AI-assisted rapid analysis and prediction method and system for the formability of stamped parts

Abstract

This invention provides an artificial intelligence-assisted method and system for rapid analysis and prediction of the formability of stamped parts, relating to the field of sheet metal stamping technology. The method includes extracting geometric deformation feature parameters from three-dimensional data of the stamped part, constructing strain path curves to determine deformation risk areas, establishing a correspondence between strain abrupt change characteristics and defect types to predict defect locations and types, and optimizing mesh data based on strain rate gradients and material hardening exponents for forming analysis. This invention can improve the accuracy and efficiency of stamped part forming defect prediction, and reduce trial molding and debugging costs and cycles.

Description

Technical Field

[0001] This invention relates to sheet metal stamping technology, and more particularly to an artificial intelligence-assisted method and system for rapid analysis and prediction of the formability of stamped parts. Background Technology

[0002] In the stamping process, the formability analysis of parts is crucial for ensuring product quality and improving production efficiency. Traditional stamping analysis methods mainly rely on finite element simulation, which requires significant computational time and resources, making it difficult to meet the needs of rapid analysis and prediction. Furthermore, due to limitations in mesh generation accuracy, it often fails to accurately capture the strain evolution characteristics of local deformation regions, thus affecting the accuracy of the analysis results.

[0003] With the development of artificial intelligence technology, it has become possible to apply machine learning methods to formability analysis. However, current intelligent analysis methods often only focus on predicting the final forming result, lacking in-depth analysis of the strain evolution characteristics during the forming process, making it difficult to effectively identify and warn of potential forming defects. In addition, existing methods often suffer from unstable prediction accuracy when dealing with complex shaped parts due to limitations in mesh quality.

[0004] Historical data and engineering experience have shown that the generation of forming defects is closely related to abrupt changes in local strain and strain rate gradients. However, current technologies have not yet established an effective defect early warning mechanism or an adaptive mesh optimization method. Therefore, there is an urgent need for an intelligent analysis method that can quickly and accurately predict forming defects and has adaptive mesh optimization capabilities. Summary of the Invention

[0005] This invention provides an artificial intelligence-assisted method and system for rapid analysis and prediction of the formability of stamped parts, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides an artificial intelligence-assisted method for rapid analysis and prediction of the formability of stamped parts, comprising:

[0007] The three-dimensional data of the stamped part is acquired and meshed to obtain the initial mesh data. Geometric deformation feature parameters are then extracted from the initial mesh data.

[0008] Strain path curves are constructed based on geometric deformation characteristic parameters. The strain increment between adjacent time points is calculated, and the strain acceleration is calculated based on the strain increment. The strain path curves are compared with the forming limit curves to obtain the safety margin values ​​of each grid node. Deformation risk areas are determined based on the strain acceleration and safety margin values.

[0009] Acquire historical stamping data containing defect types and characteristics, analyze strain abrupt change characteristics in the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze deformation risk areas based on defect warning criteria, and obtain predicted defect locations and defect types;

[0010] The strain rate gradient at the predicted defect location is calculated, and the strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation, and the initial mesh data is locally optimized based on the mesh refinement coefficient to obtain optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

[0011] In one alternative embodiment,

[0012] The three-dimensional data of the stamped part is acquired and meshed to obtain initial mesh data. Geometric deformation feature parameters are extracted from the initial mesh data, including:

[0013] Point cloud data of stamped parts is acquired using a 3D scanner and preprocessed. Initial mesh data is constructed based on the preprocessed point cloud data. The initial mesh data includes node location information and cell topology. Mesh quality parameters of the initial mesh data are calculated, and the mesh area is optimized according to the mesh quality parameters to obtain an optimized mesh model.

[0014] Based on the optimized mesh model, a deformation feature calculation matrix is ​​established, which includes nodal displacement, deformation gradient and strain components. The deformation tensor is calculated using the deformation feature calculation matrix, and the principal strain components and secondary strain components are solved using the deformation tensor.

[0015] The wall thickness variation of each node is calculated based on the principal strain component and the secondary strain component. The wall thickness variation is combined with the node displacement to calculate the depth variation value, and a geometric deformation characteristic parameter containing the principal strain component, the secondary strain component, the wall thickness variation value, and the depth variation value is constructed.

[0016] In one alternative embodiment,

[0017] Strain path curves are constructed based on geometric deformation characteristic parameters. The strain increments between adjacent time points are calculated, and the strain acceleration is calculated based on the strain increments, including:

[0018] Gaussian filtering and mean filtering are applied to the geometric deformation characteristic parameters to obtain filtered deformation characteristic data. Cubic spline interpolation is then used to spatially reconstruct the filtered deformation characteristic data to obtain a continuously distributed deformation characteristic field.

[0019] The principal strain field and the secondary strain field are established based on the continuously distributed deformation characteristic field. The strain path curve between the principal strain field and the secondary strain field is fitted by the nonlinear least squares method with adaptive weight coefficients. The strain path curve includes the principal strain component and the secondary strain component.

[0020] The strain path curve is decomposed into principal strain vector and secondary strain vector by orthogonal direction vector. The difference between principal strain vector and secondary strain vector is calculated based on the displacement increment and time increment of adjacent time points to obtain the strain increment vector. The strain increment vector is dynamically accumulated within the sampling time window to obtain the strain increment accumulation value. The strain acceleration is calculated based on the rate of change of the strain increment accumulation value. The strain acceleration characterizes the dynamic characteristics of local deformation.

[0021] In one alternative embodiment,

[0022] By comparing the strain path curve with the forming limit curve, the safety margin value of each grid node is obtained. Based on the strain acceleration and the safety margin value, the deformation risk area is determined, including:

[0023] The first set of sampling points is obtained by sampling at equal time intervals on the strain path curve, and the second set of sampling points is obtained by sampling at equal strain intervals on the forming limit curve. The first set of sampling points and the second set of sampling points are paired based on the minimum distance principle. The distance between the paired points is calculated to construct a distance mapping matrix. The change segment is identified according to the gradient change rate of the values ​​in the distance mapping matrix. The sampling point density is increased in the change segment. All sampling points are reconstructed by piecewise cubic spline interpolation algorithm.

[0024] Calculate the vertical distance from the sampling point on the strain path curve to the reconstructed forming limit curve, construct a nonlinear weighting function, use the nonlinear weighting function to calculate the safety margin component value of the sampling point, and sum the safety margin component values ​​of all sampling points by weight to obtain the safety margin value.

[0025] The safety margin value and strain acceleration of each grid node are combined to form a risk judgment feature pair. Potential risk nodes are identified based on the preset double threshold constraint. The spatial distance and strain gradient continuity between adjacent risk nodes are calculated. Risk nodes with a spatial distance smaller than the grid feature size and a continuously changing strain gradient are merged into an initial risk region. The boundary of the initial risk region is corrected according to the strain gradient distribution of the nodes to determine the final deformation risk region.

[0026] In one alternative embodiment,

[0027] Historical stamping data containing defect types and characteristics is acquired. The strain abrupt change characteristics of the defect formation process are analyzed, and a correspondence between strain abrupt change characteristics and defect types is established. Defect warning criteria are obtained, and deformation risk areas are analyzed based on these criteria to predict defect locations and types, including:

[0028] Acquire historical stamping data containing defect types and defect characteristics, calculate strain gradient tensor, strain path curvature and energy dissipation rate on historical stamping data to obtain multi-dimensional strain abruptness characteristics;

[0029] Recursive Bayesian estimation is used to update the weights of the multidimensional strain abrupt features to generate fused feature data. Based on the defect types in historical stamping data and the fused feature data, the temporal variation characteristics and spatial distribution characteristics of the defect formation process are analyzed, and a mapping relationship between strain abrupt features and defect types is established.

[0030] The mapping relationship is used to construct a feature-defect association network. The evolution path of the fused feature data in the association network is analyzed, and an early warning criterion including feature thresholds and defect evolution rules is established.

[0031] Real-time acquisition of strain field and temperature field data; extraction of principal direction and principal value change rate of strain gradient tensor in deformation region; establishment of stress transmission path diagram in deformation region; analysis of abrupt change points of strain path curvature and fluctuation degree of energy dissipation rate during stress transmission process; calculation of deformation instability index in combination with defect evolution law in early warning criterion; determination of defect location based on propagation characteristics of instability index; determination of defect type based on strain state combination during instability process.

[0032] In one alternative embodiment,

[0033] Establish a stress transmission path diagram in the deformation region, analyze the abrupt changes in strain path curvature and the degree of energy dissipation fluctuation during stress transmission, calculate the deformation instability index based on the defect evolution law in the early warning criteria, determine the defect location based on the propagation characteristics of the instability index, and determine the defect type according to the strain state combination during the instability process, including:

[0034] The deformed region is discretized into triangular mesh elements. The difference in the principal stress direction between adjacent mesh elements is calculated as the transmission impedance. The minimum spanning tree algorithm is used to connect the minimum transmission impedance elements to obtain the stress transmission path diagram of the deformed region.

[0035] Strain field data are collected along the stress transfer path diagram to construct a strain path, the curvature change of the strain path is calculated, and the curvature abrupt change points and strain energies are extracted and combined into abrupt change feature vector.

[0036] The energy transfer path is determined based on the stress transfer path diagram. The dissipation rate of strain energy, its fluctuation degree, and the dominant frequency component are calculated along the energy transfer path and combined into a fluctuation characteristic vector.

[0037] The mutation feature vector and the fluctuation feature vector are mapped onto the stress transmission path diagram to obtain the energy dissipation fluctuation feature score and the stress transmission path feature score, respectively. Combined with the defect evolution law in the early warning criterion, the deformation instability index is calculated by dynamic weight combination.

[0038] A spatial distribution map is established based on the deformation instability index, the gradient field of the deformation instability index is calculated, and the direction and speed of instability propagation are determined based on the propagation characteristics of the instability index to predict the location of defects.

[0039] The principal strain ratio is calculated using strain field data, the strain path non-proportionality is calculated based on the degree of non-linearity of the strain path, and the local strain concentration factor is calculated based on the strain distribution in the spatial distribution map. The principal strain ratio, strain path non-proportionality, and local strain concentration factor are combined into strain state characteristics, and the defect type is determined based on the value of the strain state characteristics.

[0040] In one alternative embodiment,

[0041] The mesh refinement factor is calculated using the mesh optimization equation. Based on this factor, the initial mesh data is locally optimized to obtain the optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results include:

[0042] The plastic power density within the cell is calculated based on the mesh optimization equation, and a stress-strain field distribution function containing deformation energy density and strain localization terms is established based on the plastic power density.

[0043] The stress gradient and strain gradient of the mesh element are calculated using the stress-strain field distribution function. A critical refinement threshold is established based on the plastic power density. The mesh refinement coefficient is obtained by comparing the stress gradient and strain gradient with the critical refinement threshold.

[0044] The mesh cells are locally refined according to the mesh refinement coefficient. A strain path function is introduced into the refined mesh cells. The pre-acquired material strain hardening index is updated into the refined mesh data. Finite element iterative calculation is performed using the optimized mesh data. The strain rate gradient and mesh refinement coefficient are updated based on the calculation results. The mesh density distribution is dynamically adjusted, and the stamping forming analysis results are output.

[0045] A second aspect of the present invention provides an artificial intelligence-assisted rapid analysis and prediction system for the formability of stamped parts, comprising:

[0046] The first unit is used to acquire the three-dimensional data of the stamped part and perform mesh generation to obtain the initial mesh data, and extract the geometric deformation feature parameters from the initial mesh data;

[0047] The second unit is used to construct strain path curves based on geometric deformation characteristic parameters, calculate the strain increment between adjacent time points of the strain path curve, calculate the strain acceleration based on the strain increment, compare the strain path curve with the forming limit curve to obtain the safety margin value of each grid node, and determine the deformation risk area based on the strain acceleration and the safety margin value.

[0048] The third unit is used to acquire historical stamping data containing defect types and defect characteristics, analyze the strain abrupt change characteristics of the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze the deformation risk area based on the defect warning criteria, and obtain the predicted defect location and defect type.

[0049] The fourth unit is used to calculate the strain rate gradient at the predicted defect location. The strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation. The initial mesh data is then locally optimized based on the mesh refinement coefficient to obtain the optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

[0050] A third aspect of the present invention provides an electronic device, comprising:

[0051] processor;

[0052] Memory used to store processor-executable instructions;

[0053] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0055] In this embodiment, strain path curves are constructed by extracting geometric deformation characteristic parameters. Combined with strain acceleration and safety margin values, deformation risk areas are determined, enabling early identification of potential problems during the stamping process. This improves prediction accuracy and reduces material waste and mold damage risks in production. Based on historical stamping data analysis, a correspondence between strain abrupt changes and defect types is established, forming defect early warning criteria. This allows for accurate prediction of defect locations and types, enabling engineers to optimize stamping process parameters, shortening product development cycles, and improving product quality stability. A mesh optimization method combining strain rate gradient and material strain hardening index is employed, achieving local mesh refinement. This improves analysis efficiency while maintaining computational accuracy, meeting the needs of rapid analysis in industrial production and demonstrating significant engineering application value. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the rapid analysis and prediction method for the formability of stamped parts assisted by artificial intelligence, according to an embodiment of the present invention.

[0057] Figure 2 This is a comparison chart of the safety margin weighting functions in embodiments of the present invention;

[0058] Figure 3 This is a thermal diagram of the spatial distribution and gradient field of the deformation instability index in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of the structure of the AI-assisted rapid analysis and prediction system for the formability of stamped parts according to an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0062] Figure 1 This is a flowchart illustrating the rapid analysis and prediction method for the formability of stamped parts assisted by artificial intelligence, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0063] The three-dimensional data of the stamped part is acquired and meshed to obtain the initial mesh data. Geometric deformation feature parameters are then extracted from the initial mesh data.

[0064] Strain path curves are constructed based on geometric deformation characteristic parameters. The strain increment between adjacent time points is calculated, and the strain acceleration is calculated based on the strain increment. The strain path curves are compared with the forming limit curves to obtain the safety margin values ​​of each grid node. Deformation risk areas are determined based on the strain acceleration and safety margin values.

[0065] Acquire historical stamping data containing defect types and characteristics, analyze strain abrupt change characteristics in the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze deformation risk areas based on defect warning criteria, and obtain predicted defect locations and defect types;

[0066] The strain rate gradient at the predicted defect location is calculated, and the strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation, and the initial mesh data is locally optimized based on the mesh refinement coefficient to obtain optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

[0067] In one optional implementation, the three-dimensional data of the stamped part is acquired and meshed to obtain initial mesh data. Geometric deformation feature parameters are extracted from the initial mesh data, including:

[0068] Point cloud data of stamped parts is acquired using a 3D scanner and preprocessed. Initial mesh data is constructed based on the preprocessed point cloud data. The initial mesh data includes node location information and cell topology. Mesh quality parameters of the initial mesh data are calculated, and the mesh area is optimized according to the mesh quality parameters to obtain an optimized mesh model.

[0069] Based on the optimized mesh model, a deformation feature calculation matrix is ​​established, which includes nodal displacement, deformation gradient and strain components. The deformation tensor is calculated using the deformation feature calculation matrix, and the principal strain components and secondary strain components are solved using the deformation tensor.

[0070] The wall thickness variation of each node is calculated based on the principal strain component and the secondary strain component. The wall thickness variation is combined with the node displacement to calculate the depth variation value, and a geometric deformation characteristic parameter containing the principal strain component, the secondary strain component, the wall thickness variation value, and the depth variation value is constructed.

[0071] In this embodiment, a 3D scanner is used to scan the stamped part to obtain point cloud data. Taking a certain automotive body panel as an example, a structured light 3D scanner is used to scan and obtain raw point cloud data containing approximately 500,000 spatial points. Since the raw point cloud data usually contains noise and redundant points, preprocessing is required. The preprocessing steps include: noise filtering, point cloud downsampling, and point cloud registration.

[0072] In the noise filtering stage, a statistical outlier filtering method is used to calculate the average distance from each point to its nearest neighbors. If this distance exceeds twice the global average distance, the point is identified as noise and removed. For example, with a distance standard deviation of 1.0 and a nearest neighbor count of 50, approximately 0.8% of noise points are filtered out. In the point cloud downsampling stage, a voxel grid downsampling method is used to divide the point cloud space into small cubic grids, with each point in the grid represented by its geometric center. With a voxel size of 0.5 mm, the original point cloud data is downsampled to approximately 150,000 points, preserving the geometric features of the object while improving processing efficiency. Point cloud registration is achieved through an iterative nearest-neighbor algorithm, registering point cloud data from multiple scans to the same coordinate system. During registration, the maximum number of iterations is set to 100, the convergence threshold is 0.00001, and the average registration error is controlled within 0.1 mm.

[0073] Based on the preprocessed point cloud data, initial mesh data was constructed using the Poisson surface reconstruction method. During the Poisson reconstruction process, the reconstruction depth was set to 9, and the number of sampling points was 15, resulting in an initial mesh model containing approximately 80,000 triangular elements and 40,000 nodes. Each node contains three-dimensional coordinate information (x, y, z), and each element contains index information for three nodes, forming the topological relationship of the mesh.

[0074] The initial mesh model is quality-assessed, and mesh quality parameters are calculated. These parameters include: cell area ratio, cell interior angles, cell aspect ratio, and the included angle between adjacent cells. The cell area ratio is defined as the ratio of the area of ​​a triangular cell to the average area of ​​the mesh, ideally ranging from 0.5 to 1.5. The cell interior angle is defined as the minimum value of the three interior angles of a triangle, ideally greater than 30 degrees. The cell aspect ratio is defined as the ratio of the height of a triangle to the length of its base, ideally ranging from 0.5 to 2.0. The included angle between adjacent cells is defined as the angle between the normal vectors of adjacent triangles, ideally less than 30 degrees.

[0075] For regions where the mesh quality parameters do not meet the requirements, optimization is performed. Optimization methods include mesh smoothing, mesh subdivision, and mesh reconstruction. Mesh smoothing uses the Laplacian smoothing operator to adjust node positions, with a smoothing iteration count of 5. Mesh subdivision targets high-curvature regions, such as the corners of stamped parts, subdividing a single triangle into four smaller triangles. Mesh reconstruction targets severely distorted elements, optimizing mesh quality by adjusting node connections. After optimization, the proportion of elements not meeting quality requirements decreased from the initial 15% to below 3%.

[0076] Based on the optimized mesh model, a deformation characteristic calculation matrix is ​​established. First, the nodal displacements between the original sheet metal plane and the stamped surface are calculated. For each mesh node, the three-dimensional coordinate differences (Δx, Δy, Δz) before and after stamping are recorded, forming a nodal displacement matrix. Then, the deformation gradient is calculated. Within each element, the in-plane deformation gradient components are calculated using the displacement differences of the three nodes of the element. For a typical region of the stamped part, the in-plane component of the deformation gradient ranges from approximately 0.8 to 1.2, indicating that there is a certain degree of tensile or compressive deformation in this region.

[0077] Strain components are calculated based on deformation gradient. The strain components include normal strain and shear strain, obtained through deformation gradient calculation. For the forming process of a certain automobile fender, the maximum normal strain in the critical deformation region is approximately 0.18, and the maximum shear strain is approximately 0.12. A deformation tensor is constructed using the strain components, and the principal and secondary strain components are solved through tensor decomposition. The principal strain components represent the main direction and degree of deformation, while the secondary strain components represent the secondary deformation direction and degree. For the deep-drawing region of the stamped part, the average principal strain is approximately 0.15, and the average secondary strain is approximately -0.08, which conforms to the principle of approximately constant material volume.

[0078] The wall thickness variation at each node is calculated based on the principal and secondary strain components. The wall thickness variation is obtained by combining the principal and secondary strains. For automotive panel stampings, the wall thickness reduction rate in the drawing region is approximately 12% to 15%, the reduction rate in the corner region can reach up to 18%, while the wall thickness increase rate in the flange region is approximately 5% to 8%.

[0079] Finally, the wall thickness change is combined with the node displacement to calculate the depth change value. The depth change value is defined as the distance between the node's displacement in the forming direction and the original sheet plane. For cup-shaped stampings, the depth change value in the central region is approximately 20mm to 25mm, and the depth change value in the edge region is approximately 5mm to 8mm.

[0080] In this embodiment, region optimization is performed by introducing mesh quality parameters, which improves the geometric accuracy and computational stability of subsequent analyses. By establishing a deformation feature calculation matrix and extracting core data such as nodal displacements and strain components, the local deformation behavior during the stamping process can be comprehensively reflected. Furthermore, by combining the principal strain components, secondary strain components, and wall thickness and depth variations, multidimensional geometric deformation feature parameters are constructed, enabling precise quantification of complex deformation regions. This provides a data foundation for subsequent strain path construction, risk area identification, and defect prediction, thereby significantly improving the accuracy and real-time performance of stamping formability analysis.

[0081] In one optional implementation, constructing a strain path curve based on geometric deformation characteristic parameters, calculating the strain increment of the strain path curve between adjacent time points, and calculating the strain acceleration based on the strain increment includes:

[0082] Gaussian filtering and mean filtering are applied to the geometric deformation characteristic parameters to obtain filtered deformation characteristic data. Cubic spline interpolation is then used to spatially reconstruct the filtered deformation characteristic data to obtain a continuously distributed deformation characteristic field.

[0083] The principal strain field and the secondary strain field are established based on the continuously distributed deformation characteristic field. The strain path curve between the principal strain field and the secondary strain field is fitted by the nonlinear least squares method with adaptive weight coefficients. The strain path curve includes the principal strain component and the secondary strain component.

[0084] The strain path curve is decomposed into principal strain vector and secondary strain vector by orthogonal direction vector. The difference between principal strain vector and secondary strain vector is calculated based on the displacement increment and time increment of adjacent time points to obtain the strain increment vector. The strain increment vector is dynamically accumulated within the sampling time window to obtain the strain increment accumulation value. The strain acceleration is calculated based on the rate of change of the strain increment accumulation value. The strain acceleration characterizes the dynamic characteristics of local deformation.

[0085] For example, geometric deformation characteristic parameters are acquired. These parameters are typically obtained through image processing techniques, digital image correlation techniques, or other sensing devices, and include, but are not limited to, data such as displacement field and strain field. For instance, for a metal sheet undergoing stretching, displacement data points on its surface can be collected, including 1000 data points for displacement in the x and y directions.

[0086] The acquired geometric deformation feature parameters are filtered. In this embodiment, a combination of Gaussian filtering and mean filtering is used. Gaussian filtering employs a Gaussian kernel with a standard deviation of 1.5 and a window size of 5×5 pixels; mean filtering uses a sliding window of 3×3 pixels. This filtering combination effectively eliminates random noise and outliers in the data. For example, an outlier of 0.85mm in the original displacement data is corrected to 0.32mm, which better matches the distribution of surrounding data.

[0087] After filtering, cubic spline interpolation is used to spatially reconstruct these discrete data points, obtaining a continuously distributed deformation feature field. 100 control points are selected during interpolation, and the interpolation precision is set to 0.001 to ensure sufficient smoothness and accuracy of the reconstructed deformation feature field. Through cubic spline interpolation, the data originally collected in a 5mm×5mm grid is reconstructed into a continuously distributed pattern, allowing deformation values ​​to be calculated at any location.

[0088] Based on the reconstructed continuous deformation feature field, principal strain fields and secondary strain fields are established. The principal strain field represents the strain component in the direction of the maximum principal strain, and the secondary strain field represents the strain component in the direction perpendicular to the principal strain. In practice, the eigenvalues ​​and eigenvectors of the strain tensor are calculated for each spatial point. Larger eigenvalues ​​correspond to the principal strain, and smaller eigenvalues ​​correspond to the secondary strain. For example, at a certain point, the calculated principal strain value is 0.12, with a direction of 45 degrees; the secondary strain value is -0.05, with a direction of 135 degrees.

[0089] A nonlinear least squares method with adaptive weighting coefficients is used to fit the strain path curve between the principal strain field and the secondary strain field. The weighting coefficients are dynamically adjusted according to the reliability of the data points, with higher weights for points with higher reliability and lower weights for points with lower reliability. Initially, all weights are set to 1.0, and during iteration, they are adjusted based on the residual magnitude; points with residuals exceeding a threshold of 0.01 have their weights halved. A quadratic polynomial is chosen as the basic fitting function during the fitting process, with a maximum of 100 iterations and a convergence accuracy of 0.0001. Through fitting, the strain path curve characterizing the relationship between the principal strain and the secondary strain is obtained. For example, for a certain metal plate, the fitted relationship between the secondary strain and the principal strain can be expressed as: secondary strain equals -0.32 multiplied by the square of the principal strain minus 0.15 multiplied by the principal strain plus 0.002.

[0090] The strain path curve is decomposed into orthogonal direction vectors to obtain the principal strain vector and the secondary strain vector. The strain vector contains both magnitude and direction information, which together describe the deformation state of the material. For example, at t = 10s, the principal strain vector at a certain point is (0.15, 30°), and the secondary strain vector is (-0.07, 120°).

[0091] The strain increment vector is calculated based on the strain vectors at adjacent time points. For time points t and t+Δt, the difference between the principal strain vector and the difference between the secondary strain vector are calculated to obtain the strain increment vector. Specifically, if the principal strain vector is (0.15, 30°) at t=10s and (0.16, 31°) at t=11s, then the principal strain increment vector is (0.01, 1°).

[0092] The strain increment vector is dynamically accumulated within a sampling time window to obtain the cumulative strain increment value. A 5-second time window with a 1-second sliding step is chosen to reflect the deformation accumulation trend over a recent period. For example, within the 5-second window from t=10s to t=15s, the cumulative principal strain increment is 0.05, and the cumulative secondary strain increment is -0.02. Finally, the strain acceleration is calculated based on the rate of change of the cumulative strain increment value. Strain acceleration characterizes the rate of strain change and is an important indicator for predicting material failure. In the calculation, the cumulative strain increment values ​​of three consecutive time windows are taken, and the rate of change is calculated based on the central difference method. For example, for the cumulative principal strain increment values ​​of t=10s, t=15s, and t=20s, which are 0.05, 0.08, and 0.15 respectively, the calculated principal strain acceleration at t=15s is 0.02 / s². 2 .

[0093] The above method enables the calculation of strain acceleration from geometric deformation characteristic parameters, which can be used for dynamic characteristic analysis and failure prediction during material or structural deformation. By constructing strain path curves and combining vector decomposition and dynamic accumulation analysis of the primary and secondary strain fields, the nonlinear deformation behavior of local areas during stamping can be accurately captured, improving the accuracy of describing complex geometric deformations. By introducing Gaussian filtering and cubic spline interpolation, smooth reconstruction of deformation characteristic data is achieved, enhancing the continuity of strain paths and computational stability. The strain acceleration calculated based on strain increments and their rate of change helps identify transient deformation anomalies and provide early warning of potential instability or crack risks, thereby improving the intelligent sensing capability and predictive analysis accuracy of the forming process.

[0094] In one optional implementation, the strain path curve is compared with the forming limit curve to obtain the safety margin value for each grid node. The deformation risk area is determined based on the strain acceleration and the safety margin value, including:

[0095] The first set of sampling points is obtained by sampling at equal time intervals on the strain path curve, and the second set of sampling points is obtained by sampling at equal strain intervals on the forming limit curve. The first set of sampling points and the second set of sampling points are paired based on the minimum distance principle. The distance between the paired points is calculated to construct a distance mapping matrix. The change segment is identified according to the gradient change rate of the values ​​in the distance mapping matrix. The sampling point density is increased in the change segment. All sampling points are reconstructed by piecewise cubic spline interpolation algorithm.

[0096] Calculate the vertical distance from the sampling point on the strain path curve to the reconstructed forming limit curve, construct a nonlinear weighting function, use the nonlinear weighting function to calculate the safety margin component value of the sampling point, and sum the safety margin component values ​​of all sampling points by weight to obtain the safety margin value.

[0097] The safety margin value and strain acceleration of each grid node are combined to form a risk judgment feature pair. Potential risk nodes are identified based on the preset double threshold constraint. The spatial distance and strain gradient continuity between adjacent risk nodes are calculated. Risk nodes with a spatial distance smaller than the grid feature size and a continuously changing strain gradient are merged into an initial risk region. The boundary of the initial risk region is corrected according to the strain gradient distribution of the nodes to determine the final deformation risk region.

[0098] For example, the strain path curve data and the forming limit curve data of the material for each grid node during the forming process are acquired. The system obtains the changes in principal and secondary strains of each node during the forming process from finite element simulation or real-time monitoring to form the strain path curve. The forming limit curve of the material can be obtained through standard tests or retrieved from a material database.

[0099] For sampling the strain path curve, the system samples at equal time intervals. For example, in a simulation with a total forming time of 1 second, one point is collected every 0.05 seconds, resulting in 20 sampling points in the first group. Each sampling point contains the principal strain value and the secondary strain value, recorded as (ε1, ε2). For example, the sampling point of a certain node at t = 0.25 seconds might be (0.12, 0.05), indicating that the principal strain is 0.12 and the secondary strain is 0.05.

[0100] For sampling the forming limit curve, the system samples at equal strain intervals. For example, on a curve with a principal strain range of 0 to 0.6, a point is collected every 0.03 principal strain values, resulting in 21 second set of sampling points. In the biaxial tensile region (where the secondary strain is positive), a sampling point might be (0.36, 0.18); in the uniaxial tensile region (where the secondary strain is negative), a sampling point might be (0.45, -0.15).

[0101] The system pairs the first set of sampling points with the second set of sampling points based on the principle of minimum distance. For each sampling point on the strain path curve, the Euclidean distance between it and all sampling points on the forming limit curve is calculated, and the point with the smallest distance is selected as its pairing point. For example, the point (0.24, 0.08) on the strain path curve may be paired with the point (0.27, 0.09) on the forming limit curve, with a distance of 0.0374. The system constructs a distance mapping matrix for the distance values ​​between all paired points. By analyzing the gradient rate of change of the distance values ​​in the matrix, segments with large changes are identified. For example, if the distances from three consecutive sampling points to the forming limit curve are 0.15, 0.08, and 0.03, respectively, then the gradient change in this segment is large, and the sampling point density needs to be increased. In the changing segments, the system inserts additional sampling points, for example, inserting four additional points between two points that were originally spaced 0.05 seconds apart, reducing the sampling interval to 0.01 seconds.

[0102] After adjusting the sampling point density, the system reconstructs the forming limit curve using a piecewise cubic spline interpolation algorithm to ensure the curve's smoothness and continuity. During reconstruction, appropriate boundary conditions are set for each segment of the curve to ensure the continuity of the first and second derivatives of adjacent segments at the connection points. To calculate the safety margin, the system first calculates the vertical distance from each sampling point on the strain path curve to the reconstructed forming limit curve. For example, the vertical distance from the point (0.18, 0.06) on the strain path curve to the forming limit curve might be 0.12. The system constructs a nonlinear weighting function, giving greater weight to points closer to the forming limit curve. For example, a weighting function W(d) = 1 / (1+5d) can be used, where d is the vertical distance. When d = 0.12, the weight is 0.625; when d = 0.05, the weight is 0.8.

[0103] The system uses this nonlinear weighting function to calculate the safety margin component value of each sampling point, which is the product of the vertical distance and the corresponding weight. Then, the safety margin component values ​​of all sampling points are weighted and summed to obtain the safety margin value of the grid node. For example, if the safety margin component values ​​of a node's sampling points are 0.075, 0.096, 0.048, etc., the weighted summed safety margin value is 0.108.

[0104] Simultaneously, the strain acceleration, i.e., the rate of change of the strain rate, is calculated for each grid node. For example, if the principal strain of a node increases from 0.28 to 0.35 between t = 0.4 seconds and t = 0.5 seconds, with a strain rate of 0.7; and the principal strain increases from 0.35 to 0.46 between t = 0.5 seconds and t = 0.6 seconds, with a strain rate of 1.1, then the strain acceleration of this node at t = 0.5 seconds is 0.4.

[0105] The system uses the safety margin value and strain acceleration of each grid node to form a risk assessment feature pair. Potential risk nodes are identified based on preset dual threshold constraints (e.g., safety margin value less than 0.1 and strain acceleration greater than 0.3). For identified risk nodes, the system calculates their spatial distance and strain gradient continuity. If the spatial distance between adjacent risk nodes is less than the grid feature size (e.g., 2 mm) and the strain gradient changes continuously (gradient difference less than 15%), these nodes are merged into an initial risk region. For example, nodes A (50, 30), B (52, 30), and C (51, 32) have a spatial distance of less than 2 mm and strain gradients of 0.023, 0.025, and 0.026 respectively, with a difference of less than 15%, and can be merged into one initial risk region. The boundary of the initial risk region is corrected according to the strain gradient distribution of the nodes. At the boundary, if the strain gradient changes abruptly (rate of change greater than 30%), the boundary position is adjusted. For example, at the boundary node D(54, 30), the strain gradient abruptly changes from 0.025 to 0.016, a change rate of 36%, exceeding the 30% threshold. Therefore, node D is excluded from the risk region. In this way, the system identifies the final deformation risk region, providing accurate risk warnings for the stamping process.

[0106] Figure 2 This is a comparison chart of the safety margin weighting functions in embodiments of the present invention, as shown below. Figure 2 As shown in the figure, this diagram compares weight functions with different safety margins. The solid line represents the nonlinear weight function of this technical solution, the dashed line represents the linear weight function used in the traditional Zhenga algorithm, and the dotted-dash line represents the exponential weight function used in the MLFCD algorithm. The nonlinear weight function of this technical solution has three distinct characteristic ranges: when the vertical distance d < 0.3, the function exhibits high sensitivity characteristics near P1 (0.2, 0.78), with a slope of approximately -1.25; when 0.3 ≤ d < 0.6, the function exhibits medium sensitivity characteristics near P2 (0.4, 0.45), with a slope of approximately -0.78; and when d ≥ 0.6, the function exhibits low sensitivity characteristics near P3 (0.58, 0.10), with a slope of approximately -0.25. This piecewise nonlinear design allows the weighting function to exhibit different sensitivities in different risk ranges. Particularly when the vertical distance is small (close to the forming limit curve), the weight value changes more sensitively, more accurately reflecting the forming limit state of the material. Conversely, when the vertical distance is large (far from the forming limit curve), the weight value changes relatively smoothly, avoiding misjudgments caused by oversensitivity. Compared to the traditional Zhenga algorithm and MLFCD algorithm, the nonlinear weighting function of this technical solution exhibits superior recognition characteristics in risk assessment, reducing the false alarm rate in high-risk areas by 31.7% and the false alarm rate in low-risk areas by 28.9%, resulting in an overall improvement in recognition accuracy of 40.2%.

[0107] Existing technologies for identifying deformation risks during stamping typically rely on static strain criteria or single safety margin assessments, which struggle to capture dynamic changes during the forming process. This is particularly problematic in areas with complex and rapidly evolving strain paths and forming limits, leading to delayed risk assessments and insufficient accuracy in region identification. This application addresses these issues by constructing a distance mapping between the strain path curve and the forming limit curve, introducing gradient rate of change analysis to achieve dynamic sampling densification of high-change sections, effectively improving the accuracy of curve reconstruction. Furthermore, it incorporates a nonlinear weighting function to differentiate the safety margin assessment of each sampling point, ensuring that the safety margin value more accurately reflects the deformation margin of each node relative to the instability boundary. Simultaneously, by introducing strain acceleration into the risk assessment process and constructing bivariate judgment feature pairs, it overcomes the shortcomings of existing methods that rely solely on static indicators to identify rapidly deforming trend regions. Finally, it clusters potential risk nodes and corrects region boundaries based on spatial distance and strain gradient continuity, improving the coherence and accuracy of risk region identification. The above improvements, based on the capture of dynamic evolution features, achieve accurate early warning and spatial positioning of local forming instability trends without significantly increasing computational complexity, thereby enhancing the timeliness and reliability of risk identification.

[0108] In one optional implementation, historical stamping data containing defect types and characteristics is acquired, strain abrupt change characteristics of the defect formation process are analyzed, a correspondence between strain abrupt change characteristics and defect types is established, defect warning criteria are obtained, and deformation risk areas are analyzed based on the defect warning criteria to obtain predicted defect locations and defect types, including:

[0109] Acquire historical stamping data containing defect types and defect characteristics, calculate strain gradient tensor, strain path curvature and energy dissipation rate on historical stamping data to obtain multi-dimensional strain abruptness characteristics;

[0110] Recursive Bayesian estimation is used to update the weights of the multidimensional strain abrupt features to generate fused feature data. Based on the defect types in historical stamping data and the fused feature data, the temporal variation characteristics and spatial distribution characteristics of the defect formation process are analyzed, and a mapping relationship between strain abrupt features and defect types is established.

[0111] The mapping relationship is used to construct a feature-defect association network. The evolution path of the fused feature data in the association network is analyzed, and an early warning criterion including feature thresholds and defect evolution rules is established.

[0112] Real-time acquisition of strain field and temperature field data; extraction of principal direction and principal value change rate of strain gradient tensor in deformation region; establishment of stress transmission path diagram in deformation region; analysis of abrupt change points of strain path curvature and fluctuation degree of energy dissipation rate during stress transmission process; calculation of deformation instability index in combination with defect evolution law in early warning criterion; determination of defect location based on propagation characteristics of instability index; determination of defect type based on strain state combination during instability process.

[0113] For example, historical stamping data containing defect types and characteristics is first acquired. During implementation, 500 sets of stamping test data under different process parameters can be extracted from the database. Each set of data includes complete strain field data, temperature field data, final workpiece quality rating, and defect distribution images for each stage of the forming process. This historical data covers five common defect types: wrinkles, cracks, springback, surface scratches, and material thinning.

[0114] The strain gradient tensor, strain path curvature, and energy dissipation rate are calculated from the acquired historical stamping data to obtain multi-dimensional strain abrupt change characteristics. Specifically, the workpiece surface is divided into 10,000 mesh elements. Differential calculations are performed on the displacement data of each mesh element at 200 time steps during the forming process to obtain strain field data. For each mesh element, the strain gradient tensor between adjacent time steps is calculated, including the principal direction and amplitude changes; the strain path curvature, i.e., the rate of change of the strain direction, is calculated; and the energy dissipation rate, i.e., the change of plastic power per unit time, is calculated. For example, for a certain region of the workpiece during the forming process, the principal value of the strain gradient tensor changes from 0.05 to 0.25, the strain path curvature increases from 0.02 rad / mm to 0.15 rad / mm, and the energy dissipation rate increases from 2 J / s to 18 J / s.

[0115] Recursive Bayesian estimation was employed to update the weights of multi-dimensional strain abrupt change features, generating fused feature data. Initially, the weights for the strain gradient tensor, strain path curvature, and energy dissipation rate were set to 0.3, 0.3, and 0.4, respectively. Then, these weights were recursively updated based on the correlation between defect locations and each feature in historical data. After 50 iterations, the weights converged to 0.45 for the strain gradient tensor, 0.35 for the strain path curvature, and 0.2 for the energy dissipation rate. Using the updated weights, the multi-dimensional features of each grid cell were weighted and fused to generate a comprehensive feature value.

[0116] Based on defect type and fusion feature data from historical stamping forming data, this study analyzes the temporal variation and spatial distribution characteristics of defects during their formation process, establishing a mapping relationship between strain abrupt changes and defect types. Specifically, for each defect type, fusion feature data of the defective and non-defective regions during the forming process are extracted to construct a feature vector. Comparative analysis reveals that wrinkle defects are typically accompanied by a more than 3-fold increase in strain path curvature within a short time (within 15ms); the principal value of the strain gradient tensor in cracked defect regions exceeds 0.3 at the end of forming (the last 25% of the forming stage); and the energy dissipation rate fluctuation in material thinning regions exceeds five times that of the surrounding regions.

[0117] A feature-defect correlation network is constructed by mapping relationships. The evolution path of fused feature data in the correlation network is analyzed to establish early warning criteria that include feature thresholds and defect evolution patterns. In implementation, 200 typical defect samples from historical data are used to construct a five-layer correlation network, with each layer representing a different forming stage. Network nodes represent feature states, and edges represent state transition probabilities. Through network analysis, early warning criteria for each defect type are obtained: when the rate of change of the principal value of the strain gradient tensor in a certain area exceeds 0.02 / s and the duration exceeds 0.5s, it is judged as a high-risk area for wrinkling; when the number of abrupt changes in the curvature of the strain path is less than 100 mm², it is judged as a high-risk area for wrinkling. 2 If there are more than 5 cracks within a time step and the energy dissipation rate fluctuation coefficient is greater than 0.8, it is judged as a high risk of cracking; if the principal direction of the strain gradient tensor changes by more than 30° in adjacent time steps and the energy dissipation rate is less than 60% of the average value of the surrounding area, it is judged as a high risk of rebound.

[0118] Real-time acquisition of strain and temperature field data is performed to extract the principal directions and rates of change of the principal values ​​of the strain gradient tensor in the deformation region, thus establishing a stress transmission path diagram in the deformation region. In practice, 12 strain sensors and 8 temperature sensors distributed at key locations in the mold are used, with a sampling frequency of 200Hz, to acquire data during the forming process in real time. Based on the acquired data, the principal directions and rates of change of the principal values ​​of the strain gradient tensor for each mesh element are calculated, and a stress transmission path diagram is plotted, displaying the direction and intensity of stress transmission from high to low.

[0119] The deformation instability index is calculated by analyzing the abrupt changes in strain path curvature and the fluctuations in energy dissipation rate during stress transfer, combined with the defect evolution law in the early warning criteria. In practice, locations where the strain path curvature changes by more than 150% within a short time (50ms) are identified as abrupt change points; the ratio of the standard deviation to the mean of the energy dissipation rate for each grid element is calculated as the fluctuation level; based on the early warning criteria, the instability index F is calculated as: F = 0.45 × (strain gradient change rate / threshold) + 0.35 × (number of curvature abrupt changes / threshold) + 0.2 × (energy fluctuation level / threshold). When the F value exceeds 1, the region is considered to have a defect risk.

[0120] The location of defects is determined based on the propagation characteristics of the instability index, and the type of defect is determined based on the combination of strain states during the instability process. During implementation, for regions with an instability index F greater than 1, their spatial distribution and temporal evolution characteristics are analyzed. If the F value spreads radially and the central F value exceeds 1.5, the center point is identified as the defect initiation location; if the F value exhibits a banded distribution with a bandwidth less than 5 mm, the center line of the banded region is identified as the defect location. Based on the strain state combination characteristics of the defect region, such as the principal value of the strain gradient, peak curvature, and energy dissipation characteristics, and in accordance with the warning criteria, the specific type of defect—wrinkle, crack, springback, surface scratch, or material thinning—is determined.

[0121] In this embodiment, by fusing and modeling multidimensional strain abrupt change features in historical stamping data, a feature-defect correlation network is constructed and its evolutionary patterns are extracted, achieving in-depth mining of defect formation mechanisms and type identification. Dynamically updating feature weights through recursive Bayesian estimation improves the accuracy of expressing the contribution of different features to defect evolution. Real-time acquisition of strain and temperature field data, combined with stress transmission path and instability index propagation analysis, enables early warning, precise location, and type identification of defects. Overall, this method significantly enhances the ability to predict defects, identify risks, and trace the source of defects in complex stamping processes, providing support for high-precision quality control and intelligent decision-making.

[0122] In one optional implementation, a stress transmission path diagram of the deformation region is established, the abrupt change points of strain path curvature and the degree of energy dissipation rate fluctuation during the stress transmission process are analyzed, the deformation instability index is calculated in conjunction with the defect evolution law in the early warning criterion, the defect location is determined based on the propagation characteristics of the instability index, and the defect type is determined according to the strain state combination during the instability process, including:

[0123] The deformed region is discretized into triangular mesh elements. The difference in the principal stress direction between adjacent mesh elements is calculated as the transmission impedance. The minimum spanning tree algorithm is used to connect the minimum transmission impedance elements to obtain the stress transmission path diagram of the deformed region.

[0124] Strain field data are collected along the stress transfer path diagram to construct a strain path, the curvature change of the strain path is calculated, and the curvature abrupt change points and strain energies are extracted and combined into abrupt change feature vector.

[0125] The energy transfer path is determined based on the stress transfer path diagram. The dissipation rate of strain energy, its fluctuation degree, and the dominant frequency component are calculated along the energy transfer path and combined into a fluctuation characteristic vector.

[0126] The mutation feature vector and the fluctuation feature vector are mapped onto the stress transmission path diagram to obtain the energy dissipation fluctuation feature score and the stress transmission path feature score, respectively. Combined with the defect evolution law in the early warning criterion, the deformation instability index is calculated by dynamic weight combination.

[0127] A spatial distribution map is established based on the deformation instability index, the gradient field of the deformation instability index is calculated, and the direction and speed of instability propagation are determined based on the propagation characteristics of the instability index to predict the location of defects.

[0128] The principal strain ratio is calculated using strain field data, the strain path non-proportionality is calculated based on the degree of non-linearity of the strain path, and the local strain concentration factor is calculated based on the strain distribution in the spatial distribution map. The principal strain ratio, strain path non-proportionality, and local strain concentration factor are combined into strain state characteristics, and the defect type is determined based on the value of the strain state characteristics.

[0129] For example, for the deformation area to be detected, full-field strain data of the deformation area is acquired using digital image correlation (DIC). In this embodiment, a high-resolution camera is used to photograph the surface of the deformed specimen to obtain grayscale images before and after deformation. The surface displacement field is calculated using a sub-pixel matching algorithm to obtain the strain distribution information of the deformation area. For example, for a 100mm × 100mm metal plate specimen, a 5-megapixel camera can be used to obtain displacement field data with a resolution of 0.05mm / pixel. The deformation area is discretized into triangular mesh elements. In this embodiment, the Delaunay triangulation algorithm is used to divide the deformation area into uniformly sized triangular elements, generating approximately 10,000 triangular mesh elements with a side length of approximately 0.5mm. For each triangular element, its principal stress direction is calculated based on the strain data acquired by DIC.

[0130] The difference in principal stress directions between adjacent triangular mesh elements is calculated as the transfer impedance. Specifically, for any two adjacent triangular elements A and B, their respective principal stress directions θA and θB are extracted, and the angle difference Δθ = |θA - θB| between the two directions is calculated. This angle difference is normalized to a value between 0 and 1 as the transfer impedance value. For example, when the principal stress directions of the two elements are completely consistent, the transfer impedance is 0; when the principal stress directions of the two elements are perpendicular, the transfer impedance is 1. After calculating the transfer impedance between all adjacent elements, the minimum spanning tree algorithm is used to connect the elements with the minimum transfer impedance. In the implementation process, the Prim algorithm can be used to construct the minimum spanning tree, starting from an element at the boundary of the deformation region, and gradually adding adjacent elements with the minimum transfer impedance until all elements are connected into an acyclic graph structure, thus obtaining the stress transfer path diagram of the deformation region.

[0131] Strain path data is collected along the path in the stress transfer path diagram to construct the strain path. In this embodiment, the values ​​of principal strains ε1 and ε2 are recorded along each node on the path, forming a strain path {(ε1, 1, ε2, 1), (ε1, 2, ε2, 2), ..., (ε1, n, ε2, n)}. The curvature change of the strain path is then calculated. The curvature at each point on the path is calculated using the three-point method. When the curvature value exceeds a threshold of 0.5, the point is marked as a curvature abrupt change point. In practical applications, for aluminum alloy specimens, when loaded to 80% of the yield stress, five curvature abrupt change points were detected in the area around the defect, with abrupt change values ​​ranging from 0.6 to 0.8.

[0132] The energy transfer path is determined based on the stress transfer path diagram, and the strain energy per unit volume and its dissipation rate are calculated along the energy transfer path. For each point on the path, the strain energy per unit volume U is calculated, and its spatial gradient ΔU / Δx is calculated as the strain energy dissipation rate. The spectral characteristics of the dissipation rate are analyzed by Fast Fourier Transform, and the dominant frequency component and fluctuation amplitude are extracted. In this embodiment, the fluctuation amplitude of the dissipation rate in the healthy material region is within ±5%, while the fluctuation amplitude in the defective region can reach ±30%, and the dominant frequency component is concentrated between 0.3Hz and 0.5Hz.

[0133] The mutation feature vector and fluctuation feature vector are mapped onto the stress transmission path diagram to obtain the energy dissipation fluctuation feature score and the stress transmission path feature score, respectively. In this embodiment, a scoring system from 0 to 100 is used, with the score in the healthy region typically below 20, while the score in the defective region exceeds 60. Combining the defect evolution law of the material, the deformation instability index is calculated through dynamic weight combination. For aluminum alloy materials, the mutation feature weight is set to 0.6, and the fluctuation feature weight is set to 0.4. The calculated instability index is generally less than 0.2 in the healthy region, while it can reach above 0.7 in the defective region. A spatial distribution map is established based on the deformation instability index, and its gradient field is calculated. The direction with the largest gradient value is identified as the instability propagation direction, and the magnitude of the gradient value corresponds to the instability propagation speed. In practical applications, when there is a region in the instability index gradient field with a gradient value greater than 0.1 / mm, and this region is radially distributed, the center of this region can be determined as the defect location. Finally, the principal strain ratio λ = ε1 / ε2 was calculated using strain field data, the strain path nonproportionality (NPL) was calculated, and combined with the local strain concentration factor (LCF), a strain state characteristic vector (λ, NPL, LCF) was formed. The defect type was determined based on the range of values ​​of this characteristic vector: when λ > 5 and NPL < 0.2, it was identified as a crack-type defect; when 1 < λ < 3 and NPL > 0.5, it was identified as a void-type defect; when λ ≈ 1 and LCF > 3, it was identified as an inclusion-type defect. For example, for a steel plate containing a crack defect, λ = 7.2, NPL = 0.15, and LCF = 3.6 were measured, successfully identifying the crack-type defect and accurately locating its position with an error of less than 1 mm.

[0134] Figure 3 The spatial distribution and gradient field thermogram of the deformation instability index are shown in the embodiment of the present invention. Figure 3As shown in the figure, the deformation area is represented by an 8×4 grid, with the horizontal axis ranging from 0-140mm and the vertical axis ranging from 0-60mm. Each grid cell clearly displays a specific instability index value, ranging from 0.12 to 0.95, with the color transitioning from light blue (low value) to dark blue (high value), forming a visually intuitive gradient. The core instability area is located at coordinates (80mm, 40mm), where the instability index reaches its highest value of 0.95. This location is specifically marked by a black circle as the predicted defect location. Three arrows mark the directions of instability expansion from this point outwards, pointing to the upper left, upper, and upper right, indicating a tendency for the instability area to expand in multiple directions. The instability index gradient field is steepest in the central region; as it spreads outwards from the grid (80mm, 40mm), the index value rapidly decreases from 0.95 to 0.88, 0.72, etc., forming a clear instability gradient. This spatial distribution map clearly reveals the propagation characteristics of the deformation instability index, providing an intuitive basis for predicting defect locations and determining the direction and speed of instability propagation. Compared with traditional thermodynamic methods, it can more accurately identify potential high-risk areas. In particular, the identification of high gradient areas of the instability index is of great guiding significance for implementing preventive measures.

[0135] Existing technologies for defect prediction in the stamping process largely rely on static thresholds or local strain exceedance judgments, making it difficult to effectively identify the dynamic evolution process of defect formation. Furthermore, the lack of systematic modeling of stress and energy transmission paths leads to insufficient accuracy in predicting defect location and type. This application proposes discretizing the deformation region into a triangular mesh and constructing a transmission impedance based on the difference in the principal stress directions. A minimum spanning tree algorithm is used to construct a stress transmission path map, dynamically capturing the actual stress propagation path. Furthermore, multi-source dynamic feature vectors, such as strain path curvature abrupt changes and energy dissipation fluctuations, are extracted based on this, and a deformation instability index is constructed using a dynamic weight combination method, achieving continuous monitoring and early warning of defect evolution trends. Regarding defect identification, this application introduces principal strain ratio, strain path non-proportionality, and local strain concentration factor as comprehensive strain state features, significantly improving the discriminative power of defect type determination compared to traditional methods that rely solely on strain amplitude. These improvements, based on the stress-strain-energy coupling transmission relationship, construct a complete dynamic instability assessment framework, enabling earlier and more accurate identification and location of potential defects during the forming process, significantly improving the intelligence and foresight of stamping quality control.

[0136] In one optional implementation, a mesh refinement factor is calculated using a mesh optimization equation. The initial mesh data is then locally optimized based on this refinement factor to obtain optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results include:

[0137] The plastic power density within the cell is calculated based on the mesh optimization equation, and a stress-strain field distribution function containing deformation energy density and strain localization terms is established based on the plastic power density.

[0138] The stress gradient and strain gradient of the mesh element are calculated using the stress-strain field distribution function. A critical refinement threshold is established based on the plastic power density. The mesh refinement coefficient is obtained by comparing the stress gradient and strain gradient with the critical refinement threshold.

[0139] The mesh cells are locally refined according to the mesh refinement coefficient. A strain path function is introduced into the refined mesh cells. The pre-acquired material strain hardening index is updated into the refined mesh data. Finite element iterative calculation is performed using the optimized mesh data. The strain rate gradient and mesh refinement coefficient are updated based on the calculation results. The mesh density distribution is dynamically adjusted, and the stamping forming analysis results are output.

[0140] For example, the geometric model of the part corresponding to the stamped structure is first discretized, and then transformed into an initial mesh model required for structural finite element analysis using two-dimensional or three-dimensional meshing techniques. This initial mesh model contains multiple finite elements, each associated with material properties, boundary conditions, and load paths. The mesh density of the initial mesh is usually set according to the geometric complexity and forming path; for example, denser meshes are used in structurally sensitive areas such as corners, stiffeners, and openings, while larger elements are used in flat areas to reduce computational complexity.

[0141] Based on the initial finite element mesh model, mesh optimization equations are constructed to evaluate the deformation response intensity of each mesh element. These equations comprehensively assess the deformation energy density distribution and strain localization trend within the element, and establish a plastic power density index accordingly. Plastic power density is a parameter characterizing the rate of plastic energy consumption during loading, which can be described by the product of material hardening behavior and local strain rate. This index reflects the plastic activity of the mesh element at the current loading stage; regions with higher plastic power density indicate more severe deformation and more concentrated risk, thus requiring finer-grained simulation analysis.

[0142] In each grid cell, the stress-strain field distribution function is calculated based on the aforementioned plastic power density. This distribution function comprises two main components: a deformation energy density term and a strain localization term. The deformation energy density term describes the energy intensity absorbed per unit volume of material during plastic deformation and is typically directly related to the plastic power density. The strain localization term describes the strain gradient trend in a specific direction and can reflect potential locations prone to instability, such as local shear bands and necking regions.

[0143] Based on the stress-strain field distribution function, the stress gradient and strain gradient within each mesh element are extracted. The stress gradient represents the rate of change of stress between different points within the mesh, while the strain gradient represents the steepness of the strain distribution in space. High stress or strain gradients typically appear at deformation boundaries, abrupt geometric changes, or multi-path stress concentration areas, and are important indicators of potential defects or instability. Next, a critical refinement threshold is constructed based on the plastic power density. This threshold serves as the criterion for determining whether mesh refinement is necessary, measuring whether the current mesh element requires further refinement. Specifically, when the plastic power density of a certain element exceeds this critical threshold, or the combined result of its stress gradient and strain gradient reaches a specified critical state, it can be determined that the element has a potential instability trend and requires local refinement.

[0144] In practice, by setting multiple refinement levels corresponding to different critical refinement coefficients, regions with different response intensities are assigned different mesh refinement factors. The mesh refinement coefficient is used to control the accuracy of the control unit meshing; a higher value indicates a higher degree of refinement. Based on the refinement coefficient, the initial mesh data is adjusted cell by cell. High-risk areas are optimized through local densification, while low-risk areas maintain their original mesh density or are appropriately simplified. During the optimization process, it is necessary to maintain the continuity and quality of the mesh, such as controlling parameters like control unit angles and area ratios, to avoid generating excessively small or highly distorted cells.

[0145] In the optimized mesh structure described above, a strain path function is introduced to describe the evolution of the deformation history. The strain path function is a function that reflects the changing trend of the strain state of a material at different loading stages. It can be obtained by fitting historical strain data and reflects the principal strain direction, strain amplitude, and strain increment characteristics of the material at a specific time point and loading path. The strain path function is mapped to each mesh element within the refined region, enabling these elements to dynamically represent the actual deformation path in subsequent analyses.

[0146] Simultaneously, the pre-acquired material strain hardening index is updated in the newly generated refined mesh data. The strain hardening index is a parameter describing the hardening ability of a material during plastic deformation; different materials, or even the same material, exhibit different hardening behaviors at different temperatures and loading rates. Introducing the material strain hardening index helps to accurately calculate the stress response behavior of materials, improving the accuracy of forming analysis.

[0147] After mesh refinement and parameter updates, iterative finite element calculations are performed based on the optimized mesh data. This process employs explicit or implicit finite element methods to dynamically simulate the entire stamping process and outputs results such as stress field, strain field, and thickness distribution in real time. By re-evaluating the strain rate gradient and mesh refinement factor in each iteration step, the mesh density distribution can be dynamically adjusted. For example, if a rapid strain concentration trend is found in the leading edge region of the part in the tenth iteration, the refinement factor for that region is increased by one level to further refine the mesh and obtain higher simulation accuracy.

[0148] Taking the simulation of automotive door panel stamping as an example, in the initial meshing stage, the door panel contour is divided into medium-density meshes, and the plastic power density threshold is set as the initial judgment standard. During the calculation, it was found that the strain localization was obvious in the corner areas of the door panel and the roots of the tension ribs, and the plastic power density was much higher than the average value. At the same time, the stress gradient direction was concentrated. It was decided to increase the refinement factor of this area from 1.0 to 2.5 to automatically complete the local refinement. After updating the mesh, the strain hardening index curve of the material being duplex steel as a function of temperature was loaded into each mesh element, and an explicit finite element method was used for iterative solution. In the later stages of iteration, the system continued to dynamically adjust the mesh density based on the strain rate change trend, making the prediction of necking location and crack tendency area in the simulation results more accurate and highly consistent with the crack location of the test piece.

[0149] Ultimately, the system outputs stamping analysis results, including but not limited to: element stress-strain cloud maps, thickness variation maps, and crack risk zone prediction maps. These results can be used to guide actual die compensation, blank holder force adjustment, and process path optimization, significantly improving the yield and stability of actual stamping production.

[0150] In this embodiment, by introducing plastic power density and stress-strain gradient analysis, dynamic optimization of the mesh refinement process is achieved. This enables accurate identification of areas with severe deformation and local mesh refinement, thereby improving the resolution and accuracy of the finite element analysis. By constructing a strain path function and updating material hardening parameters, the simulation process more closely resembles actual forming behavior, enhancing the physical realism of the model. Simultaneously, the mesh refinement and iterative adjustment process can adaptively adapt to deformation evolution trends, effectively avoiding overcomputation and improving overall computational efficiency. This approach significantly enhances the ability of stamping simulation to predict instability risks such as cracks and necking, providing a highly reliable reference for mold design and process optimization.

[0151] Figure 4 This is a schematic diagram of the structure of the AI-assisted rapid analysis and prediction system for the formability of stamped parts according to an embodiment of the present invention. Figure 4 As shown, the system includes:

[0152] The first unit is used to acquire the three-dimensional data of the stamped part and perform mesh generation to obtain the initial mesh data, and extract the geometric deformation feature parameters from the initial mesh data;

[0153] The second unit is used to construct strain path curves based on geometric deformation characteristic parameters, calculate the strain increment between adjacent time points of the strain path curve, calculate the strain acceleration based on the strain increment, compare the strain path curve with the forming limit curve to obtain the safety margin value of each grid node, and determine the deformation risk area based on the strain acceleration and the safety margin value.

[0154] The third unit is used to acquire historical stamping data containing defect types and defect characteristics, analyze the strain abrupt change characteristics of the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze the deformation risk area based on the defect warning criteria, and obtain the predicted defect location and defect type.

[0155] The fourth unit is used to calculate the strain rate gradient at the predicted defect location. The strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation. The initial mesh data is then locally optimized based on the mesh refinement coefficient to obtain the optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

[0156] A third aspect of the present invention provides an electronic device, comprising:

[0157] processor;

[0158] Memory used to store processor-executable instructions;

[0159] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0160] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0161] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI-assisted rapid analysis and prediction method for the formability of stamped parts, characterized in that, include: The three-dimensional data of the stamped part is acquired and meshed to obtain the initial mesh data. Geometric deformation feature parameters are then extracted from the initial mesh data. Strain path curves are constructed based on geometric deformation characteristic parameters. The strain increment between adjacent time points is calculated, and the strain acceleration is calculated based on the strain increment. The strain path curves are compared with the forming limit curves to obtain the safety margin values ​​of each grid node. Deformation risk areas are determined based on the strain acceleration and safety margin values. Acquire historical stamping data containing defect types and characteristics, analyze strain abrupt change characteristics in the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze deformation risk areas based on defect warning criteria, and obtain predicted defect locations and defect types; The strain rate gradient at the predicted defect location is calculated, and the strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation, and the initial mesh data is locally optimized based on the mesh refinement coefficient to obtain optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

2. The method according to claim 1, characterized in that, The three-dimensional data of the stamped part is acquired and meshed to obtain initial mesh data. Geometric deformation feature parameters are extracted from the initial mesh data, including: Point cloud data of stamped parts is acquired using a 3D scanner and preprocessed. Initial mesh data is constructed based on the preprocessed point cloud data. The initial mesh data includes node location information and cell topology. Mesh quality parameters of the initial mesh data are calculated, and the mesh area is optimized according to the mesh quality parameters to obtain an optimized mesh model. Based on the optimized mesh model, a deformation feature calculation matrix is ​​established, which includes nodal displacement, deformation gradient and strain components. The deformation tensor is calculated using the deformation feature calculation matrix, and the principal strain components and secondary strain components are solved using the deformation tensor. The wall thickness variation of each node is calculated based on the principal strain component and the secondary strain component. The wall thickness variation is combined with the node displacement to calculate the depth variation value, and a geometric deformation characteristic parameter containing the principal strain component, the secondary strain component, the wall thickness variation value, and the depth variation value is constructed.

3. The method according to claim 1, characterized in that, Strain path curves are constructed based on geometric deformation characteristic parameters. The strain increments between adjacent time points are calculated, and the strain acceleration is calculated based on the strain increments, including: Gaussian filtering and mean filtering are applied to the geometric deformation characteristic parameters to obtain filtered deformation characteristic data. Cubic spline interpolation is then used to spatially reconstruct the filtered deformation characteristic data to obtain a continuously distributed deformation characteristic field. The principal strain field and the secondary strain field are established based on the continuously distributed deformation characteristic field. The strain path curve between the principal strain field and the secondary strain field is fitted by the nonlinear least squares method with adaptive weight coefficients. The strain path curve includes the principal strain component and the secondary strain component. The strain path curve is decomposed into principal strain vector and secondary strain vector by orthogonal direction vector. The difference between principal strain vector and secondary strain vector is calculated based on the displacement increment and time increment of adjacent time points to obtain the strain increment vector. The strain increment vector is dynamically accumulated within the sampling time window to obtain the strain increment accumulation value. The strain acceleration is calculated based on the rate of change of the strain increment accumulation value. The strain acceleration characterizes the dynamic characteristics of local deformation.

4. The method according to claim 1, characterized in that, By comparing the strain path curve with the forming limit curve, the safety margin value of each grid node is obtained. Based on the strain acceleration and the safety margin value, the deformation risk area is determined, including: The first set of sampling points is obtained by sampling at equal time intervals on the strain path curve, and the second set of sampling points is obtained by sampling at equal strain intervals on the forming limit curve. The first set of sampling points and the second set of sampling points are paired based on the minimum distance principle. The distance between the paired points is calculated to construct a distance mapping matrix. The change segment is identified according to the gradient change rate of the values ​​in the distance mapping matrix. The sampling point density is increased in the change segment. All sampling points are reconstructed by piecewise cubic spline interpolation algorithm. Calculate the vertical distance from the sampling point on the strain path curve to the reconstructed forming limit curve, construct a nonlinear weighting function, use the nonlinear weighting function to calculate the safety margin component value of the sampling point, and sum the safety margin component values ​​of all sampling points by weight to obtain the safety margin value. The safety margin value and strain acceleration of each grid node are combined to form a risk judgment feature pair. Potential risk nodes are identified based on the preset double threshold constraint. The spatial distance and strain gradient continuity between adjacent risk nodes are calculated. Risk nodes with a spatial distance smaller than the grid feature size and a continuously changing strain gradient are merged into an initial risk region. The boundary of the initial risk region is corrected according to the strain gradient distribution of the nodes to determine the final deformation risk region.

5. The method according to claim 1, characterized in that, Historical stamping data containing defect types and characteristics is acquired. The strain abrupt change characteristics of the defect formation process are analyzed, and a correspondence between strain abrupt change characteristics and defect types is established. Defect warning criteria are obtained, and deformation risk areas are analyzed based on these criteria to predict defect locations and types, including: Acquire historical stamping data containing defect types and defect characteristics, calculate strain gradient tensor, strain path curvature and energy dissipation rate on historical stamping data to obtain multi-dimensional strain abruptness characteristics; Recursive Bayesian estimation is used to update the weights of the multidimensional strain abrupt features to generate fused feature data. Based on the defect types in historical stamping data and the fused feature data, the temporal variation characteristics and spatial distribution characteristics of the defect formation process are analyzed, and a mapping relationship between strain abrupt features and defect types is established. The mapping relationship is used to construct a feature-defect association network. The evolution path of the fused feature data in the association network is analyzed, and an early warning criterion including feature thresholds and defect evolution rules is established. Real-time acquisition of strain field and temperature field data; extraction of principal direction and principal value change rate of strain gradient tensor in deformation region; establishment of stress transmission path diagram in deformation region; analysis of abrupt change points of strain path curvature and fluctuation degree of energy dissipation rate during stress transmission process; calculation of deformation instability index in combination with defect evolution law in early warning criterion; determination of defect location based on propagation characteristics of instability index; determination of defect type based on strain state combination during instability process.

6. The method according to claim 5, characterized in that, Establish a stress transmission path diagram in the deformation region, analyze the abrupt changes in strain path curvature and the degree of energy dissipation fluctuation during stress transmission, calculate the deformation instability index based on the defect evolution law in the early warning criteria, determine the defect location based on the propagation characteristics of the instability index, and determine the defect type according to the strain state combination during the instability process, including: The deformed region is discretized into triangular mesh elements. The difference in the principal stress direction between adjacent mesh elements is calculated as the transmission impedance. The minimum spanning tree algorithm is used to connect the minimum transmission impedance elements to obtain the stress transmission path diagram of the deformed region. Strain field data are collected along the stress transfer path diagram to construct a strain path, the curvature change of the strain path is calculated, and the curvature abrupt change points and strain energies are extracted and combined into abrupt change feature vector. The energy transfer path is determined based on the stress transfer path diagram. The dissipation rate of strain energy, its fluctuation degree, and the dominant frequency component are calculated along the energy transfer path and combined into a fluctuation characteristic vector. The mutation feature vector and the fluctuation feature vector are mapped onto the stress transmission path diagram to obtain the energy dissipation fluctuation feature score and the stress transmission path feature score, respectively. Combined with the defect evolution law in the early warning criterion, the deformation instability index is calculated by dynamic weight combination. A spatial distribution map is established based on the deformation instability index, the gradient field of the deformation instability index is calculated, and the direction and speed of instability propagation are determined based on the propagation characteristics of the instability index to predict the location of defects. The principal strain ratio is calculated using strain field data, the strain path non-proportionality is calculated based on the degree of non-linearity of the strain path, and the local strain concentration factor is calculated based on the strain distribution in the spatial distribution map. The principal strain ratio, strain path non-proportionality, and local strain concentration factor are combined into strain state characteristics, and the defect type is determined based on the value of the strain state characteristics.

7. The method according to claim 1, characterized in that, The mesh refinement factor is calculated using the mesh optimization equation. Based on this factor, the initial mesh data is locally optimized to obtain the optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results include: The plastic power density within the cell is calculated based on the mesh optimization equation, and a stress-strain field distribution function containing deformation energy density and strain localization terms is established based on the plastic power density. The stress gradient and strain gradient of the mesh element are calculated using the stress-strain field distribution function. A critical refinement threshold is established based on the plastic power density. The mesh refinement coefficient is obtained by comparing the stress gradient and strain gradient with the critical refinement threshold. The mesh cells are locally refined according to the mesh refinement coefficient. A strain path function is introduced into the refined mesh cells. The pre-acquired material strain hardening index is updated into the refined mesh data. Finite element iterative calculation is performed using the optimized mesh data. The strain rate gradient and mesh refinement coefficient are updated based on the calculation results. The mesh density distribution is dynamically adjusted, and the stamping forming analysis results are output.

8. An AI-assisted rapid analysis and prediction system for the formability of stamped parts, used to implement the method described in any one of claims 1-7, characterized in that, include: The first unit is used to acquire the three-dimensional data of the stamped part and perform mesh generation to obtain the initial mesh data, and extract the geometric deformation feature parameters from the initial mesh data; The second unit is used to construct strain path curves based on geometric deformation characteristic parameters, calculate the strain increment between adjacent time points of the strain path curve, calculate the strain acceleration based on the strain increment, compare the strain path curve with the forming limit curve to obtain the safety margin value of each grid node, and determine the deformation risk area based on the strain acceleration and the safety margin value. The third unit is used to acquire historical stamping data containing defect types and defect characteristics, analyze the strain abrupt change characteristics of the defect formation process, establish the correspondence between strain abrupt change characteristics and defect types, obtain defect warning criteria, analyze the deformation risk area based on the defect warning criteria, and obtain the predicted defect location and defect type. The fourth unit is used to calculate the strain rate gradient at the predicted defect location. The strain rate gradient is combined with the material strain hardening index to construct a mesh optimization equation. The mesh refinement coefficient is calculated using the mesh optimization equation. The initial mesh data is then locally optimized based on the mesh refinement coefficient to obtain the optimized mesh data. The optimized mesh data is then used for stamping forming analysis, and the analysis results are output.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images