement point sequence to generate a dynamically adjusted measurement point coordinate sequence.

3. The multi-point hardness testing method according to claim 1, characterized in that, The process of performing hardness testing according to the measurement point coordinate sequence, calibrating surface morphology interference, integrating spatial coordinates, and outputting a calibration hardness matrix includes: The coordinate sequence of the measuring points is obtained, and the hardness probe array is controlled to perform point-by-point indentation detection to obtain the raw hardness data. The surface morphology interference component is extracted from the original hardness data, and laser confocal compensation calibration is performed to obtain the calibrated hardness value. The calibrated hardness values ​​are integrated according to spatial coordinates to output a calibrated hardness matrix.

4. The multi-point hardness testing method according to claim 1, characterized in that, The process of inputting the calibration hardness matrix and material parameters into a hybrid model to predict the stress field, marking low-confidence regions, and correcting the stress distribution using finite element analysis to generate a corrected stress field includes: Obtain the calibration hardness matrix and material parameter library, input them into the GNN-FEA hybrid model, perform stress field prediction, and obtain the preliminary stress distribution.

5. The multi-point hardness testing method according to claim 4, characterized in that, The process of obtaining the preliminary stress distribution also includes: Confidence scores are extracted from the preliminary stress distribution, low-confidence regions are marked, and the coordinates of the region to be verified are obtained. Local finite element analysis is performed on the coordinates of the area to be verified to generate a corrected stress field.

6. The multi-point hardness testing method according to claim 1, characterized in that, The matching and correction of the stress field with the CAD model, the inversion of the peak stress region and the correction of welding residual stress, and the generation of a stress concentration region report include: Obtain the modified stress field and the equipment CAD model, perform geometric topology matching, and obtain the stress-geometry correlation mapping; Peak stress regions are extracted from the stress-geometric correlation mapping, and residual stress inversion calculations are performed to obtain high-risk stress coordinates. Welding residual stress is corrected for high-risk stress coordinates, and a stress concentration area report is generated.

7. The multi-point hardness testing method according to claim 1, characterized in that, The process of identifying new measurement point requirements based on stress concentration reports and confidence scores, dynamically updating the robotic arm path, and feeding back the information to the measurement point planning module includes: Obtain stress concentration area reports and confidence scores, identify areas with insufficient detection, and determine the need for new measurement points; The coordinate increments are extracted from the new measurement point requirements, and the robotic arm path is dynamically updated to obtain an optimized measurement point sequence. The optimized measurement point sequence is fed back to the measurement point planning module to complete the closed-loop adjustment.

8. The multi-point hardness testing method according to claim 1, characterized in that, The process involves combining stress field analysis with industry standards to classify risk levels, marking out-of-limit areas, integrating CAD models, and generating a visual decision report, including: Obtain the stress field and industry stress threshold standards, classify the risk levels, and obtain a risk heat map; Exceeding limits areas are extracted from the risk heat map, and three-dimensional annotations and maintenance priority rankings are performed to obtain a maintenance recommendation table; The maintenance recommendation form is integrated with the equipment CAD model to generate a visual decision report.

9. The multi-point hardness testing method according to claim 1, characterized in that, The calibration hardness matrix and material parameters are input into the hybrid model to predict the stress field. The hybrid model is a GNN-FEA hybrid model, which is used to establish the correlation between hardness and residual stress.

10. A multi-point hardness testing system for special equipment, applied to the multi-point hardness testing method for special equipment according to any one of claims 1-9, characterized in that, include: The measuring point planning module is used to acquire the equipment CAD model and historical stress data, classify risk levels and densify the mesh of high-stress areas, and optimize the generation of measuring point coordinate sequences. The hardness acquisition module is used to perform hardness testing according to the coordinate sequence of the measurement points, calibrate surface morphology interference, integrate spatial coordinates, and output a calibration hardness matrix. The stress calculation module is used to input the calibration hardness matrix and material parameters into the hybrid model to predict the stress field, mark low confidence areas, and generate the corrected stress field by correcting the stress distribution through finite element method. The stress inversion module is used to match and correct the stress field with the CAD model, invert the peak stress region and correct the welding residual stress, and generate a stress concentration region report. The closed-loop optimization module is used to identify new measurement point requirements based on stress concentration reports and confidence scores, dynamically update the robotic arm path, and feed it back to the measurement point planning module. The decision output module is used to combine stress field and industry standards to classify risk levels, mark out-of-limit areas, integrate CAD models, and generate a visual decision report.

Images