An error compensation method and system for sheet metal

Abstract

This invention discloses an error compensation method and system for sheet metal. The method includes: acquiring measured deviation data of sheet metal and theoretical part digital models; generating an adaptive parametric mesh based on local curvature characteristics; mapping the deviation data and weighted fitting to construct an error field; synthesizing a set of deviation vectors and extracting the dominant deformation mode; inputting mold deformation and material fluctuation parameters; generating a predicted deviation field through finite element simulation; obtaining an optimized predicted deviation field through residual optimization; extracting compensation vectors and inversely mapping them to the mold surface; determining the spatial distribution of compensation amount; generating adjustment instructions; and obtaining a mold error compensation and correction scheme. The method and system of this invention can achieve accurate and efficient compensation of sheet metal forming errors, meeting the high-precision and high-consistency production requirements of sheet metal processing in the field of industrial control technology.

Description

Technical Field

[0001] This invention relates to the field of industrial control system technology, and in particular to an error compensation method and system for sheet metal. Background Technology

[0002] Currently, in the field of industrial control systems, with the continuous improvement of the precision requirements for sheet metal parts and the increasing demand for complex curved surface processing in industries such as automobiles and aerospace, the error control level of sheet metal forming process, as a core manufacturing link, is directly related to the product assembly quality and production stability.

[0003] Existing sheet metal error compensation methods in the industry mainly rely on manual trial and error adjustments or experience-based corrections at single measuring points. Examples include manual mold repair after multiple trial moldings, overall compensation using the average deviation, or local adjustments that ignore the spatial coupling characteristics of errors. However, this approach is clearly insufficient in complex operating environments. Because the dominant deformation mode is not separated, compensation lacks specificity, easily leading to improvements in some areas while new deviations appear in others. Manual mold repair is inefficient, relies heavily on experience, and is difficult to adapt to the error distribution of complex curved sheet metal surfaces. Furthermore, single measuring point data cannot reflect the spatial correlation characteristics of errors, especially in large covering parts or multi-process forming scenarios, making it difficult to achieve accurate compensation across the entire area, thus limiting accuracy improvements.

[0004] In summary, existing technologies are insufficient to achieve accurate and efficient compensation for sheet metal forming errors, and cannot meet the high-precision and high-consistency production requirements of sheet metal processing in the field of industrial control technology. Summary of the Invention

[0005] This invention provides an error compensation method and system for sheet metal, so as to achieve accurate and efficient compensation of sheet metal forming errors and meet the production needs of high precision and high consistency in sheet metal processing in the field of industrial control technology.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an error compensation method for sheet metal, comprising: Obtain the theoretical part model and the measured deviation data of the surface of the sheet metal part; The local curvature characteristics of the theoretical part's digital model are obtained, and the mesh node spacing parameters are determined based on the local curvature characteristics to generate an adaptive parametric mesh. The measured deviation data is mapped onto the parameterized grid to obtain discrete deviation data. The discrete deviation data is then weighted and fitted, and the fitting result is converted into a structured error field. Based on the error field and the unit normal vector of the theoretical part model surface, the deviation value of each uniform sampling point in the error field is calculated, and the deviation value is vector synthesized with the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points. Extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode; A finite element simulation model is constructed based on the dominant deformation mode. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. The residual matrix is ​​calculated based on the predicted bias field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain the optimized predicted bias field. If it does not exceed the threshold, the predicted bias field is used as the optimized predicted bias field. Based on the optimized prediction deviation field, the compensation vector for each region is determined, and the compensation vector is converted into specific adjustment instructions to obtain the mold error compensation and correction scheme.

[0007] Secondly, the present invention provides an error compensation system for sheet metal, comprising: The data acquisition module is used to acquire the theoretical part model and the measured deviation data of the surface of the sheet metal parts; The mesh generation module is used to obtain the local curvature characteristics of the theoretical part's digital model, determine the mesh node spacing parameters based on the local curvature characteristics, and generate an adaptive parametric mesh. The error field construction module is used to map the measured deviation data onto the parameterized grid to obtain discrete deviation data, perform weighted fitting on the discrete deviation data, and convert the fitting result into a structured error field. The deviation vector module is used to calculate the deviation value of each uniform sampling point in the error field based on the error field and the unit normal vector of the surface of the theoretical part digital model, and to vector synthesize the deviation value with the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points. The deformation mode module is used to extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode. The prediction simulation module is used to construct a finite element simulation model based on the dominant deformation mode. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. The compensation and correction module is used to calculate the residual matrix based on the predicted deviation field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain the optimized predicted deviation field. If it does not exceed the threshold, the predicted deviation field is used as the optimized predicted deviation field. The scheme generation module is used to determine the compensation vector of each region based on the optimized prediction deviation field, and convert the compensation vector into specific adjustment instructions to obtain the mold error compensation correction scheme.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains the measured deviation data of sheet metal parts and the theoretical part model, performs gridding processing based on local curvature characteristics, generates an adaptive parameterized grid, establishes a spatial mapping relationship and obtains an error field by weighted fitting, breaks through the limitation that traditional single measurement point data cannot reflect spatial correlation, explores the global distribution characteristics of sheet metal errors, eliminates the one-sided interference of isolated deviations, provides high-precision basic data support for error analysis, effectively improves the construction accuracy of complex surface error fields, and solves the problem of lack of targeted adjustment in manual trial and error.

[0009] (2) This invention combines the error field and the theoretical numerical model unit normal vector to generate a set of deviation vectors, extracts orthogonal feature vectors and filters the dominant deformation mode, inputs mold deformation and material fluctuation parameters to simulate and predict the deviation field, breaks through the limitation of traditional methods that cannot separate the dominant deformation mode, accurately captures the deformation characteristics of multi-factor coupling, provides multi-dimensional basis for compensation calculation, significantly improves the pertinence and accuracy of error compensation, and makes up for the defects of limited accuracy of trial and error adjustment.

[0010] (3) This invention optimizes the prediction deviation field through residual matrix, extracts the compensation vector and projects it onto the mold surface through inverse mapping, determines the spatial distribution of compensation amount and generates adjustment instructions, solves the limitations of traditional full-domain coarse compensation, provides accurate compensation basis for mold correction, solves the problems of insufficient full-domain compensation and long trial molding cycle of complex sheet metal, takes into account compensation accuracy and production efficiency, and meets the industrial field's demand for high precision and high consistency in sheet metal processing. Attached Figure Description

[0011] Figure 1 This is a schematic flowchart of an error compensation method for sheet metal provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of an error compensation system for sheet metal provided in the second embodiment of the present invention. Detailed Implementation

[0012] 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.

[0013] Reference Figure 1The first embodiment of the present invention provides an error compensation method for sheet metal, comprising the following steps: S101, Obtain the theoretical part model and the measured deviation data of the surface of the sheet metal part; S102, Obtain the local curvature characteristics of the theoretical part digital model, determine the mesh node spacing parameters based on the local curvature characteristics, and generate an adaptive parameterized mesh; S103, the measured deviation data is mapped onto the parameterized grid to obtain discrete deviation data, the discrete deviation data is weighted and fitted, and the fitting result is converted into a structured error field; S104, Based on the error field and the unit normal vector of the theoretical part model surface, calculate the deviation value of each uniform sampling point in the error field, and perform vector synthesis of the deviation value and the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points; S105, extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode; S106, Based on the dominant deformation mode, a finite element simulation model is constructed. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. S107, Calculate the residual matrix based on the predicted deviation field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, adjust the simulation parameters and re-simulate to obtain the optimized predicted deviation field; if it does not exceed the threshold, use the predicted deviation field as the optimized predicted deviation field. S108, Based on the optimized prediction deviation field, determine the compensation vector for each region, convert the compensation vector into specific adjustment instructions, and obtain the mold error compensation and correction scheme.

[0014] In step S101, the theoretical part model and the measured deviation data of the surface of the sheet metal part are obtained, including: Retrieve the preset theoretical part model to ensure that the theoretical part model is consistent with the design dimensions of the sheet metal part; The original point cloud data of the sheet metal part surface is collected by scanning equipment, and the original point cloud data is rigidly registered with the theoretical part digital model. The normal distance from each point to the surface of the theoretical part digital model is calculated to obtain discrete deviation data. Outliers in the discrete deviation data are removed, and the proportion of missing data is calculated. If the proportion of missing data exceeds a preset data integrity threshold, supplementary data is collected; otherwise, the discrete deviation data is used as the measured deviation data.

[0015] It should be noted that, firstly, when retrieving the preset theoretical part model to ensure consistency with the design dimensions of the sheet metal part, the theoretical part model is stored in the PLM system on the local server in the common STEP format, containing complete 3D geometric information and design dimension annotations. After retrieval, key dimensions are verified using professional CAD software, including the part's length, width, key hole diameter, and center distance, etc. The verification deviation must be controlled within 0.05mm. If it exceeds this range, the model is retrieved again or updated to ensure consistency between the model and the design requirements of the actual sheet metal part in production. For example, retrieving the theoretical part model of an automotive door panel and verifying its design dimensions such as 1200mm length, 800mm width, and 10mm key hole diameter, etc., confirms that they are completely consistent with the model annotations, thus meeting the requirements for subsequent error analysis.

[0016] Subsequently, raw point cloud data of the sheet metal part surface is acquired using a scanning device. A high-precision laser line scanning device is selected, with a sampling frequency of 1000Hz to ensure that the point cloud density meets the measurement requirements of complex curved surfaces. First, feature alignment is performed based on the common key geometric features (such as the center of the positioning hole and boundary corners) of the theoretical digital model and the measured point cloud to obtain the initial pose of the point cloud relative to the digital model. On this basis, rigid registration adopts the Iterative Closest Point (ICP) algorithm, using the key feature points of the theoretical digital model as the reference, with the number of iterations set to 50 and a convergence threshold of 0.02mm. The process automatically stops when the change in the mean square error between two adjacent iterations is less than this threshold, ensuring the spatial alignment accuracy between the registered point cloud and the digital model.

[0017] Subsequently, the normal distance is calculated by solving for the shortest perpendicular distance from each point to the theoretical surface. A positive distance indicates that the actual point is outside the digital model, while a negative distance indicates that it is inside. After calculation, all deviation data are subjected to minimum-maximum normalization and mapped to the interval [-1, 1] (-0.5mm corresponds to -1, 0.5mm corresponds to 1), eliminating the influence of differences in numerical range. For example, the original point cloud data of automotive door panel sheet metal parts, containing 500,000 discrete points, after ICP registration, shows that the normal distance from a certain point to the surface of the theoretical digital model is 0.3mm, which is 0.6 after normalization. This value, together with other point data, constitutes the discrete deviation dataset.

[0018] Next, outliers in the discrete deviation data are removed using a 3-standard-deviation method. The mean and standard deviation of all deviation data are calculated first, and extreme values ​​exceeding the mean ± 3 standard deviations are removed to prevent individual outliers from interfering with overall data quality. The percentage of missing data is calculated by dividing the number of missing points by the total number of point clouds. The data integrity threshold is determined based on the tolerance boundary analysis of the surface reconstruction algorithm for discrete data missing values. Specifically, through hole repair and surface fitting tests on historical sample sets, it was found that when the data missing rate is below 5%, the surface fitting residuals based on the moving least squares (MLS) method can stably converge within a preset accuracy range (e.g., 0.01 mm). Therefore, 5% is set as the basic threshold. Based on this, dynamic hierarchical adjustments are made in conjunction with the geometric topological features of the parts: For small, simple sheet metal parts, due to their sparse geometric features, the registration algorithm is highly sensitive to point cloud coverage, so the threshold needs to be strictly tightened to 3% to ensure the calculation accuracy of the rigid transformation matrix; while for large, complex parts such as automotive body panels, considering the physical scanning blind spots in deep cavities and flanged areas, and their rich high-order curvature features with strong geometric constraints, allowing for more data compensation through interpolation algorithms, the threshold is appropriately relaxed to 8%. This threshold has been verified through multiple batches of production and can avoid over-collection while ensuring data integrity.

[0019] For example, a sheet metal part has 500,000 discrete deviation data points. After removing 1,200 outliers, the number of missing points is 15,000, which is 3% of the total. This is lower than the basic threshold of 5%, so the data is considered valid measured deviation data and no further data collection is required.

[0020] In step S102, the local curvature characteristics of the theoretical part's digital model are obtained, and the mesh node spacing parameters are determined based on the local curvature characteristics to generate an adaptive parametric mesh, including: Calculate the curvature value of each point on the surface of the theoretical part digital model. If the curvature value exceeds the preset curvature judgment threshold, set the mesh nodes according to the preset densified node spacing; if it does not exceed the threshold, set the mesh nodes according to the preset sparse node spacing. The mesh nodes are arranged according to their spatial positions to form an adaptive parametric mesh covering the surface of the theoretical part's digital model; Verify the uniformity of node distribution in the parameterized mesh. If the verification fails, adjust the node positions until the requirements are met.

[0021] It should be noted that, firstly, the curvature value of each point on the surface of the theoretical part digital model is calculated. If the curvature value exceeds the preset curvature judgment threshold, the mesh nodes are set according to the preset densified node spacing. If it does not exceed the threshold, the mesh nodes are set according to the preset sparse node spacing. The curvature value calculation adopts the average curvature algorithm based on triangular facets. The surface triangular mesh of the digital model is analyzed by CAD software, and the curvature is estimated by combining the information of the one-ring neighboring vertices of each vertex.

[0022] In this implementation case, the curvature threshold is determined by analyzing the statistical distribution characteristics of the Gaussian curvature of the surface of the theoretical part's digital model. Specifically, the curvature values ​​of all discrete sampling points on the model surface are calculated, a curvature frequency histogram is constructed, and the curvature value corresponding to the cumulative frequency distribution reaching 80% is used as the critical point to distinguish between high curvature sensitive areas and low curvature flat areas. This statistical boundary ensures that the meshing strategy covers most of the geometric features that are prone to stress concentration in sheet metal forming (such as fillets and flanges). Based on the sample statistics of typical body sheet metal parts, the mathematical expectation of this critical value is stable at 0.02mm. - ¹ is near this value, therefore it is set as the base threshold. For complex curved sheet metal (such as automotive body panels), due to their rich geometric details, to balance computational efficiency and feature capture capability, the threshold can be adjusted upwards to 0.03mm based on the 90th percentile value. - ¹; For simple planar sheet metal, the threshold is lowered to 0.01mm based on the 70th percentile value. - ¹. This threshold setting has been verified through multiple batches of digital model testing to achieve adaptive mesh density allocation. **The spacing between dense nodes is set to 3mm, and the spacing between sparse nodes is set to 15mm to ensure dense meshing in high-curvature areas to capture details, and sparse meshing in low-curvature areas to reduce computational load. For example, the curvature value of the rounded transition area in the automotive door panel digital model is 0.025mm. - ¹, If the value exceeds the basic threshold, set nodes at 3mm intervals; the curvature value of the door panel plane area is 0.008mm. - ¹, Set nodes at 15mm intervals to achieve mesh adaptation.

[0023] Subsequently, the mesh nodes are arranged according to their spatial positions to form an adaptive parametric mesh covering the surface of the theoretical part's digital model. The arrangement method employs surface parametric coordinate mapping, using the UV coordinates of the digital model as a reference. Mesh nodes are evenly distributed across the UV parameter space of the surface and then mapped back to a three-dimensional Cartesian coordinate system, ensuring that the nodes are uniformly distributed along the surface without overlap. Before arrangement, the surface of the digital model is divided into partitions. Each partition is arranged independently before being stitched together to avoid misalignment of nodes across partitions. For example, the digital model of an automotive door panel is divided into three partitions: a planar area, a rounded corner area, and an edge area. Nodes are arranged separately and then stitched together to form a parametric mesh that completely covers the door panel surface, with approximately 20,000 nodes in total. This ensures complete coverage while controlling the mesh size.

[0024] Finally, the uniformity of node distribution in the parameterized mesh is verified. If the verification fails, the node positions are adjusted until the requirements are met. The coefficient of variation (COP) is used to measure the non-uniformity. Specifically, the system iterates through all topologically adjacent node pairs in the mesh, calculates their Euclidean distances to construct a spacing dataset, and then calculates the standard deviation and arithmetic mean of this dataset. The ratio of the standard deviation to the mean is defined as the non-uniformity. This ratio eliminates the influence of the average mesh size (dimensions) and simply reflects the dispersion of node distribution; a larger value indicates more significant local differences in density.

[0025] It should be noted that the uniformity threshold is set based on the statistical results of mesh sensitivity analysis of finite element simulation accuracy. Specifically, multi-scale mesh generation tests were conducted on typical historical sheet metal parts, and a convergence curve of "mesh quality index - simulation prediction error" was plotted. Statistical analysis revealed that when the non-uniformity (CV value) is below 0.3, the marginal contribution of further improving mesh uniformity to computational accuracy decreases significantly, while the computation time for mesh generation increases exponentially. Therefore, 0.3 was set as the basic threshold (inflection point value) balancing accuracy and efficiency. Based on this, scenario-based dynamic adjustments were made: for high-precision error analysis scenarios (such as Class A surface compensation), the top 20% quantile (i.e., 0.2) was selected as the threshold based on the statistical distribution of historical high-precision case libraries to suppress oscillations during numerical fitting; for fast simulation scenarios, the top 80% quantile (i.e., 0.4) was selected as the threshold to prioritize iteration speed. If the non-uniformity exceeds the threshold during verification, the Laplace smoothing algorithm is used to move nodes from dense regions to sparse regions, or to perform topological reconstruction (split or collapse) on overly dense / sparse regions, and then the non-uniformity is recalculated. For example, if the initial non-uniformity of a mesh is 0.35, exceeding the basic threshold of 0.3, after smoothing, the non-uniformity is reduced to 0.28, which meets the requirements.

[0026] In step S103, the measured deviation data is mapped onto the parameterized grid to obtain discrete deviation data. The discrete deviation data is then weighted and fitted, and the fitting result is converted into a structured error field, including: The measured deviation data is mapped onto the parameterized grid through spatial interpolation to obtain discrete deviation data; Based on the spatial distance between nodes, the discrete deviation data is weighted and fitted to obtain the fitting result; The error of the fitting result is calculated. If the error exceeds a preset fitting accuracy threshold, the weight coefficients are adjusted and the fitting is refitted. If the error does not exceed the threshold, the continuous deviation distribution of the fitting result is obtained as the error field.

[0027] It should be noted that, firstly, when the measured deviation data is transmitted to each node of the parameterized grid through spatial interpolation, and the spatial mapping relationship between the two is established, the inverse distance weighted interpolation (IDW) method is selected. This method can not only accurately transmit the deviation data, but also allocate weights according to the spatial distance between nodes, making the influence of the nearest measured points on the grid nodes more significant.

[0028] During interpolation, each grid node is used as the center, and the search radius is set to twice the average spacing between grid nodes to ensure that a sufficient number of measured points participate in the calculation. The weight is the reciprocal of the distance, with closer points having larger weights. Then, all weights involved in the calculation are normalized to ensure that the sum of the weights is 1. The measured deviation data needs to be min-max normalized first, mapped to the interval [-1, 1] to avoid the influence of numerical range differences on the interpolation results.

[0029] For example, the distances of three measured deviation points around a certain grid node are 2mm, 4mm, and 6mm, respectively. The calculated weights are 0.6, 0.3, and 0.1, respectively. After weighted summation, the mapping deviation value of the node is 0.22mm, and the spatial mapping is successfully established.

[0030] Subsequently, a weighted least squares method based on a moving window is used for local surface reconstruction. First, a 3×3 grid region containing itself and its surrounding neighboring nodes is dynamically locked as an independent fitting unit, centered on the current node to be calculated. This window size ensures a sufficient number of sampling points (9 points) to solve the quadratic surface equation while avoiding the introduction of irrelevant noise from cross-regional data. Next, a bivariate quadratic polynomial function with six undetermined coefficients is constructed as the fitting model. Compared to a linear plane, this model has the ability to describe complex hypercurvature deformation and can accurately capture the nonlinear springback and torsion characteristics of sheet metal surfaces. Then, a weighted calculation is performed to calculate the Euclidean distance from each node in the fitting unit to the fitting center. The weights are set to the reciprocal of the distance and normalized (ensuring the sum of the weights is 1), so that nodes closer to the center have stronger control over the shape of the fitted surface. Finally, an objective function that minimizes the weighted sum of squared residuals is established. The coefficients of the above polynomial are determined by solving the linear equations through partial derivatives, thereby obtaining a continuous and smooth deviation distribution surface in this local region.

[0031] For example, the mapping deviation values ​​of 9 nodes in a certain partition are between 0.15 and 0.25 mm. After performing a quadratic polynomial fitting with distance weights, a continuous deviation curve for the region is obtained, which not only preserves the deviation characteristics of each node, but also achieves a smooth transition and avoids abrupt changes.

[0032] The error of the fitting result is calculated. If the error exceeds a preset fitting accuracy threshold, the weighting coefficients are adjusted and the fit is re-fitted. If the error does not exceed the threshold, the continuous deviation distribution of the fitting result is obtained as the error field. It is important to note that the fitting accuracy threshold is set based on the statistical data of sheet metal error fitting over the past year. The fitting residuals of all valid fitting cases are statistically analyzed. The basic threshold is set to 0.02 mm. For complex curved sheet metal surfaces, due to larger deviation fluctuations, the threshold can be increased to 0.03 mm. For simple planar sheet metal surfaces, where the deviation is stable, the threshold can be decreased to 0.01 mm. This threshold has been verified through multiple batches and can ensure both fitting accuracy and computational efficiency. The fitting error is calculated as the mean of the absolute differences between the fitted value and the mapped value for each node. If it exceeds a threshold, the weight of nearby nodes can be increased for refitting. Specifically, first, the current calculation error is compared with the set threshold to determine the deviation factor. Then, a local gain coefficient positively correlated with this deviation factor is constructed, i.e., the greater the error exceeds the threshold, the larger the gain coefficient becomes. Subsequently, several (e.g., 3) measured reference points closest to the calculation center are identified, and their original weights are multiplied by this local gain coefficient to enhance the tight constraint of the nearest neighbor points on the fitting result, forcing the fitted surface to deform locally to more closely resemble the measured value. Finally, the weights of the remaining distant points are kept unchanged, and all the adjusted new weights are normalized to ensure that the sum of the weights is conserved to 1. The surface is then refitted using the updated weight parameters until the error converges to the threshold range.

[0033] For example, if the fitting error for a certain region is 0.025 mm and the base threshold is 0.02 mm, it indicates a significant deviation in error. The system automatically generates a gain coefficient greater than 1 (e.g., 1.05) based on the deviation ratio, amplifying the weights of the three nearest measured points (e.g., increasing them from 0.6 to 0.63). After normalization and balancing, the dominant role of the nearest neighbors is significantly enhanced, and the error after refitting decreases to 0.018 mm, meeting the accuracy requirements.

[0034] In step S104, based on the error field and the unit normal vector of the theoretical part's digital model surface, the deviation value of each uniform sampling point in the error field is calculated. The deviation value is then vector-synthesized with the unit normal vector of the corresponding sampling point to obtain a set of deviation vectors on the uniform sampling points, including: Uniform sampling points are selected on the surface of the theoretical part's digital model according to a preset sampling density threshold; Calculate the unit normal vector of each of the uniform sampling points on the surface of the theoretical part's digital model; The deviation value of each uniform sampling point in the error field is extracted, and the deviation value is vector synthesized with the corresponding unit normal vector to obtain the deviation vector of each uniform sampling point. The vectors are then summarized to form a set of deviation vectors on the uniform sampling points.

[0035] It should be noted that, firstly, when selecting uniform sampling points on the surface of the theoretical part's digital model according to the preset sampling density threshold, the sampling density threshold is set based on the spatial frequency distribution characteristics of the error field and experimental statistics on the fidelity of surface reconstruction. Specifically, through analysis of the springback error spectrum of typical historical sheet metal parts, it was found that when the sampling density reaches 5 points per square centimeter, the error surface reconstructed from discrete points can cover more than 95% of the deformation features (including macroscopic twisting and microscopic wavy lines), meeting the requirements of conventional industrial-grade precision. Therefore, this is set as the basic threshold. For complex curved sheet metal surfaces such as automotive body panels, which have high-frequency local undulations and stress concentration areas, the threshold can be increased to 8 points per square centimeter to prevent feature loss due to undersampling. For simple planar sheet metal, whose deformation mode is mainly low-frequency overall bending, this threshold has been tested with multiple batches of digital models and can control the amount of computation while ensuring sampling accuracy.

[0036] It is worth noting that the selection process employs a UV parameter space uniform sampling method based on isoparametric lines. This method first extracts the topological parameterized domain of the digital model and normalizes it to the [0,1] unit plane. Then, based on the determined total number of samples, it calculates the equidistant walk distances in the U and V directions to generate two-dimensional parameter mesh nodes. Finally, through the inverse mapping calculation of the surface equation, it maps back to three-dimensional space to obtain the sampling points, ensuring that the points are uniformly distributed on the surface without overlap. For example, the theoretical surface area of ​​a car door panel digital model is 2000 square centimeters. At 5 sampling points per square centimeter, a total of 10,000 uniform sampling points are selected to uniformly cover the planar area, rounded corner area, and edge area of ​​the door panel.

[0037] Subsequently, when calculating the unit normal vector of each sampling point on the surface of the theoretical part's digital model, the surface normal vector calculation function of the CAD software was used, based on the triangular facet data of the digital model surface. For each sampling point, the normal directions of its five neighboring triangular facets were selected, and the initial normal vector was obtained by vector averaging. Then, the vector magnitude was normalized to 1 to ensure it was a unit vector, and its direction consistently pointed outwards from the digital model (avoiding confusion between positive and negative values). After calculation, the vector was verified. If the angle between the vector of a sampling point and the neighboring vectors exceeded 30°, it was recalculated to ensure directional continuity. For example, for a sampling point in the rounded corner area of ​​a door panel, the initial vector obtained by averaging the normal directions of the neighboring triangular facets was (0.2, 0.3, 0.93). After normalization, the magnitude was 1, accurately pointing outwards from the digital model and consistent with the vector directions of the surrounding sampling points.

[0038] Next, the deviation value of each sampling point in the error field is extracted. This deviation value is obtained from the error field through bilinear interpolation. The error field is continuously distributed. During interpolation, the deviation values ​​of the four grid nodes surrounding the sampling point are weighted and calculated, with the weight allocation following the geometric rule of diagonal area ratio. Specifically, the sampling point falls within a rectangular cell enclosed by four grid nodes. This rectangle is divided into four sub-rectangles, and the weight of each grid node is numerically equal to the ratio of the area of ​​its sub-rectangle along its diagonal direction to the total area of ​​the grid cell. This rule is mathematically equivalent to performing linear interpolation in two coordinate directions, ensuring that the closer a sampling point is to a node, the larger the diagonal area corresponding to that node, thus obtaining a larger weight contribution. Furthermore, the sum of the weights of the four nodes is strictly 1. The deviation value has been normalized to [-1, 1]. During vector synthesis, the deviation value is multiplied by each component of the unit normal vector to obtain a directional deviation vector (which has both magnitude and direction, reflecting the spatial orientation of the actual deviation). Finally, all deviation vectors are summarized in the order of the sampling points to form a set, which is the deviation vector set.

[0039] For example, the interpolation deviation value of a certain sampling point is 0.6 (corresponding to an actual deviation of 0.3mm), the unit normal vector is (0.2, 0.3, 0.93), and the synthesized deviation vector is (0.12, 0.18, 0.558). By summarizing such vectors from 10,000 sampling points, a complete set of deviation vectors is formed, providing basic data for the subsequent extraction of the dominant deformation mode.

[0040] In step S105, mutually orthogonal feature vectors are extracted from the set of deviation vectors, and the main contributing components are extracted from the feature vectors as the dominant deformation mode, including: Calculate the contribution rate corresponding to each feature vector, sort them from largest to smallest, and sum them up to obtain the cumulative contribution rate; If the cumulative contribution rate exceeds the preset contribution judgment threshold, the corresponding feature vector is extracted; if it does not exceed the threshold, subsequent feature vectors are extracted until the requirements are met, and the extracted feature vectors are used as the dominant deformation mode.

[0041] It should be noted that, firstly, when extracting mutually orthogonal eigenvectors from the set of deviation vectors, Principal Component Analysis (PCA) is used to construct a sample matrix with sampling points as samples and deviation components as dimensions. The matrix is ​​then subjected to mean-neutralization to eliminate the influence of the average displacement of the data. Next, the covariance matrix of the sample matrix is ​​calculated to quantify the linear correlation between deviation changes at different spatial locations. Subsequently, eigenvalue decomposition is performed on the covariance matrix to obtain a series of eigenvalues ​​and their corresponding eigenvectors. Finally, the eigenvectors are sorted in descending order according to the magnitude of the eigenvalues, with vectors having larger eigenvalues ​​representing the dominant deformation patterns with higher contributions.

[0042] In this implementation, the deviation vector set needs to be first subjected to min-max normalization, mapping it to the [-1,1] interval to ensure data scale uniformity. By constructing the covariance matrix of the deviation vector set, eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvalue sequence and the corresponding eigenvector set. The direction of the eigenvectors represents the main direction of the deformation pattern, and the magnitude of the eigenvalues ​​reflects the contribution of that pattern to the overall deviation. For example, after PCA processing, the deviation vector set of a car door panel yields 10 eigenvectors; the first three eigenvalues ​​are significantly larger than the others, corresponding to the main deformation trends.

[0043] Next, the contribution rate corresponding to each eigenvector is calculated. After sorting them from largest to smallest, the cumulative contribution rate is obtained by summing them. The contribution rate is calculated by dividing a single eigenvalue by the sum of all eigenvalues, and the result is expressed as a percentage. During sorting, the eigenvectors are sorted from largest to smallest according to their eigenvalues, and then the contribution rates are summed sequentially to obtain the cumulative contribution rate. This value reflects the explanatory power of the first k eigenvectors for the overall bias.

[0044] For example, the contribution rates of the feature vectors of a certain sheet metal part are 60%, 30%, 5%, 3%... respectively. The cumulative contribution rate of the first two is 90%, and the contribution rate of the first three is 95%, clearly showing the contribution ratio of the main deformation modes.

[0045] If the cumulative contribution rate exceeds the preset contribution threshold, the corresponding feature vector is extracted. If it does not exceed the threshold, subsequent feature vectors are extracted until the requirement is met. The contribution threshold is set based on the analysis of sheet metal deformation data over the past year. Statistical analysis of feature decomposition results for a large number of different types of sheet metal revealed that when the cumulative contribution rate reaches 85%, the first k feature vectors can cover the main deformation information. Therefore, the basic threshold is set to 85%. For complex curved sheet metal, such as automotive body panels, the deformation patterns are more complex, and the threshold can be increased to 90% to ensure coverage of all key deformations. For simple planar sheet metal with a single deformation pattern, the threshold can be decreased to 80% to simplify calculations. This threshold has been verified through multiple batches and can control the number of dominant deformation modes while ensuring accuracy. The extracted feature vectors are used as the dominant deformation modes.

[0046] For example, the cumulative contribution rate of the first two feature vectors of a complex curved sheet metal is 88%, which does not reach the 90% threshold. After adding the third feature vector, the cumulative contribution rate is 96%, which meets the requirements. Finally, these three feature vectors are extracted as the dominant deformation modes, which correspond to the three main deformations: central bulge, edge warping, and local depression.

[0047] In step S106, a finite element simulation model is constructed based on the dominant deformation modes. Preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode, including: The dominant deformation mode is mapped to the finite element mesh node system, and the simulation boundary conditions are established by combining the preset mold deformation parameters and preset material fluctuation parameters to construct the finite element simulation model. The nodal displacement response under each dominant deformation mode is calculated using the finite element simulation model. The normal distance is obtained by analyzing the displacement response, and the local prediction deviation field corresponding to each mode is generated. By superimposing all the local prediction bias fields, the overall prediction bias field is obtained.

[0048] It should be noted that the elastic deformation of the mold is obtained through stress testing under the working conditions of the mold. Strain gauges are used to collect strain data at key parts of the mold, which is then converted into elastic deformation. The value is usually in the range of 0-0.5mm. Local wear is calculated by laser scanning of the mold surface to determine the normal distance between the actual and theoretical surfaces. The value is usually in the range of 0-0.3mm. Material thickness fluctuation is obtained from the inspection data of sheet metal raw materials, statistically analyzing the thickness difference of materials in the same batch. The value is usually in the range of 0-0.2mm. Strength fluctuation is obtained by tensile testing to determine the fluctuation range of the material's yield strength, directly using its measured physical values. For example, if the measured elastic deformation of a mold is 0.2mm and the material thickness fluctuation is 0.1mm, these measured physical parameters will be directly used as boundary conditions input into the model to fully reflect the actual physical state of the mold and material.

[0049] Subsequently, the dominant deformation modes, mold deformation parameters, and material fluctuation parameters are input into the simulation model. The deviation distribution corresponding to each dominant deformation mode is calculated, and the deviation distributions of each mode are superimposed to obtain the overall predicted deviation field. The simulation model uses the ABAQUS finite element model, with tetrahedral elements and an element size of 2 mm. The mold part is assigned steel properties (Young's modulus 210 GPa, Poisson's ratio 0.3), and the sheet metal part is assigned values ​​according to the actual material properties. The input mold deformation and material fluctuation parameters are all actual physical values. The model calculates the nodal displacement response under the current physical boundary conditions based on the principle of elastoplastic mechanics, thereby determining the deviation distribution of each dominant deformation mode. During superposition, weights are assigned according to the contribution rate of the dominant deformation modes; the higher the contribution rate, the greater the weight. The weighted summation yields the overall predicted deviation field.

[0050] Finally, the rationality of the predicted deviation field is verified. If the irrationality exceeds the preset reasonable simulation threshold, the parameters are adjusted and the simulation is repeated. The reasonable simulation threshold is set based on the comparative statistics of sheet metal simulation and measured data over the past year. The degree of agreement between the predicted deviation field and the measured error field is statistically analyzed in all valid simulation cases. The basic threshold is set at 0.15mm, meaning that the proportion of areas in the predicted deviation field exceeding 0.15mm and opposite to the measured deviation direction does not exceed 10%. For complex curved sheet metal surfaces, due to their complex deformation, this threshold can be increased to 0.2mm; for simple planar sheet metal surfaces, it can be decreased to 0.1mm. This threshold has been verified through multiple batches and can effectively filter out reasonable prediction results. The irrationality is calculated as the proportion of the area in the predicted deviation field that exceeds the reasonable deviation range (-0.15mm to 0.15mm) and does not match the measured deviation trend. If it exceeds 10%, it is considered unreasonable.

[0051] For example, in a certain prediction deviation field, 15% of the area has a deviation value of -0.2mm (opposite to the measured positive deviation trend). The unreasonableness of 15% exceeds 10%. The mold elastic deformation parameter is adjusted (from 0.2mm to 0.25mm) and the simulation is repeated until the unreasonableness meets the requirements.

[0052] In step S107, a residual matrix is ​​calculated based on the predicted bias field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain an optimized predicted bias field. If the norm does not exceed the threshold, the predicted bias field is used as the optimized predicted bias field, including: Spatial registration is performed between the predicted bias field and the error field, and interpolation is performed node by node to generate an initial residual matrix; Calculate the norm of the residual matrix. If the norm exceeds a preset residual judgment threshold, identify the simulation parameter that has the greatest impact on the residual and adjust it to obtain the adjusted parameter. The adjusted parameters are input into the simulation model to regenerate the prediction bias field. The residual matrix is ​​calculated repeatedly until the norm meets the requirements, thus obtaining the optimized prediction bias field.

[0053] It should be noted that, firstly, when spatially registering the predicted deviation field and the error field, to overcome the inconsistency in node distribution between the finite element simulation mesh and the measured parametric mesh, a rigid body transformation matrix based on key feature points (such as the center of the positioning hole) is first used to unify them to the same measurement coordinate system. Then, using shape function interpolation technology, the unstructured node deviation data output from the finite element simulation is resampled and mapped to the corresponding mesh nodes of the error field, ensuring strict spatial alignment between the two fields. Based on this, the residual matrix is ​​generated by subtracting the corresponding node deviation values ​​of the predicted deviation field and the error field point by point, with each element representing the difference in deviation for the same node in the two fields. The norm is calculated using the Frobenius norm, which is the square root of the sum of the squares of all elements, comprehensively reflecting the overall residual magnitude.

[0054] It should be noted that the residual judgment threshold is set based on the statistical distribution characteristics of the residual norm of historical simulation-measured calibration data. Specifically, the system collects calibration residual data from a large number of historical cases, constructs a cumulative probability distribution function (CDF) of the residual norm, and selects the norm value when the cumulative probability reaches 85% as the basic convergence boundary for judging the simulation accuracy. Statistical data shows that this boundary value is stable around 0.08mm, so it is set as the basic threshold. Based on this, dynamic adjustments are made according to the complexity of the parts: complex curved sheet metal has significant nonlinear deformation, making simulation convergence difficult, and can be adjusted upward to 0.1mm based on the 90th percentile; simple planar sheet metal is sensitive to flatness requirements, and can be adjusted downward to 0.06mm based on the 75th percentile. This threshold has been verified through multiple batches and can accurately judge whether the simulation results meet the accuracy requirements. The Morris screening method in global sensitivity analysis is used to identify the simulation parameters with the greatest impact. The partial derivatives of each parameter (mold elastic deformation, local wear, material thickness fluctuation, strength fluctuation) with respect to the residual norm are calculated, and the parameter with the largest absolute value of the partial derivative is the key parameter. During adjustment, a rigid fixed step size is abandoned in favor of adaptive correction based on sensitivity gradient. First, the adjustment direction is determined by adjusting along the negative gradient direction according to the sign of the sensitivity coefficient. If increasing the parameter leads to an increase in residual, the parameter is decreased; conversely, the decrease is not. Second, the adjustment step size is calculated. This step size is proportional to the extent to which the current residual exceeds the threshold and inversely proportional to the magnitude of the sensitivity coefficient. A preset damping factor (e.g., 0.6) is introduced to prevent iterative oscillation. Finally, a physical boundary truncation check is performed to ensure that the adjusted parameter remains within the valid range of the physical quantity (e.g., the elastic deformation of the mold must be between 0-0.5 mm). If it exceeds this range, it is forcibly clamped to the boundary value to ensure that the parameter remains within the reasonable engineering range.

[0055] For example, the residual norm of a certain prediction bias field and error field is 0.09 mm, exceeding the basic threshold of 0.08 mm. Sensitivity analysis revealed that the elastic deformation of the mold is a key parameter, and its sensitivity is negative (i.e., increasing the deformation can reduce the residual). Based on the out-of-tolerance ratio, a suggested step size of 0.02 mm was set. The original parameter of 0.2 mm was adjusted positively to 0.22 mm, and after confirming that this value did not exceed the physical upper limit of 0.5 mm, it was input into the model for the next round of simulation.

[0056] For example, the residual norm of a certain prediction deviation field and error field is 0.09 mm, which exceeds the basic threshold of 0.08 mm. Sensitivity analysis revealed that the elastic deformation of the mold has the greatest impact. The original parameter of 0.2 mm was adjusted to 0.23 mm by +15%, resulting in the adjusted parameter.

[0057] Subsequently, the adjusted parameters are input into the simulation model to regenerate the prediction bias field. The residual matrix is ​​calculated repeatedly until the norm meets the requirements. When the optimized prediction bias field is obtained, the prediction bias field is regenerated using the previous ABAQUS finite element model, with consistent mesh generation and boundary conditions. Only the adjusted simulation parameters are replaced, ensuring the consistency of the simulation conditions. During the repetition process, the residual matrix and norm are recalculated after each adjustment until the norm falls below the residual judgment threshold. The maximum number of repetitions is set to 5. If the requirements are still not met after 5 repetitions, the residual judgment threshold can be appropriately relaxed (maximum not exceeding 0.02 mm) to avoid infinite loops.

[0058] For example, the adjusted mold elastic deformation amount of 0.23mm is input into the model, the predicted deviation field is regenerated, and the residual norm is calculated to be 0.07mm, which is lower than the basic threshold of 0.08mm and meets the accuracy requirements. This predicted deviation field is the optimized predicted deviation field, which can not only accurately reflect the actual error distribution of sheet metal, but also provide a reliable basis for subsequent compensation calculations.

[0059] In step S108, compensation vectors for each region are determined based on the optimized prediction deviation field, and the compensation vectors are converted into specific adjustment instructions to obtain a mold error compensation and correction scheme, including: Extract the deformation values ​​of each region in the optimized prediction deviation field, and calculate the reverse displacement component as a compensation vector based on the deformation values; The compensation vector is projected onto the mold surface using the inverse mapping method to obtain the discrete compensation amount for each region of the mold. Spatial interpolation is performed on the discrete compensation quantities to form a continuous spatial distribution of compensation quantities; Based on the spatial distribution of the compensation amount, an adjustment instruction for mold processing is generated, and the adjustment instruction is input into the CNC system to obtain a mold error compensation and correction scheme.

[0060] It should be noted that, firstly, the deformation values ​​of each region in the optimized prediction deviation field are extracted. When calculating the reverse displacement component as the compensation vector based on the deformation values, the deformation values ​​are extracted from the discrete grid nodes of the optimized prediction deviation field. The deformation values ​​of each node reflect the actual deviation magnitude and direction of the sheet metal in that region. After extraction, minimum-maximum normalization is required, and the values ​​are mapped to the interval [-1, 1] (-0.5mm corresponds to -1, and 0.5mm corresponds to 1).

[0061] The reverse displacement component is calculated by inverting the deformation value. If the deformation value is positive, the compensation vector is negative, indicating that the mold surface needs to be adjusted in the opposite direction, and vice versa. This ensures that the compensation vector can offset the actual deviation of the sheet metal. For example, if the deformation value of a certain area is 0.2mm (0.4 after normalization), the calculated reverse displacement component is -0.2mm (-0.4 after normalization). This vector is the compensation vector for that area, clarifying the direction and magnitude of mold adjustment.

[0062] It is worth noting that the inverse mapping method uses Thin Plate Spline Interpolation (TPS) based on the principle of energy minimization. This method can not only accurately establish the spatial correspondence between the sheet metal and the mold surface, but also ensure the global smoothness of the mapping transformation by minimizing the bending energy function, thus effectively adapting to the nonlinear mapping requirements of complex curved surfaces. Before projection, the geometric correspondence between the sheet metal and the mold needs to be established. Using the theoretical part's digital model as an intermediary and a topological reference, the projections of the compensation vector nodes-surfaces on the sheet metal are mapped to the corresponding nodes on the mold surface according to geometric constraints. The discrete compensation amounts obtained after mapping need to be physically constrained and denormalized to the actual physical range (-0.5mm to 0.5mm) to ensure that the values ​​meet the mold processing accuracy requirements. For example, if the compensation vector of a certain area of ​​the sheet metal is -0.2mm, after TPS inverse mapping, the discrete compensation amount of the corresponding mold surface node is 0.2mm, indicating that the mold node needs to be adjusted outward by 0.2mm along the normal direction.

[0063] Next, spatial interpolation is performed on the discrete compensation quantities to form a continuous spatial distribution of compensation quantities. The spatial interpolation uses radial basis function (RBF) interpolation, and the kernel function is selected as a Gaussian function to ensure that the distribution of compensation quantities after interpolation is continuous and smooth without abrupt changes.

[0064] It should be noted that the interpolation accuracy threshold is set based on the statistical distribution characteristics of interpolation residuals in historical mold processing data. Specifically, by analyzing a large number of mold surface reconstruction cases over the past year, a cumulative probability distribution function (CDF) for the interpolation residuals is constructed, and the residual value when the cumulative probability reaches 95% is defined as the acceptable accuracy boundary. Statistical data shows that for conventional molds, this boundary value is stable around 0.01mm, so it is set as the basic threshold. For complex curved sheet metal surfaces, due to their complex geometry, the allowable residual fluctuation range is wider (based on the 98th quantile), and can be adjusted upwards to 0.015mm; for simple planar sheet metal surfaces, the flatness requirements are extremely high (based on the 90th quantile), and can be adjusted downwards to 0.008mm. This threshold has been verified through multiple batches and can ensure the smoothness of mold surface adjustments. The interpolation accuracy is calculated as the mean of the absolute differences between the interpolated compensation amount and the discrete compensation amount. If it exceeds the threshold, the interpolation method can be adjusted to inverse distance weighted interpolation (IDW), and the continuous compensation amount distribution can be recalculated. For example, the accuracy of RBF interpolation in a certain area is 0.012mm, which exceeds the basic threshold of 0.01mm. After adjusting to IDW interpolation, the accuracy drops to 0.009mm, which meets the requirements and forms a continuous spatial distribution of compensation.

[0065] Finally, adjustment instructions for mold processing are generated based on the spatial distribution of the compensation amount. These instructions are then input into the CNC system to obtain the mold error compensation and correction scheme. The tool position coordinates are determined based on the compensation amount. A vector superposition algorithm is used to obtain the coordinates and normal vectors of the original machining path points on the mold surface. The new coordinates are the original coordinates plus the product of the unit normal vector and the compensation amount. Path planning uses adaptive G-code generated by the equal residual height method. The instruction format includes key parameters such as tool position coordinates, spindle speed, feed rate, and depth of cut. The target coordinates are obtained by superimposing the compensation amount along the normal direction of the mold surface. The spindle speed can be set to 5000 rpm, the feed rate to 100 mm / min, and the depth of cut is set in segments according to the compensation amount (0.05 mm when the compensation amount is ≤0.1 mm, and 0.1 mm when it is >0.1 mm) to avoid excessive single-cutting amounts affecting the mold surface quality. After the generated adjustment instructions are input into the CNC machining system, the system drives the tool to perform precise machining of the mold according to the instructions, ultimately obtaining the mold error compensation and correction scheme.

[0066] For example, based on the spatial distribution of the compensation amount, the adjustment command for a certain mold is generated as "G01 X100.2 Y50.3 Z-5.0F100 S5000", which means that the tool position point is adjusted to 100.2mm along the X-axis, 50.3mm along the Y-axis, and -5.0mm along the Z-axis, with a feed rate of 100mm / min and a spindle speed of 5000rpm. After executing this command, the mold compensation machining in the corresponding area can be completed.

[0067] In summary, this invention discloses an error compensation method and system for sheet metal, comprising: acquiring measured deviation data of sheet metal and theoretical part digital models; generating an adaptive parametric mesh according to local curvature to construct an error field; synthesizing a set of deviation vectors and extracting the dominant deformation mode; inputting mold deformation and material fluctuation parameters to simulate and predict the deviation field, and obtaining an optimized predicted deviation field through residual optimization; extracting compensation vectors and inversely mapping them to the mold surface to determine the spatial distribution of compensation amount, generating adjustment instructions, and obtaining a mold error compensation and correction scheme. This method can achieve accurate and efficient compensation of sheet metal forming errors, meeting the high-precision and high-consistency production requirements of sheet metal processing in the field of industrial control technology.

[0068] Reference Figure 2 The second embodiment of the present invention provides an error compensation system for sheet metal, comprising: The data acquisition module is used to acquire the theoretical part model and the measured deviation data of the surface of the sheet metal parts; The mesh generation module is used to obtain the local curvature characteristics of the theoretical part's digital model, determine the mesh node spacing parameters based on the local curvature characteristics, and generate an adaptive parametric mesh. The error field construction module is used to map the measured deviation data onto the parameterized grid to obtain discrete deviation data, perform weighted fitting on the discrete deviation data, and convert the fitting result into a structured error field. The deviation vector module is used to calculate the deviation value of each uniform sampling point in the error field based on the error field and the unit normal vector of the surface of the theoretical part digital model, and to vector synthesize the deviation value with the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points. The deformation mode module is used to extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode. The prediction simulation module is used to construct a finite element simulation model based on the dominant deformation mode. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. The compensation and correction module is used to calculate the residual matrix based on the predicted deviation field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain the optimized predicted deviation field. If it does not exceed the threshold, the predicted deviation field is used as the optimized predicted deviation field. The scheme generation module is used to determine the compensation vector of each region based on the optimized prediction deviation field, and convert the compensation vector into specific adjustment instructions to obtain the mold error compensation correction scheme.

[0069] It should be noted that the error compensation system for sheet metal provided in this embodiment of the invention is used to execute all the process steps of the error compensation method for sheet metal in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0070] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0071] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An error compensation method for sheet metal, characterized in that, include: Obtain the theoretical part model and the measured deviation data of the surface of the sheet metal part; The local curvature characteristics of the theoretical part's digital model are obtained, and the mesh node spacing parameters are determined based on the local curvature characteristics to generate an adaptive parametric mesh. The measured deviation data is mapped onto the parameterized grid to obtain discrete deviation data. The discrete deviation data is then weighted and fitted, and the fitting result is converted into a structured error field. Based on the error field and the unit normal vector of the theoretical part model surface, the deviation value of each uniform sampling point in the error field is calculated, and the deviation value is vector synthesized with the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points. Extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode; A finite element simulation model is constructed based on the dominant deformation mode. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. The residual matrix is ​​calculated based on the predicted bias field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain the optimized predicted bias field. If it does not exceed the threshold, the predicted bias field is used as the optimized predicted bias field. Based on the optimized prediction deviation field, the compensation vector for each region is determined, and the compensation vector is converted into specific adjustment instructions to obtain the mold error compensation and correction scheme.

2. The error compensation method for sheet metal according to claim 1, characterized in that, The acquisition of theoretical part numerical models and measured surface deviation data of sheet metal parts includes: Retrieve the preset theoretical part model to ensure that the theoretical part model is consistent with the design dimensions of the sheet metal part; The original point cloud data of the sheet metal part surface is collected by scanning equipment, and the original point cloud data is rigidly registered with the theoretical part digital model. The normal distance from each point to the surface of the theoretical part digital model is calculated to obtain discrete deviation data. Outliers in the discrete deviation data are removed, and the proportion of missing data is calculated. If the proportion of missing data exceeds a preset data integrity threshold, supplementary data is collected; otherwise, the discrete deviation data is used as the measured deviation data.

3. The error compensation method for sheet metal according to claim 1, characterized in that, The process of obtaining the local curvature features of the theoretical part's digital model, determining the mesh node spacing parameters based on the local curvature features, and generating an adaptive parametric mesh includes: Calculate the curvature value of each point on the surface of the theoretical part digital model. If the curvature value exceeds the preset curvature judgment threshold, set the mesh nodes according to the preset densified node spacing; if it does not exceed the threshold, set the mesh nodes according to the preset sparse node spacing. The mesh nodes are arranged according to their spatial positions to form an adaptive parametric mesh covering the surface of the theoretical part's digital model; Verify the uniformity of node distribution in the parameterized mesh. If the verification fails, adjust the node positions until the requirements are met.

4. The error compensation method for sheet metal according to claim 1, characterized in that, The step of mapping the measured deviation data onto the parameterized grid to obtain discrete deviation data, performing weighted fitting on the discrete deviation data, and converting the fitting result into a structured error field includes: The measured deviation data is mapped onto the parameterized grid through spatial interpolation to obtain discrete deviation data; Based on the spatial distance between nodes, the discrete deviation data is weighted and fitted to obtain the fitting result; The error of the fitting result is calculated. If the error exceeds a preset fitting accuracy threshold, the weight coefficients are adjusted and the fitting is refitted. If the error does not exceed the threshold, the continuous deviation distribution of the fitting result is obtained as the error field.

5. The error compensation method for sheet metal according to claim 1, characterized in that, The step involves calculating the deviation value of each uniform sampling point in the error field based on the error field and the unit normal vector of the theoretical part's digital model surface, and then vector synthesizing the deviation value with the unit normal vector of the corresponding sampling point to obtain a set of deviation vectors on the uniform sampling points, including: Uniform sampling points are selected on the surface of the theoretical part's digital model according to a preset sampling density threshold; Calculate the unit normal vector of each of the uniform sampling points on the surface of the theoretical part's digital model; The deviation value of each uniform sampling point in the error field is extracted, and the deviation value is vector synthesized with the corresponding unit normal vector to obtain the deviation vector of each uniform sampling point. The vectors are then summarized to form a set of deviation vectors on the uniform sampling points.

6. The error compensation method for sheet metal according to claim 1, characterized in that, The step of extracting the main contributing component from the feature vector as the dominant deformation mode includes: Calculate the contribution rate corresponding to each feature vector, sort them from largest to smallest, and sum them up to obtain the cumulative contribution rate; If the cumulative contribution rate exceeds the preset contribution judgment threshold, the corresponding feature vector is extracted; if it does not exceed the threshold, subsequent feature vectors are extracted until the requirements are met, and the extracted feature vectors are used as the dominant deformation mode.

7. The error compensation method for sheet metal according to claim 1, characterized in that, The finite element simulation model is constructed based on the dominant deformation modes. Preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode, including: The dominant deformation mode is mapped to the finite element mesh node system, and the simulation boundary conditions are established by combining the preset mold deformation parameters and preset material fluctuation parameters to construct the finite element simulation model. The nodal displacement response under each dominant deformation mode is calculated using the finite element simulation model. The normal distance is obtained by analyzing the displacement response, and the local prediction deviation field corresponding to each mode is generated. By superimposing all the local prediction bias fields, the overall prediction bias field is obtained.

8. The error compensation method for sheet metal according to claim 1, characterized in that, The step of calculating the residual matrix based on the predicted bias field and the error field, and if the norm of the residual matrix exceeds a preset residual norm threshold, adjusting the simulation parameters and resimulating to obtain an optimized predicted bias field; if it does not exceed the threshold, using the predicted bias field as the optimized predicted bias field, includes: Spatial registration is performed between the predicted bias field and the error field, and interpolation is performed node by node to generate an initial residual matrix; Calculate the norm of the residual matrix. If the norm exceeds a preset residual judgment threshold, identify the simulation parameter that has the greatest impact on the residual and adjust it to obtain the adjusted parameter. The adjusted parameters are input into the simulation model to regenerate the prediction bias field. The residual matrix is ​​calculated repeatedly until the norm meets the requirements, thus obtaining the optimized prediction bias field.

9. The error compensation method for sheet metal according to claim 1, characterized in that, The step of determining the compensation vector for each region based on the optimized predicted deviation field, converting the compensation vector into specific adjustment instructions, and obtaining the mold error compensation and correction scheme includes: Extract the deformation values ​​of each region in the optimized prediction deviation field, and calculate the reverse displacement component as a compensation vector based on the deformation values; The compensation vector is projected onto the mold surface using the inverse mapping method to obtain the discrete compensation amount for each region of the mold. Spatial interpolation is performed on the discrete compensation quantities to form a continuous spatial distribution of compensation quantities; Based on the spatial distribution of the compensation amount, an adjustment instruction for mold processing is generated, and the adjustment instruction is input into the CNC system to obtain a mold error compensation and correction scheme.

10. An error compensation system for sheet metal, characterized in that, include: The data acquisition module is used to acquire the theoretical part model and the measured deviation data of the surface of the sheet metal parts; The mesh generation module is used to obtain the local curvature characteristics of the theoretical part's digital model, determine the mesh node spacing parameters based on the local curvature characteristics, and generate an adaptive parametric mesh. The error field construction module is used to map the measured deviation data onto the parameterized grid to obtain discrete deviation data, perform weighted fitting on the discrete deviation data, and convert the fitting result into a structured error field. The deviation vector module is used to calculate the deviation value of each uniform sampling point in the error field based on the error field and the unit normal vector of the surface of the theoretical part digital model, and to vector synthesize the deviation value with the unit normal vector of the corresponding sampling point to obtain the set of deviation vectors on the uniform sampling points. The deformation mode module is used to extract mutually orthogonal feature vectors from the set of deviation vectors, and extract the main contributing part from the feature vectors as the dominant deformation mode. The prediction simulation module is used to construct a finite element simulation model based on the dominant deformation mode. The preset mold deformation parameters and preset material fluctuation parameters are input into the finite element simulation model to obtain the prediction deviation field corresponding to each dominant deformation mode. The compensation and correction module is used to calculate the residual matrix based on the predicted deviation field and the error field. If the norm of the residual matrix exceeds a preset residual norm threshold, the simulation parameters are adjusted and the simulation is repeated to obtain the optimized predicted deviation field. If it does not exceed the threshold, the predicted deviation field is used as the optimized predicted deviation field. The scheme generation module is used to determine the compensation vector of each region based on the optimized prediction deviation field, and convert the compensation vector into specific adjustment instructions to obtain the mold error compensation correction scheme.

Images