Real-time drawing and verifying method for machined part pattern of bending machine numerical control system

Abstract

The invention relates to the technical field of real-time verification of bending machines, in particular to a real-time drawing verification method for a machined part pattern of a bending machine numerical control system. The method comprises the following steps: obtaining an original pattern and identifying geometric element features, generating feature region data, carrying out load safety margin calculation to obtain a safety threshold boundary, disassembling a principal stress direction and constructing a continuous stress flow field, carrying out constraint reconstruction on the geometric element features, and identifying variable contour data. The method comprises the following steps: acquiring target bending data, planning a bending track, predicting a pre-machining shape, generating a real-time machining gap through gap comparison with the target data, detecting elastic springback of a bent part, performing compensation parameter mapping based on springback data, simulating secondary machining, and finally performing real-time comparison and dynamic machining parameter fine adjustment. The machining process is optimized, and the machining efficiency and the finished product quality of the numerical control system of the bending machine are integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time verification of bending machines, and particularly to a method for real-time drawing verification of processed part drawings of a numerical control system for bending machines. Background Art

[0002] In the actual production process, due to multiple influences such as material properties, equipment accuracy, and environmental factors, the actual forming of bent parts often fails to achieve the expected design effect, resulting in problems such as reduced production efficiency and material waste. Existing bending processing verification methods mostly rely on experience and tests. This subjectivity and uncertainty lead to unstable machining accuracy and are difficult to meet the requirements of modern manufacturing for high precision and high consistency. In addition, many traditional technologies lack corresponding analysis tools when facing complex geometric shapes and are unable to effectively handle non-linear deformation and multi-axis interaction effects. Secondly, existing technologies often ignore the dynamic characteristics of materials. Especially under high frequency and high stress, the influence of material behavior changes on bending accuracy has not been fully considered, resulting in difficult prediction of the shape and dimensional accuracy of bent parts in actual applications. The existence of springback phenomenon further increases the complexity of process adjustment. Existing springback compensation methods rely on fixed parameters, lack flexibility and adaptability, and cannot achieve the best compensation effect. Summary of the Invention

[0003] Based on this, it is necessary to provide a method for real-time drawing verification of processed part drawings of a numerical control system for bending machines to solve at least one of the above technical problems.

[0004] To achieve the above object, the method for real-time drawing verification of processed part drawings of a numerical control system for bending machines includes the following steps: Step S1: Obtain the original drawing of the bent workpiece; identify the geometric element features of the original drawing of the bent workpiece, and perform boundary feature cutting based on the geometric element features to generate feature region data; perform load safety margin calculation on the feature region data based on the preset material data of the processed part to obtain the safety threshold boundary; Step S2: Decompose the safety threshold boundary in the principal stress direction and construct a continuous stress flow field; perform geometric constraint reconstruction on the geometric element features based on the continuous stress flow field and identify variable contour data; Step S3: Obtain the target bending data; perform bending trajectory planning on the variable contour data based on the target bending data to obtain a planned bending sequence; predict the pre-processed bending shape according to the planned bending sequence; Step S4: Perform corresponding position projection on the target bending data and the pre-processed bending shape, and perform gap comparison analysis to generate real-time processing gaps; perform elastic springback detection on the variable contour data based on the real-time processing gaps to obtain the elastic springback data of the bent part; Step S5: Perform springback compensation based on the springback data of the bent part to generate springback compensation data for bending; map the compensation parameters to the original drawing of the bent workpiece according to the springback compensation data for bending to obtain the gap compensation parameters; Step S6: Perform secondary machining simulation on the variable contour data according to the gap compensation parameters to generate secondary machining simulation data; perform real-time comparison based on the secondary machining simulation data and the target bending data, and perform fine-tuning of dynamic machining parameters according to the real-time comparison data.

[0005] The present invention provides a basic basis for the machining process by obtaining the original drawing of the bent workpiece. The identification of geometric element features and the generation of feature region data by cutting lay a foundation for subsequent analysis. The calculation of the load safety margin ensures the safety of the workpiece during use. The construction of the safety threshold boundary provides a reference for optimizing machining parameters. The implementation of the principal stress direction decomposition makes the stress analysis more accurate. The establishment of a continuous stress flow field promotes the reconstruction of geometric constraints. The identification of variable contour data lays a solid foundation for subsequent bending trajectory planning. The acquisition of target bending data provides a practical operation guide for the bending process. The bending trajectory planning enhances the flexibility and accuracy of bending. The prediction of the pre-machined bending shape provides an estimate of the actual machining result. The generation of real-time machining gaps reflects the deviation between the machining accuracy and the target. The elastic springback detection ensures a comprehensive understanding of the springback characteristics of the bent part. The generation of springback compensation data for bending provides a necessary basis for optimizing the finished product accuracy. The mapping of gap compensation parameters realizes the precise adjustment of the original drawing. The generation of secondary machining simulation data provides dynamic feedback for verifying the machining process. The analysis of real-time comparison data realizes the optimization of the machining process by fine-tuning machining parameters, overall improving the machining efficiency and finished product quality of the CNC system of the bending machine, and providing effective guarantee and support for achieving high-precision bending. Description of the Drawings

[0006] Figure 1 It is a schematic diagram of the step flow for the real-time drawing verification method of the workpiece drawing of the CNC system of the bending machine; Figure 2 It is a schematic diagram of the detailed implementation steps of Step S2; The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0007] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0008] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus the repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0009] It should be understood that although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0010] To achieve the above object, please refer to Figures 1 to 2 , a real-time drawing verification method for the machining part drawing of a bending machine numerical control system, comprising the following steps: Step S1: Obtain the original drawing of the bending workpiece; identify the geometric element features of the original drawing of the bending workpiece, and perform boundary feature cutting based on the geometric element features to generate feature region data; perform load safety margin calculation on the feature region data based on the preset workpiece material data to obtain the safety threshold boundary; Step S2: Decompose the safety threshold boundary in the principal stress direction and construct a continuous stress flow field; perform geometric constraint reconstruction on the geometric element features based on the continuous stress flow field and identify the variable contour data; Step S3: Obtain the target bending data; perform bending trajectory planning on the variable contour data based on the target bending data to obtain the planned bending sequence; predict the pre-processed bending shape according to the planned bending sequence; Step S4: Project the target bending data and the pre-processed bending shape at corresponding positions and perform gap comparison analysis to generate the real-time machining gap; perform elastic springback detection on the variable contour data based on the real-time machining gap to obtain the elastic springback data of the bent part; Step S5: Perform springback compensation based on the elastic springback data of the bent part to generate the bending springback compensation data; perform compensation parameter mapping on the original drawing of the bending workpiece according to the bending springback compensation data to obtain the gap compensation parameters; Step S6: Perform secondary machining simulation on the variable contour data according to the gap compensation parameter to generate secondary machining simulation data; perform real-time comparison based on the secondary machining simulation data and the target bending data, and perform dynamic adjustment of machining parameters according to the real-time comparison data.

[0011] The present invention provides a basic basis for the machining process by obtaining the original drawing of the bent workpiece. The identification of geometric element features and the generation of feature area data from cutting lay a foundation for subsequent analysis. The calculation of the load safety margin ensures the safety of the workpiece during use. The construction of the safety threshold boundary provides a reference for the optimization of machining parameters. The implementation of the principal stress direction decomposition makes the stress analysis more accurate. The establishment of the continuous stress flow field promotes the reconstruction of geometric constraints. The identification of variable contour data lays a solid foundation for the subsequent bending trajectory planning. The acquisition of target bending data provides a practical operation guide for the bending process. The bending trajectory planning enhances the flexibility and accuracy of bending. The prediction of the pre-machined bending shape provides an estimate of the actual machining result. The generation of the real-time machining gap reflects the deviation between the machining accuracy and the target. The elastic springback detection ensures a comprehensive understanding of the springback characteristics of the bent workpiece. The generation of bending springback compensation data provides a necessary basis for optimizing the finished product accuracy. The mapping of the gap compensation parameter realizes the precise adjustment of the original drawing. The generation of secondary machining simulation data provides dynamic feedback for the verification of the machining process. The analysis of real-time comparison data realizes the optimization of the machining process by fine-tuning machining parameters, overall improving the machining efficiency and finished product quality of the numerical control system of the bending machine, and providing effective guarantee and support for achieving high-precision bending.

[0012] In the embodiment of the present invention, the method for real-time drawing verification of the machining part drawing of the numerical control system of the bending machine includes the following steps: Step S1: Obtain the original drawing of the bent workpiece; identify the geometric element features of the original drawing of the bent workpiece, and perform boundary feature cutting based on the geometric element features to generate feature area data; calculate the load safety margin for the feature area data based on the preset workpiece material data to obtain the safety threshold boundary; In this embodiment, when obtaining the original drawing of the bent workpiece, a high-precision two-dimensional CAD drawing input module is used to access the drawing files in AutoCAD format (.dwg) or STEP format (.stp). During the drawing reading process, the boundary recognition function is called and the Boundary Scan Segmentation algorithm is used to convert the drawing contour data into a set of line segments. All sets of line segments need to meet the constraint conditions that the length threshold is greater than 1.0 mm and the included angle tolerance is less than 1.5°. When identifying geometric element features, the RANSAC (Random Sample Consensus algorithm) is applied to perform linear fitting and arc extraction on the boundary line segments. A corner point mark is established at the position where the adjacent included angle change of the line segment combination structure is greater than 3°. Subsequently, the contour closure determination module is called for the area between the marked nodes to complete the area division. Based on the identified geometric element structure, the primitive-level partitioning algorithm is executed to complete the boundary feature cutting. The generated feature area data includes the closed contour path, internal structure element number, and attribute label of each area. The boundary feature cutting precision error needs to be less than ±0.2 mm. Then, the material information corresponding to the drawing of this workpiece is retrieved from the database. The material number is GB-T700 Q235, the yield strength is set to 235 MPa, the elastic modulus is 210 GPa, and the Poisson's ratio is 0.3. Based on the above material parameters, the finite element micro-element distribution function is called to perform grid element load distribution on each feature area. The Von Mises stress is calculated for each element using the equivalent stress method. The loading method is standard constant pressure loading, and the pressure direction is perpendicular to the normal direction of the area boundary. The element distribution range is controlled within a grid spacing of 0.5 mm. According to the principle that the maximum element stress does not exceed the yield limit, the critical load boundary line of each feature area is extracted and defined as the safety threshold boundary.

[0013] Step S2: Decompose the safety threshold boundary in the principal stress direction and construct a continuous stress flow field; based on the continuous stress flow field, perform geometric constraint reconstruction on the geometric element features and identify the variable contour data; In this embodiment, when decomposing the safety threshold boundary in the principal stress direction, the eigenvalue decomposition operation is performed on each boundary element node using the equivalent stress tensor matrix to extract the major principal stress direction (Major Principal Stress Direction). The main direction range is limited to 0° to 180°. Based on this, a direction vector field is established. The discrete vector field is extended into a continuous stress flow field using the structural interpolation method (Structural Interpolation Method). The distribution density of the vector field is set to no less than 4 direction vector nodes per square millimeter. Subsequently, this continuous stress flow field is superimposed on the original geometric element feature coordinate system, and the direction offset fitting operation is performed to reconstruct the geometric constraints, restricting the offset deformation angle not to exceed ±1.8°. The reconstructed boundary curve is then subjected to fitting and smoothing, with the fitting error controlled within 0.15 mm. Finally, based on the reconstructed curve, the deformation-sensitive segment is identified under the constraints of curvature continuity and length, and marked as variable contour data. The data structure includes the curvature change rate, segment length, average principal stress direction, and number index.

[0014] Step S3: Obtain the target bending data; perform bending trajectory planning on the variable contour data based on the target bending data to obtain a planned bending sequence; predict the pre-processed bending shape according to the planned bending sequence. In this embodiment, when obtaining the target bending data, a standard bending process flow table is imported through a human-machine interaction terminal, including process number, bending sequence, angle setting, linear displacement, and tool parameters. The target bending angle range is set from 30° to 150°, and the minimum bending radius is limited to 1.5 mm. Based on this data, a trajectory control algorithm based on path optimization is performed on the variable contour data for bending trajectory planning. The variable step-size path evolution module is enabled in the control algorithm, and the trajectory node spacing is not greater than 2 mm. The bending path needs to construct a continuous and smooth path within the allowable minimum curvature radius range, and the control point sequence is recorded. Finally, a planned bending sequence is obtained. The storage format of the bending sequence is a structured data set including curvature control points, bending angles, and loading positions. Subsequently, based on the planned bending sequence, a middleware deformation simulation module is constructed to perform non-linear large deformation simulation in the ANSYS Workbench environment. The simulated material properties are consistent with the previous material parameters, the simulation loading speed is 5 mm / s, and the number of simulation steps is 200 steps. The stability of the intermediate configuration is used as the stop determination condition to generate the predicted pre-processed bending shape. This data is a set of deformed nodes with timestamps, including the spatial positions, principal strain directions, and curvature vector information of each point.

[0015] Step S4: Project the target bending data and the pre-processed bending shape at corresponding positions, and perform gap comparison and analysis to generate real-time processing gaps; perform elastic springback detection on the variable contour data based on the real-time processing gaps to obtain the elastic springback data of the bent part. In this embodiment, when performing the corresponding position projection operation on the target bending data and the pre-processed bending shape, the bidirectional projection comparison function is called to perform full-profile mapping on the two shape models. The error matching function based on point cloud density weighting is used to calculate the local normal distance difference at each mapping node. The error tolerance is set to ±0.25 mm, and the real-time processing gap data is generated. Subsequently, based on this gap data, the non-linear deformation response model is called to establish the elastic springback displacement path model between the corresponding nodes. The reaction displacement threshold is set to 0.15 mm, and the maximum springback angle deviation is 2°. The elastic springback data of the bent part is calculated by reverse-pushing the springback path, and the output result is the matrix structure data including the displacement vector direction, springback angle, and residual stress mapping.

[0016] Step S5: Perform springback compensation based on the elastic springback data of the bent part to generate the bending springback compensation data; perform compensation parameter mapping on the original drawing of the bent part according to the bending springback compensation data to obtain the gap compensation parameters. In this embodiment, based on the elastic springback data of the bent part, the highly sensitive springback region is extracted by the springback absorption region recognition algorithm, and its local contour data is subjected to stress inverse solution reconstruction operation, and the corrected configuration function is generated. The regional deformation gain parameter constructed in combination with this corrected function is superimposed on the original springback data to form the bending springback compensation data. The compensation data includes the angle increment, line segment extension ratio, and local curvature adjustment vector. Under the guidance of this data, the reverse geometric mapping is performed on the original drawing of the bent part, and the compensation amount is mapped into the boundary curve of the original drawing. The process needs to keep the boundary closure error not exceeding 0.05 mm. The output gap compensation parameters include the deformation function expression of each boundary curve and the angle correction vector matrix.

[0017] Step S6: Perform secondary processing simulation on the variable contour data according to the gap compensation parameters to generate secondary processing simulation data; perform real-time comparison based on the secondary processing simulation data and the target bending data, and perform dynamic adjustment of the processing parameters according to the real-time comparison data.

[0018] In this embodiment, the machining simulation subsystem is called according to the gap compensation parameter, and a compensated contour model is constructed in the ABAQUS simulation platform. The grid element density is set to 0.25 mm, the material properties are called from the original data model, the loading condition is set to constant-speed pressurization of the servo hydraulic system, the pressure increment is 50 N / step, the secondary machining simulation is performed, the simulation step size is 0.1 s, and the secondary machining simulation data is output. The data structure includes the configuration states of each time node during the bending process, the stress distribution map, and the node position transformation matrix. Subsequently, the data is subjected to spatial mapping reconstruction with the target bending data, three-dimensional geometric alignment is performed, and the residual distribution map is fitted using the least square error matching function. When the residual extreme value does not exceed 0.2 mm, the dynamic machining parameter fine-tuning module is activated to finely adjust the feed speed, bending angle setting, and tool travel output by the bending control system. The adjustment range of the feed speed is ±1.2 mm / s, the adjustment range of the bending angle is ±1.5°, and the adjustment accuracy of the tool travel is ±0.1 mm. Finally, a set of machining feedback correction parameters is output for the secondary closed-loop correction of the control system.

[0019] Preferably, identifying the geometric element features of the original pattern of the bent workpiece in step S1 and performing boundary feature cutting based on the geometric element features includes: Identifying the geometric element features of the original pattern of the bent workpiece; Constructing a topological relationship based on the geometric element features and drawing a connection topology diagram based on the topological relationship. Among them, the minimum distance between nodes in the connection topology diagram shall not be less than 0.1 mm, and the maximum connectivity error in the topological path shall not exceed 0.05 mm; Identifying the pattern boundary contour based on the connection topology diagram; Performing solid feature cutting on the pattern boundary contour, performing area cutting according to the minimum unit segmentation size of 1 mm², and controlling the aspect ratio of the cutting area within 1:5 to 5:1 to generate feature area data.

[0020] In this embodiment, in the real-time drawing verification method of the workpiece pattern of the bending machine numerical control system, the geometric element feature recognition operation of the original workpiece pattern is processed by means of vector graphics analysis. First, the original pattern in the input format of DXF (Drawing Exchange Format) or SVG (Scalable Vector Graphics) is imported into the CAD core engine for vectorized boundary extraction processing. The ShapeAnalysis_ShapeTolerance class in the OpenCascade geometric modeling kernel is called to uniformly transform all straight lines, arcs, broken lines, and B-spline curves and classify them into a geometric element feature set. They are grouped and numbered according to the entity primitive type, and the endpoints of the line segments, the control points of the curves, and the boundary contact points between adjacent faces are marked. All point position data are retained to 6 decimal places of precision to ensure that the error does not exceed 0.0001 mm during the boundary connection process. After generating the initial feature point set and line segment set, it enters the topological relationship construction link. The topological relationship construction is jointly executed by the graph traversal and the minimum connection optimization strategy. First, the directed graph structure in the Boost Graph Library is called. The starting point and the ending point of the geometric element feature are used as the nodes of the graph respectively, and the line segments and curves are defined as the edges of the graph. The edge weight is set as the actual length of the line segment. After constructing the initial connection graph, the Dijkstra shortest path algorithm is used to identify the minimum path set between each geometric element feature. After screening out the non-closed structure or local isolated features, it enters the topological correction stage. The connection distance between all nodes in the graph shall not be less than 0.1 mm, otherwise it is regarded as misconnection or redundant points, and the adjacent point merging strategy based on Euclidean distance calculation is executed to merge the node pairs less than 0.1 mm into a single node, and at the same time, its boundary direction vector and connectivity angle are recalculated. The maximum connectivity error of all topological paths shall not exceed 0.05 mm, all path error cumulative values are checked through the floating-point error cumulative check method to ensure the closed effectiveness of the topology map, and thus the connection topology map is drawn. In the stage of identifying the boundary contour of the pattern, the BRepBuilderAPI_MakeWire class in OpenCascade is called to construct the boundary set in the topology map into a closed contour structure. By using the method of judging the clockwise direction of the vector, all geometric element feature combinations that form the outermost contour line are screened out, and the boundary normal vector direction is corrected by judging its direction consistency, completing the reconstruction of the outer contour of the pattern. On this basis, the BRepOffsetAPI_Sewing interface is called to construct the closed contour area into a two-dimensional solid structure. Entering the entity feature cutting stage, the entity feature cutting is discretely processed by setting the parameters of the cutting grid. The BRepTools_Substitution class in OpenCascade is called to partition the contour area in a uniform grid manner, and the cutting size of the smallest unit area is set to 1 mm². The horizontal and vertical layer-by-layer scanning cutting mode is adopted, and the aspect ratio parameter of each cutting block is set between 1:5 and 5:1. The boundary area is automatically adjusted to the legal ratio of the closest range. For example, when processing a rectangular contour with a width of 3 mm and a height of 15 mm, it is divided into 5 vertical areas, each unit size is 1 mm × 3 mm, with a total of 15 sub-areas and automatically numbered as R1~R15. The geometric information of each sub-area includes corner point coordinates, boundary normal vectors, the corresponding relationship with the geometric element features in the original pattern, and the topological association number. Finally, all area data are uniformly saved as the feature area data structure.

[0021] Preferably, the load safety margin calculation for the feature area data based on the preset workpiece material data in step S1 includes: Associating the feature area data with the material type based on the preset workpiece material data to generate material distribution data; Projecting the anisotropy index of the material distribution data and constructing a directional strength matrix based on the projected data; Mapping the material attributes of the feature area data according to the directional strength matrix to obtain a set of material parameters; Reconstructing the material microstructure according to the set of material parameters and performing constitutive relationship conversion based on the material microstructure to generate a material stress tensor; Calculating the critical load distribution based on the material stress tensor and performing safety margin limitation based on the critical load distribution to obtain the safety threshold boundary.

[0022] In this embodiment, when associating the material type of the feature region data based on the preset workpiece material data, first, the workpiece drawing is imported into the drawing analysis module. This module performs voxel segmentation processing on the CAD three-dimensional model, divides the overall model into several micro-unit regions containing machining behavior characteristics, and each micro-unit contains a unique spatial coordinate index. On this basis, the basic material data of the preset workpiece is loaded. This data includes material code, crystal structure type, heat treatment status, yield strength, elastic modulus, Poisson's ratio, and hardening function surface, which are matched and selected by the material database according to the machining process version number, and the material data is mapped into each feature region micro-unit to form a spatial distribution material map. Each micro-unit is precisely bound to its corresponding material type in the spatial dimension to form material distribution data containing three primary key fields: three-dimensional coordinates, material code, and material parameters. When projecting the anisotropy index of the material distribution data, a machining direction coordinate system is constructed. This coordinate system is established based on the alignment benchmark of the upper and lower dies of the bending machine. The X direction is set as the workpiece axial direction, the Y direction is the width direction, and the Z direction is the normal loading direction. Subsequently, the single-crystal elastic modulus and shear modulus values in each direction are extracted according to the material crystal structure, and directional remapping is performed in the coordinate system to numerically project the rigid response ability of each material in the current spatial direction to generate a directional index set. After the projection is completed, a directional strength vector is established for each micro-unit, which includes the elastic response strength in three orthogonal directions and the strength levels in three shear directions, and a spatial strength matrix is constructed for subsequent loading analysis. When mapping the material properties of the feature region data according to the directional strength matrix, the material parameters of the micro-units are adjusted relying on the anisotropy information, and the strength distribution information in different directions is coupled with the basic material parameters. The elastic modulus change curve, yield strength correction value, and strain hardening interval related to the direction are recorded in each micro-unit to generate a material parameter set. This set is a composite material feature table under the three-dimensional coordinate index, and its structure is a multi-field data table. The main fields include spatial index, elastic modulus, Poisson's ratio, yield limit, and the numerical interval of yield strength in each direction. When reconstructing the material microstructure according to the material parameter set, a three-dimensional reconstruction engine is used to perform microscopic grain modeling. The average grain size is set to 10 microns, the grain orientation is generated by the ODF function distribution according to the machining and heat treatment history, the grain boundary position is drawn using the spatial minimum energy distribution method, the grain boundary thickness is set to 0.2 microns, the grain boundary slip behavior is constructed using the critical shear model defined in the GB2312-3 metal hot working data, and the intragranular slip direction is defined using the main slip system of the FCC crystal structure. These microstructures are input into the constitutive model generator, and an elastic-plastic constitutive relationship is used to generate a stress tensor field, and the tensor is expressed as σ ij, where i and j are coordinate direction indices, and the values ​​represent the stress response in each direction per unit area, in megapascals (MPa). In the process of calculating the critical load distribution based on the material stress tensor, the maximum stress in the direction of the principal axis of the stress tensor is analyzed to obtain the maximum principal stress of each micro-unit, and compared with the yield strength of the material in the current processing direction, the von Mises yield criterion is used. Determine the yield state and establish a yield area map, where s ij is the component of the deviatoric stress tensor, which represents the pure shear stress after removing the average stress in each direction from the total stress tensor, and describes the stress component that causes deformation in the material stress state. f(σ) is the vonMises yield function. function), which is used to determine whether the material has entered the yield state. When f(σ)≥0, it means that the material is in the yield or plastic state. When f(σ)<0, the material is still in the elastic state. f(σ) is the yield strength, which represents the equivalent stress value of the material when the force reaches the yield. This value is one of the material property parameters. According to the material model, heat treatment state, grain structure and other settings, the area exceeding the yield limit in the map is defined as the critical load point set. Based on the spatial aggregation density and position distribution of the point set, a three-dimensional safety boundary model is constructed. The load deviation degree of all micro-units with yield risk is calculated and marked, and finally a safety threshold boundary map is generated. This map is used as a dynamic interference judgment input in real-time processing path optimization and can be loaded into the bending machine CNC system in real time for processing pattern drawing and processing behavior regulation.

[0023] Preferably, step S2 comprises the following steps: Step S21: performing surface feature recognition on the safety threshold boundary, and identifying a feature control area based on the surface feature; performing mesh division based on the feature control area to obtain a finite element mesh; Step S22: Decomposing the principal stress direction based on the finite element network to obtain the principal stress direction vector; constructing a continuous stress flow field according to the principal stress direction vector; Step S23: performing density gradient mapping based on the continuous stress flow field, and extracting stress hot zone data based on the density gradient; performing material density distribution analysis based on the stress hot zone data to obtain material density distribution; Step S24: identifying unit stress concentration points based on material density distribution; performing local shape correction on the unit stress concentration points to obtain corrected stress concentration points; Step S25: performing stress distribution equalization adjustment on the geometric element features based on the modified stress concentration points, and reconstructing geometric constraints based on the stress distribution equalization adjustment data to generate adjustment restriction conditions; Step S26: Perform unrestricted contour recognition on the geometric element features according to the adjustment limit conditions to obtain variable contour data.

[0024] In this embodiment, the safety threshold boundary data generated in the previous stage is input into the calculation module based on Gaussian curvature in the form of three-dimensional point cloud. Gaussian curvature calculation operations are performed on each boundary surface point. A spherical neighborhood with a local window radius of 2.5 mm is used to estimate the curvature of each boundary point. A local fitting surface function z = f(x, y) is constructed for the neighborhood point set by the least squares method. Then, the principal curvatures k1 and k2 are calculated, and their product is used as the Gaussian curvature value of this point. All regions with Gaussian curvature values greater than 0.15 are marked as concave regions to form a feature control region. Subsequently, a three-dimensional hexahedron automatic mesh generation tool is called to perform volume mesh generation operations on the volume within the control region. The initial mesh density is set to 0.8 mm, and the boundary mesh is refined based on the Poisson mesh optimization algorithm to generate a finite element mesh with boundary continuity. After the finite element mesh is generated, the principal stress solution module is input based on the boundary conditions. Fixed boundary conditions are applied to each hexahedron element, and the initial external load direction is defined as the positive direction of the Y-axis. The loading amplitude is set to 250 MPa. The stress tensor components σxx, σyy, σzz, σxy, σyz, and σxz of each element are obtained by solving with a three-dimensional elastic finite element solver. The eigenvalue decomposition operation is performed on each tensor by the stress principal value decomposition method to calculate the principal stresses σ1, σ2, and σ3 and their corresponding eigenvector directions v1, v2, and v3. The first principal stress direction v1 is extracted as the principal stress direction vector set. The principal stress direction vectors are spherically interpolated according to the continuity between elements. A continuous stress flow field is constructed through the interpolated principal stress direction vectors, and this flow field is defined as the vector field T(x, y, z). The streamline smoothing operation is performed on its direction components using the Bezier curve interpolation method to finally obtain a steerable continuous vector field data set. Based on the aforementioned continuous stress flow field T(x, y, z), the density gradient projection mapping operation is adopted. The initial value of the unit density is set to 7.85 g / cm³. The unit length direction integral is performed on all paths in the flow field, and the density change rate ∇ρ(x, y, z) of each unit path is calculated. After normalizing the density change rate over the entire domain, it is used as the weight for extracting the hot zone. The weighted thermal mapping method is used to extract the high-density change region and set it as the stress hot zone. The hot zone threshold is set to a density gradient ∇ρ greater than 1.In the region of 2 g / cm³ / mm, the corresponding material density data are extracted according to the grid cell coordinate indices covered by these hot regions. The mean density distribution ρ_i within each cell is calculated by the volume averaging method, and a three-dimensional density distribution function ρ(x, y, z) is constructed. The material density distribution function ρ(x, y, z) is input into the stress concentration point identification module. The local peak detection algorithm is used to search for the maximum value of the density gradient modulus ∥∇ρ∥ for each grid cell. The window size is set to 3×3×3 cells. If the density of the central cell is greater than the densities of all its neighboring cells and the density gradient modulus is greater than 3.5 g / cm³ / mm² within this window, it is marked as a cell stress concentration point. The fifth-order Bezier surface around this point is fitted into the initial local geometric model. When performing shape correction, the Bezier control points are projected and moved along the principal stress direction by a distance d = (σ / σ_max)×Δx, where σ is the first principal stress of this cell, σ_max is the maximum principal stress of the whole field, and Δx is the side length of the cell. The corrected stress concentration point position data set is obtained through the above method. Using the corrected stress concentration point data as the input, the stress distribution of the geometric element features is evenly adjusted. The structure topology optimization module is called, and the optimization objective function is set to the maximum stress σ_minimization. Under the condition of being constrained by the stress field σ(x, y, z), the shape of the geometric element boundary line is reparameterized through the variable shape function. The shape function uses the quartic B-spline function, and the control point spacing of each shape parameter vector is limited within 0.2 mm. The control point positions are adjusted so that the stress difference Δσ_i = |σ_i - σ_mean| of each cell after correction is less than 15 MPa. After the even adjustment is completed, the obtained control point coordinate set is defined as the geometric constraint reconstruction condition and output to the next module for contour data recognition. The geometric element features are processed for contour recognition without restrictions. Using the above control point coordinate set as the constraint control input, the boundary point cloud map structure of the three-dimensional geometric model is recognized through the Graph Convolutional Network (GCN) algorithm, and the graph topology structure G(V, E) is constructed. The boundary continuity of each node V is judged, and unrestricted contour recognition is performed based on the region where the boundary curvature change rate is greater than 0.25 / mm. During the recognition process, the graph traversal operation is carried out starting from the root node at the starting point of the curved boundary, and all unrestricted contour boundary paths are recorded. Finally, the contour paths are output in the form of ordered coordinates to generate a variable contour data set.

[0025] Particularly importantly, step S23 includes: Quantify the gradient direction of the continuous stress flow field to obtain stress gradient distribution data; Perform gradient intensity classification on the stress gradient distribution data to generate gradient intensity classification data; Perform hotspot threshold fitting on the gradient intensity classification data to generate the stress hotspot threshold; Extract the hotspot range based on the stress hotspot threshold and the gradient intensity classification data to obtain the stress hotspot data; Estimate the material density corresponding to the hotspot according to the stress hotspot data; Perform material distribution mapping based on the material density corresponding to the hotspot to obtain the material density distribution.

[0026] In this embodiment, a stress tensor matrix based on each unit node is constructed, the principal stress direction vector is subjected to coordinate mapping, and a Gaussian filter is selected to denoise each principal stress direction vector in the three-dimensional continuous stress field. The width of the filter kernel is set to σ = 1.5, and the step size is Δx = 1.0 unit length. The Sobol gradient operator is used to solve the spatial derivative of each denoised node vector, and the discrete numerical derivatives of the stress change rates in the X-axis, Y-axis, and Z-axis directions are calculated respectively. Then, through the transformation of the three-dimensional Euler angles, the directional derivatives at each node are unified into the three-dimensional polar coordinate expression form, and finally, the gradient direction angle distribution matrix between nodes is formed, with the unit of radians and the range set in [0, 2π]. During the process of grading the stress gradient distribution data by gradient intensity, the maximum-minimum normalization method is used to scale the gradient modulus value corresponding to each node to between 0 and 1. The gradient modulus value is calculated by the Euclidean norm of the vector derivative, and its expression form is G = √(∂σ / ∂x)²+(∂σ / ∂y)²+(∂σ / ∂z)². Subsequently, the normalized G values are divided into five levels, and the level thresholds are set to [0, 0.2), [0.2, 0.4), [0.4, 0.6), [0.6, 0.8), [0.8, 1.0] respectively, and are marked as L1 to L5. Each node is assigned to the corresponding category according to the gradient level it belongs to, and a hierarchical mapping grid in the three-dimensional space is constructed. The side length of each grid is set to 2.5 mm. During the process of fitting the hot zone threshold for the gradient intensity classification data, the Logistic regression surface fitting method is used to fit the coordinate sets of the L4 and L5 level gradient nodes in the spatial distribution. The Logistic function is set to f(x, y, z) = 1 / (1 + e^-(a·x + b·y + c·z + d)). The fitting coefficients a = 0.12, b = -0.08, c = 0.1, and d = -1.7 are obtained through the least squares fitting method. In the three-dimensional space, the critical fitting threshold T = 0.5 is set according to the function output value. When f(x, y, z) ≥ T, the corresponding node is marked as a potential stress hot zone node, forming the spatial boundary of the hot zone threshold. During the process of extracting the hot zone range based on the stress hot zone threshold and the gradient intensity classification data, first, the nodes that meet f(x, y, z) ≥ 0 are screened out from the node sets with gradient classifications of L4 and L5.5 node groups, and then the voxel growth algorithm based on eight-neighborhood connectivity is used for cluster identification. The hot zone set with more than 50 voxels connected continuously is defined as the valid hot zone, and the boundary volume of the hot zone is constructed. Box) to record the minimum volume space of each hot zone. In the process of calculating the material density corresponding to the hot zone according to the stress hot zone data, the local structural stiffness K=E·A / L (where E is the Young's modulus of the material, A is the cross-sectional area, and L is the unit length) of the material is based on the relationship σ=K·ε (strain) with the local stress σ. Assuming that the workpiece is in a static load uniform state under the stress state, the average stress σ_avg and the average strain ε_avg within the boundary of each hot zone are used to infer the local structural stiffness K_avg=σ_avg / ε_avg. The relative density ρ_local=K_avg / K_ref of the current hot zone is further derived by the ratio of K_avg to the conventional material stiffness, where K_ref is the structural stiffness of the standard density material, which is 45GPa. The density value distribution of each hot zone voxel is obtained in the above method, and the unit is g / cm³. Finally, the local material density matrix of each hot zone is formed. In the process of material distribution mapping based on the material density corresponding to the hot zone, the inverse distance weighted interpolation method (IDW, Inverse Distance Weighted Interpolation) is used. Weighting) propagates the density value in the hot zone to the boundary of the hot zone, sets the influence radius to 10mm, sets the interpolation power index to 2, and the calculation formula is ρ(x,y,z)=∑(ρ_i / d_i. 2 ) / ∑(1 / d_i 2 ), where ρ_i is the known density in the hot zone, d_i is the distance between the current point and ρ_i, and finally a complete material density distribution map is constructed in the entire three-dimensional grid of the workpiece with a density resolution of 0.1g / cm³. The map is stored in Voxel format with a resolution of 0.5mm³. Each Voxel contains the corresponding spatial position and its density value, and the output is a three-dimensional density volume data set.

[0027] Preferably, the step S3 of performing bending trajectory planning on the variable contour data based on the target bending data includes: Perform boundary constraint mapping on target bending data to obtain boundary constraint data; Perform bending angle serialization processing based on boundary constraint data to obtain angle serialization data; Perform multi-axis collaborative path mapping on the variable contour data according to the angle serialization data to generate candidate bending trajectories; Perform bending simulation based on the candidate bending trajectory, and perform deformation field mapping based on the simulated bending data to generate deformation topology data; Detection of bending conflict structures in deformed topological data; Avoid conflict trajectories for candidate bending trajectories based on the bending conflict structure, and plan bending actions according to the conflict avoidance trajectories to obtain the planned bending sequence.

[0028] In this embodiment, during the CAD drawing input stage, B-spline fitting operation is performed on the two-dimensional bending profile. The number of control points is set to be not less than 30, and the fitting error is kept less than 0.05 mm. After fitting, the curvature continuity of the bending start point, end point, and turning point is extracted. The initial tangential vector and the end normal vector of each bending segment are calculated respectively to construct the initial boundary condition set. Subsequently, a static configuration constraint framework is introduced. Under the reference workpiece coordinate system, the boundary vertical constraint value in the Z-axis direction is set to ±0.03 mm, and the linear spread range constraint in the X-axis direction is ±1.2 mm. The boundary values are converted into the local constraint matrix of each B-spline control point through the three-dimensional space boundary projection algorithm, and finally the boundary constraint control set is formed. The boundary set corresponding to each bending segment contains six rigid parameters: three-dimensional coordinate offset, curvature offset, and direction rotation angle. All boundary data are stored in the three-dimensional tensor format with the size of [number of bending segments, 6]. During the process of serializing the bending angle based on the boundary constraint data, a polar coordinate expression is constructed with the geometric center of the workpiece as the coordinate origin. The θ angle and r radius projection are performed on each bending control point, and the tangential angle between adjacent bending segments is calculated. The angle is recorded in degrees, with the unit of degree, and the value range is within [0, 180], and the angle accuracy is controlled within 0.1°. A sorting algorithm is used to sort all the angles in ascending order to ensure that the bending path is arranged in the angle evolution direction. If there are adjacent bending segments with an angle difference less than 2°, they are merged into one segment and the angle center point is reconstructed. Finally, the bending angle sequence is output. Each sequence unit contains the start point, end point, angle value, and direction vector. The total length of the angle sequence is consistent with the number of bending segments, and the sequence is saved in the matrix form with the dimension of [n, 4]. During the process of multi-axis collaborative path mapping of the variable contour data according to the angle serialization data, a five-axis motion control algorithm is used, and the rotation accuracy of each control axis is set to ±0.01°. A five-axis attitude mapping function M(θ, x, y, z) is introduced in the path generation, where θ is the bending angle, and (x, y, z) is the bending start point coordinate. Each bending angle is mapped into the five-axis coordinate parameters through the forward kinematics model to form the spatial bending path of the variable contour. The distance between each node in the path is 0.5 mm, and the control axis angle and the tool die contact point pose are recorded at the node to generate the preliminary candidate bending trajectory path set. The trajectory set is recorded in the discrete path group manner, and each group of path data contains more than 1000 discrete points. The path format is a five-dimensional vector group: [x, y, z, θ1, θ2], where θ1 and θ2 are the tool die angles of the Y-axis and Z-axis respectively. During the process of bending simulation based on the candidate bending trajectory and deformation field mapping based on the simulated bending data, the finite element simulation engine ABAQUS is used to perform step-by-step loading simulation on each bending trajectory segment. The material model is set to the elastoplastic constitutive relationship, and the Young's modulus value in the material parameters is 210 GPa, and the Poisson's ratio is 0.3. The yield strength is set to 280 MPa. A 100-step non-linear incremental solution is carried out for each bending section during the loading process. After the solution is completed, the equivalent plastic strain (PEEQ), equivalent stress (von Mises), and principal strain direction information of the nodes in the bending area are extracted. The three-dimensional grid interpolation function is used to map the deformation information on the nodes back to the CAD space to form a three-dimensional deformation field. The deformation field is expressed in the form of a topological grid, and the size of each grid unit is 0.25 mm³. The generated deformation topology data volume is 512×512×64 grid units. Each unit inside contains stress-strain pairs, node displacement vectors, and plastic deformation index values. During the process of detecting the bending conflict structure in the deformation topology data, the vector angle analysis is carried out using the principal strain direction extracted from the deformation field tensor and the die movement direction. The angle threshold is set to 45°. If the angle between the local deformation principal strain direction and the current bending movement direction exceeds the threshold, and at the same time the equivalent plastic strain value in this area exceeds 0.12, it is determined as a bending conflict area. The voxel clustering algorithm is used to partition the continuous conflict area. The minimum boundary volume of each conflict structure needs to exceed 50 grid units. After extracting the bounding box of the conflict area, the conflict voxel set is output, and the conflict segment number is marked in the three-dimensional trajectory space to generate a conflict structure identification map. The map is represented in binary volume, where 1 represents conflict voxels and 0 represents non-conflict. During the process of avoiding conflict trajectories for the candidate bending trajectories based on the bending conflict structure and planning the bending actions according to the conflict avoidance trajectories, the path obstacle avoidance algorithm RRT* (Rapidly-Exploring Random Tree Star) is called for each conflict segment to adjust the bending trajectory. The maximum step size is set to 1.5 mm, and the maximum number of iterations is 3000 times. In the path generation, the non-conflict area is used as the passable space, and the conflict identification map is used as the obstacle body to dynamically generate new trajectory path segments. The trajectory node interpolation is smoothed by quintic B-spline, and the path curvature is controlled within 0.1, and the corner smoothing error does not exceed 0.2°. All the regenerated path segments are merged and globally re-optimized, and finally a complete bending action sequence is output. Each action contains five-axis control parameters, die contact point posture, bending angle instruction, and execution timestamp. The bending action sequence format is time-ordered structured data, and the precision requirement is one frame of control data per 0.01 s.

[0029] Preferably, the predicting the pre-processed bending shape according to the planned bending sequence in step S3 includes: Reconstructing the variable bending part structure according to the variable contour data; Performing path torque mapping on the variable bending part structure based on the planned bending sequence to obtain bending force guidance data; Performing displacement curvature folding on the bending force guidance data to generate configuration trend data; Based on the configuration trend data, geometric changes of the variable bending part structure are speculated to obtain the pre-processed bending shape.

[0030] In this embodiment, during the process of reconstructing the variable bending part structure based on the variable contour data, first, all closed contour boundaries are extracted from the 2D CAD drawing data through the Open CASCADE Technology modeling engine. The boundaries are expressed in vector format, and each boundary consists of multiple line segments and arc segments. The line segments and arcs are uniformly converted into a set of midpoint coordinates, start and end coordinates, direction angles, and normal angles. After generating the complete contour topology information, a custom geometric segmentation function is called to perform contour segmentation with a precision of 1 millimeter per segment, and the discrete point set sequence is extracted as the input coordinate set. After the coordinate set is input into the boundary structure reconstruction module, variable bending structure units are generated segment by segment according to the topological continuity of the contour. Each structure unit is numbered in a fixed order, and each boundary structure is recorded in the structure model according to its connection angle and length. When modeling, the plate thickness is uniformly set to 3 millimeters, the bending inner radius is set to 4 millimeters, and the bending angle is inferred from the included angle of the boundary segment and mapped into the 2D vector model. After all structure units are generated, they are integrated into a standard structure library data and uniformly output as a STEP format file for subsequent bending physical mapping calculations. The connection method between boundary segments is set to inelastic rotational connection, and no intersection nodes with more than two structure segments overlapping are allowed in the structure model. During the process of performing path torque mapping on the variable bending part structure based on the planned bending sequence, first, the generated STEP structure file is imported into the bending path simulator. The path simulator consists of analysis tools constructed based on the finite segment loading model. The tool supports the definition of the one-to-one mapping relationship between structure segments and loading nodes. When performing torque mapping, the number of each structure unit in the planned path is corresponded to the structure segment index. In each bending segment, virtual loading nodes are inserted at 10 equal intervals. At the loading nodes, the bending simulation displacement amounts are applied in sequence. After each displacement is applied, the reverse internal force distribution of this segment is extracted and mapped into an equivalent static moment value. The data of all loading nodes are stored as a multi-dimensional matrix, and each element of the matrix represents the local equivalent torque of the structure segment under a certain loading state. This matrix finally forms the bending force guide data, which is stored in a TXT format file with row and column labels in ASCII encoding. The structure segment numbers not included in the bending sequence are skipped without generating data. During the process of performing displacement curvature folding on the bending force guide data, the TXT format bending moment data is loaded into the deformation folding inference module. This module performs local affine transformation operations for each structure segment. When processing each structure segment, first, the corresponding torque values are read according to the loading node order, and then they are mapped into angle increments and linear displacement vectors in sequence. All displacement changes are combined and superimposed in the 2D plane. The superimposition method uses the main direction of the structure segment as the reference axis, and cumulative rotation and offset are performed according to the bending loading order. The rotation operation is achieved by constructing a 2D rotation matrix, and the offset operation is achieved by constructing a translation vector. After all structure segments are processed, a set of configuration trend data is generated, which records the endpoint displacement and angle change information of each structure segment under different loading states.The transformation information of all structural segments is uniformly encoded into a binary block structure. The data block of each structural segment contains the starting coordinate, the ending coordinate, the cumulative angular change amount, and the cumulative translation vector. After all data blocks are concatenated in sequence, a set of configuration trend trajectory files is formed. During the process of geometric change speculation on the variable bending part structure based on the configuration trend data, the configuration trend trajectory files are loaded into the structure feedforward deformation module. When inputting, this module reconstructs the complete structure connection diagram according to the structural segment number, and then sequentially reads the configuration transformation data of each segment starting from the first structural segment. When applying the transformation data of each segment, first correct the starting coordinate of the current segment according to the cumulative translation vector, then construct a local rotation transformation matrix according to the cumulative rotation angle, perform a rotation transformation on the direction vector of the current segment, and recalculate the ending coordinate of this segment. Subsequently, use the current ending coordinate as the starting coordinate of the next segment to continue performing the geometric transformation speculation operation of the next segment. After each segment operation is completed, write the new starting and ending coordinates into the output list. Finally, the output forms a set of complete updated coordinate sequences of the structural segments.

[0031] Preferably, the corresponding position projection of the target bending data and the pre-processed bending shape in step S4, and the gap comparison analysis include: Extract the target feature control points of the target bending data; Extract the actual feature control points of the pre-processed bending shape; Construct a bidirectional projection matrix based on the target feature control points and the actual feature control points; Perform corresponding position projection according to the bidirectional projection matrix to obtain corresponding position projection data; Perform distance metric calculation based on the corresponding position projection data to generate a shape deviation vector; Perform directional analysis on the shape deviation vector to obtain the shape deviation trend; Identify the critical deviation points based on the form deviation trend; Determine the regionalized gap according to the critical deviation points; Integrate and statistically analyze the regionalized gap to determine the real-time processing gap.

[0032] In this embodiment, during the process of extracting the target feature control points of the target bending data, first, the DXF file of the standard target pattern is loaded, and all vector elements are read using the custom pattern parsing module. Only the polyline structure composed of LINE elements and ARC elements is retained. The connection nodes between consecutive line segments in the pattern are used as preliminary control point candidates. The node angle calculation module is called to construct an angle θ for the two line segments before and after each connection node. The nodes with the absolute value of the angle greater than 3 degrees are marked as high-fold angle feature points. This angle threshold is set to 3 degrees according to the actual detection accuracy requirements. The marked nodes are numbered in sequence and their two-dimensional coordinate values, corresponding line segment numbers, and local angle attributes are recorded. All the extracted control points are stored in the form of an ordered array. Each control point in the array contains four basic parameters: X coordinate, Y coordinate, fold angle, and line segment number. The output format of the array is a CSV file. All data uniformly represents the accuracy with three decimal places. The angle value is accurate to one decimal place after the decimal point, and the coordinate unit is uniformly millimeters. During the process of extracting the actual feature control points of the pre-processed bending shape, first, the DXF file of the pre-processed bending pattern generated by the bending configuration prediction module is loaded. All vector data is uniformly converted into a continuous vector sequence of structural segments. The start and end point coordinates of each structural segment are extracted by calling the structural segment parsing module. Then, the angle between adjacent structural segments is calculated. The connection points with an angle value greater than 3 degrees are extracted as actual feature control points. Each control point contains its structural segment number, connection coordinates, angle information, and sequence number in the overall structure. All control point information generates a two-dimensional control point sequence matrix. The number of rows of this matrix is equal to the total number of control points, and the number of columns is 4, which are the X coordinate, Y coordinate, angle, and segment number in sequence. During the process of constructing the bidirectional projection matrix based on the target feature control points and the actual feature control points, first, the coordinate vector sets are respectively extracted from the target control point CSV file and the actual control point JSON file. The target coordinate set is denoted as G = [g1, g2,..., gn], and the actual coordinate set is denoted as A = [a1, a2,..., an]. When the number of control points is not equal, the shorter set is interpolated to make up the quantity. The interpolation method is linear interpolation, and the interval is controlled within 2 millimeters. Subsequently, the projection matrix T and the inverse matrix T' are constructed by calling the least squares registration algorithm. The projection matrix T is used to map G to the space of A, and the inverse matrix T' is used to map A back to the space of G. The projection matrix T = [R|t] is composed of the rotation matrix R and the translation vector t. R is obtained by singular value decomposition (SVD). The convergence threshold of SVD is set to 10 -8, after the entire matrix is combined to form a 4×4 homogeneous matrix form, in the process of projecting to corresponding positions according to the bidirectional projection matrix to obtain projection data at corresponding positions, each target control point gi is represented in homogeneous coordinates as [gi_x, gi_y, 1], and successively left-multiplied by the projection matrix T to obtain the corresponding projection point Ti(gi) of each gi in the actual coordinate system. Similarly, each actual control point ai is also back-projected through the T' matrix to obtain Ti'(ai), and finally a set of bidirectional mapping data point groups are formed. Each group of data contains target points, actual points, target mapping points, and actual mapping points. All data point groups are written into a unified projection data structure. Each row in this structure contains the original point coordinates, mapping point coordinates, distance difference between points, and mapping offset angle. The structure is exported in CSV format for use in the subsequent shape deviation calculation stage. During the execution of the projection operation for each pair of points, the coordinate error shall not exceed 0.1 mm. When calculating the distance metric based on the projection data at corresponding positions and generating the shape deviation vector, read the coordinate information of each pair of points in the projection data structure, and call the coordinate distance calculation module to calculate the Euclidean distance between the target control point and the actual control point. Each distance is denoted as , where di is the Euclidean distance between the target control point and the actual control point (i.e., the absolute value of the shape deviation), in millimeters, representing the bending offset at this position, gi x is the coordinate value of the i-th target feature control point in the X-axis direction, gi y is the coordinate value of the i-th target feature control point in the Y-axis direction, ai x is the coordinate value of the i-th actual feature control point in the X-axis direction, ai y is the coordinate value of the i-th actual feature control point in the Y-axis direction. At the same time, record the structure segment number and point number corresponding to each di. All distance values are output in list form and constructed into a deviation vector D = [d1, d2,..., dn]. Each element is the spatial offset of the current pair of points. The offset value unit is millimeters. The vector data precision is set to 0.001 mm. The deviation vector is output as a CSV format file, where each row contains the pair of point numbers, the corresponding segment number, the target point coordinates, the actual point coordinates, and the Euclidean distance value. When performing a directional analysis on the shape deviation vector to obtain the shape deviation trend, construct a direction vector for the target point gi and the actual point ai corresponding to each deviation value di , where vi is the i-th offset direction vector, ai x is a two-dimensional vector representing the direction and magnitude of the offset, gi x is the coordinate value of the i-th actual control point on the X-axis, and is the coordinate value of the i-th target control point on the X-axis, ai y is the coordinate value of the i-th actual control point on the Y-axis, gi yis the coordinate value of the i-th target control point on the Y-axis. All direction vectors are represented by a vector array v, and each element of the array is a two-dimensional vector. Then, an angle clustering analysis operation is performed on all direction vectors. The K-Means clustering algorithm is used to divide all deviation directions into three direction classes. The maximum number of iterations is set to 300 times, and the convergence condition is that the angle change is less than 0.5 degrees. The clustering center direction is output in vector form. Finally, each deviation direction is marked with a trend direction ID, and the trend direction ID corresponds to the deviation value one by one to form a deviation trend structure set. Each element of this structure set contains the control point number, the offset direction vector, the offset distance value, and the direction classification label. During the process of identifying critical deviation points based on the shape deviation trend, the deviation trend structure set is traversed, and the points with an offset value greater than the average value plus two standard deviations in each direction class are counted as critical deviation points. The average value μ and the standard deviation σ are calculated by standard statistical methods, and the threshold is set to μ + 2σ. All points exceeding this threshold are recorded as critical deviation points. The critical deviation points are represented by a list of point numbers, and the corresponding area segment is retrieved by combining the structure segment number. The list is output in TXT format. At the same time, during the process of identifying critical deviation points, the segment numbers where three or more critical points are continuously aggregated are marked as high error area segments. Based on the critical deviation points, the regionalization gap is determined. During the process of integrating and statistically analyzing the regionalization gap to determine the real-time processing gap, all critical deviation point numbers are read and classified according to the structure segment number. The number of critical points and the average offset in each structure segment are counted. If the number of critical points in a certain segment is greater than 3 and the average offset is greater than 1.5 mm, then this segment is marked as a gap area segment. All gap area segments are used to construct a gap segment array. Each segment in the gap segment array contains the segment number, the number of critical points, the average offset value, the maximum offset direction, and the corresponding trend direction label. Finally, all gap segment information is integrated into a real-time gap report, and the report output format is an XML document, which includes the total number of gap segments, the maximum offset segment number, the most concentrated feature point segment number, the average offset statistical table, and the corresponding coordinate index table.

[0033] Preferably, the elastic springback detection of the variable contour data based on the real-time processing gap in step S4 includes: Determine the change gap direction based on the real-time processing gap; Perform material characteristic parameterization processing on the variable contour data to obtain material characteristic parameters; Perform collision simulation based on the material characteristic parameters, and perform elastic response analysis based on the simulated collision data to obtain material elastic response data; Quantify the springback influence value according to the real-time processing gap and the change gap direction; Derive the response strength of the material elastic response data based on the springback influence value to obtain the elastic springback strength; Determine the elastic springback direction based on the change gap direction; Fuse data on the elastic springback strength and elastic springback direction to generate elastic springback data for the bent part.

[0034] In this embodiment, in the process of determining the change gap direction based on the real-time processing gap, first, the offset vectors of each corresponding control point are extracted from the deviation calculation module. The offset vector is the spatial difference between the target point coordinates and the actual point coordinates. All the extracted offset vectors form a two-dimensional vector array. Subsequently, unitized direction extraction processing is performed on all the vectors in this array, that is, all the offset vectors are normalized and mapped onto a two-dimensional direction circle to form an angle sequence. The angle sequence divides the direction interval from 0 to 360 degrees, and 5 degrees is used as the minimum statistical unit. The quantity distribution of all offset directions in each angle interval is statistically analyzed and a direction histogram is constructed. The peak extraction operation is performed on the histogram to determine the concentrated area of the maximum offset direction. The direction interval with a direction concentration higher than 75% is defined as the dominant gap direction. This direction is represented in the form of a unit vector and retained in the subsequent springback direction derivation module. In the process of parametrically processing the material characteristics of the variable contour data, the contour structure of the bent part is imported into the material mapping module. Each bent segment is numbered in the structure topology. Each numbered structure segment corresponds to a set of material parameter sets. The material parameters include elastic modulus, yield strength, strain hardening index, Poisson's ratio, and density. Each material parameter is selected by the user in the database and injected into the model. The parameter setting range is as follows: the elastic modulus is 195 to 210 GPa, the yield strength is 200 to 350 MPa, the strain hardening index is between 0.2 and 0.3, the Poisson's ratio is a fixed value of 0.3, and the density is set to 7.85 g / cm³. All the parameters establish a mapping relationship according to the structure segment index. The mapping result is a parameter matrix in the form of a multi-dimensional tensor. The parameters of each structure segment are combined in a structure body form into a material characteristic parameter file. In the process of performing collision simulation based on the material characteristic parameters and performing elastic response analysis based on the simulated collision data, the JSON format material parameter file is imported into the finite element solution environment. The mesh is divided with each structure segment as a simulation sub-domain, and the mesh accuracy is set to 0.1 mm. The material model adopts the elastoplastic constitutive relationship. The boundary condition is set to one-way compression in the bending loading direction. The simulation adopts the explicit integration method. An equally spaced step-by-step displacement loading step is applied to each structure segment, and the total number of loading steps is 100 steps, with an increment of 0.02 mm. During the simulation process, record the maximum deformation and stress response values at the nodes of each structural segment. After the simulation, output the maximum node displacement and average springback recovery distance of each segment, and construct an elastic response dataset. Each element of this dataset contains the structural segment number, maximum stress value, maximum displacement value, and springback recovery ratio. When quantifying the springback influence value according to the real-time processing gap and the change gap direction, first read the maximum offset value and direction vector corresponding to each structural segment, then project the offset direction vector onto the dominant gap direction, and retain the paragraph data with an included angle less than 15 degrees as the effective springback area. Next, multiply the springback direction vector of each structural segment by its offset to obtain the springback influence base quantity. Divide this base quantity by the average length of the corresponding paragraph and multiply by the offset value to obtain the springback influence intensity scalar for each segment. All the intensity scalar values form the springback influence vector. The springback influence vector constructs tabular data indexed by the structural segment number. Each row contains the segment number, influence intensity value, offset direction, and unit direction vector. During the process of deriving the response intensity of the material elastic response based on the springback influence value, perform a segment number alignment operation on the springback influence vector and the previously generated elastic response dataset. After the alignment is completed, perform a dot product operation on each structural segment. Combine the springback influence intensity value with the elastic recovery ratio of the corresponding segment to obtain the comprehensive springback strength index. The comprehensive springback strength value represents the geometric change intensity caused by the actual springback under the current material conditions and loading state. This strength index is sorted in sequence according to the structural segment number, and the maximum value is used as the upper limit of the overall structural springback strength. The springback strengths of all paragraphs are stored in a list form. Each piece of data includes the segment number, direction vector, intensity value, and the corresponding coordinate index position. Finally, output a springback strength distribution map in the dimension of the structural segment. During the process of determining the elastic springback direction based on the change gap direction, the springback direction derivation module compares the vector angles between the initial direction vector and the dominant offset direction vector of each structural segment. If the angle is less than 20 degrees, record this direction as the effective springback direction. If the angle is greater than 20 degrees, replace it with the symmetric direction. After normalizing all the direction vectors, uniformly convert them to the unit vector format to generate a comparison table of structural segment numbers and direction vectors. Each record in this table contains the segment number, unit direction coordinates, and the main offset angle. During the process of fusing the elastic springback strength and elastic springback direction data to generate the elastic springback data of the bent part, merge the springback strength distribution map and the unit direction vector comparison table into a unified compensation data structure. In the data structure, match the direction and intensity value for each structural segment, and calculate the direction vector multiplied by the intensity value to obtain the final springback vector. Each springback vector represents the elastic recovery direction and recovery amplitude of the structural segment during the bending process. All the springback vectors constitute the elastic springback dataset of the bent part. The dataset is mainly indexed by the segment number, and each segment contains a two-dimensional recovery vector and the vector length value. Finally, export it as a standard DXF layer file for overlay drawing comparison.

[0035] Preferably, step S5 includes the following steps: Step S51: Perform principal component decomposition on the elastic springback data of the bent part to obtain the principal springback direction set; construct a springback vector field based on the principal springback direction set; Step S52: Perform direction transformation calculation on the elastic springback data of the bent part according to the springback vector field to generate an initial compensation vector; Step S53: Perform critical point compensation simulation based on the initial compensation vector to generate local compensation data; perform global coordinated compensation according to the local compensation data to generate bent springback compensation data; Step S54: Perform feature point mapping on the original drawing of the bent workpiece according to the bent springback compensation data to obtain feature corresponding data; Step S55: Convert the bent springback compensation data into a processing parameter correction value based on the feature corresponding data to obtain a gap compensation parameter.

[0036] In this embodiment, when performing principal component decomposition on the elastic springback data of the bent part, the principal component analysis method (Principal Component Analysis, abbreviated as PCA) is used to process the springback displacement matrix composed of three-dimensional coordinates. The size of the data matrix used is N×3, where N is the number of sampling points. Each sampling point includes the springback displacement values in the X, Y, and Z directions. The singular value decomposition (Singular Value Decomposition, abbreviated as SVD) algorithm is used to process this matrix, and the first three principal component direction vectors are selected as the principal springback direction set. The principal component contribution rate threshold is set to 95%. The functions svd(A) are called in the MATLAB platform to obtain the U, S, and V matrices. The first three vectors in the V matrix are used as the principal direction vectors V1, V2, and V3. A three-dimensional springback vector field model is constructed using these three groups of vectors. The scattered point interpolation method is used to interpolate the principal direction vectors at each sampling point to generate a complete vector distribution field. This distribution field is visually mapped in a uniform grid manner in three-dimensional space, and the grid density is set to a 1mm interval to ensure sufficient direction density sampling. When performing direction transformation calculation according to the springback vector field, the orthogonal projection is performed using the angle between the springback displacement vector of each sampling point and the direction vector in the corresponding springback vector field, and the original springback vector is decomposed into the principal direction component and the remaining direction component. The principal direction component is used to generate the initial compensation vector. Let the original springback vector be R and the principal direction vector be V, then the initial compensation vector , the dot function represents the dot product operation. The set of B vectors of all points constitutes the initial compensation field. After all vectors are unified and normalized, they are multiplied by the compensation weight factor λ, which is set to 0.85 to avoid over-compensation or mis-compensation phenomena. Finally, the initial compensation vector set of all sampling points is obtained. The matrix calculation process is completed using the NumPy library in the Python environment, and the obtained result is saved as the coordinate point cloud of the initial compensation field in the format of XYZ + B_x + B_y + B_z. When performing critical point compensation simulation based on the initial compensation vector, first use the structural finite element modeling method to divide the processed bent part into a refined mesh model with a mesh size set to 1mm×1mm. Use the Abaqus platform to perform static mechanical analysis on the bending area, and establish key monitoring points focusing on the groove corners, the starting points of edge bends, and the regions with sudden changes in material thickness. Introduce the initial compensation vector as the displacement boundary condition in the model, simulate different compensation directions and amplitudes for different regions, and export the coordinate change data of the key points after the simulation, which is called local compensation data. Use the moving least squares method to interpolate and expand these local compensation points to the entire part surface. The interpolation kernel function is selected as the Gaussian kernel function with a kernel width set to 3mm. According to this fully covered compensation vector field, perform global coordinated compensation on the surface of the overall bent part through weighted average, so that the compensation vectors of all points are relatively consistent with the local compensation field, thereby generating the final bending springback compensation data. When performing feature point mapping on the original drawing of the bent part according to the bending springback compensation data, first perform surface parameterization modeling on the CAD vector data in the original drawing, use B-Spline to represent the curve contour, and select all control points as the feature point set. Let the original control point coordinates be , the corresponding coordinates in the bending springback compensation data are , construct a mapping matrix T from to . Among them, T is a non-linear affine transformation matrix, and the least-square matching method (Least-Square Matching) is used to perform fitting optimization through the function lsqnonlin() in Matlab, with the goal of minimizing , after solving, the full-pattern feature point mapping set is obtained, and the mapping result is exported as a mapping data file containing the original and compensated coordinates of the control points. When converting the bending springback compensation data into the machining parameter correction value based on the feature correspondence data, the reverse modeling method is adopted, and the coordinate difference ΔP = P′ - P is converted into the bending die edge angle adjustment value Δθ and the die stroke correction amount Δd. The material yield limit is set to 245 MPa, the elastic modulus is 210 GPa, and the bending radius R is 1.5 times the material thickness t. Under the simulation conditions, the functional relationship between ΔP, Δθ, and Δd is established, and the empirical relationships Δθ = k1×ΔP_total and Δd = k2×ΔPtotal are obtained through the experimental calibration curve, where ΔPtotal is the modulus length of the compensation vector in each direction, k1 = 8.6° / mm, and k2 = 2.3 mm / mm. All the compensation data are converted into the machining die edge adjustment parameters and the die height adjustment amount accordingly, and a gap compensation parameter table is generated in the format of workpiece number, control point number, Δθ, and Δd, which is imported into the control module of the bending machine numerical control system for the real-time drawing module to read and dynamically refresh the contour coordinate set of the machined part.

[0037] Particularly importantly, step S52 includes: Construct a feature weight matrix according to the feature points of the springback vector field; Perform a non-linear coordinate transformation on the feature weight matrix to obtain a coordinate transformation field; Perform the shortest path planning on the coordinate transformation field to generate transformation path data; Determine the compensation vector set based on the transformation path data; perform clustering analysis based on the compensation vector set to obtain the compensation feature dimension; Reconstruct the compensation vector based on the compensation feature dimension to generate an initial compensation vector.

[0038] In this embodiment, index processing is performed on the feature points in the rebound vector field. First, the spatial coordinates of each feature point are classified according to the main rebound direction, and the position and rebound vector of each feature point in the three-dimensional space are calculated by using the direction and position differences of the rebound data. Taking the length and direction of the rebound vector of each feature point as input, a feature weight matrix is constructed by using the local weighted method. Specifically, each weight in the weight matrix is calculated as a function of the Euclidean distance between two points. Assuming that the adjacency threshold is 2.5 mm, the local density weighting coefficient is defined as w = exp(-d² / (2*1.2²)), where d is the Euclidean distance between two points, and the σ value is taken as 1.2 mm, to obtain a matrix of size n×n, where n is the number of feature points. After obtaining the feature weight matrix, non-linear coordinate transformation is performed, and the Isomap method in manifold learning is selected for dimensionality reduction. By selecting the neighborhood radius of each point as 3 mm and setting the adjacency boundary as 8, the shortest path distance between points is calculated, so as to map the feature points from the three-dimensional space to the two-dimensional plane. Through this method, the coordinates of the feature points are simplified, and the position relationship of each feature point in the output two-dimensional coordinate space maintains its original local structure, generating a coordinate transformation field, and obtaining a coordinate matrix of M×2, where M is the number of feature points after mapping. On the transformed coordinate space, the path planning method based on the A* algorithm is used to determine the shortest path. The heuristic function is set as the Manhattan distance h(n)=|x - x t | + |y - y t |, where (x t , y t ) are the coordinates of the target point, and the path cost function g(n) is the cumulative cost from the starting point to the current node. During the path planning process, in order to prevent excessive backtracking, the minimum weight threshold of the path node is set as 0.6. Finally, a set of transformed path data is generated. Each path point contains its two-dimensional coordinates and the corresponding three-dimensional compensation offset. Based on the compensation offset values in the path data, a compensation vector set is constructed. Each compensation vector vᵢ = (Δxᵢ, Δyᵢ, Δzᵢ) is calculated from the difference between the original position and the target position of the transformed path point. Subsequently, clustering analysis is performed on these compensation vectors. The k-means clustering algorithm is used to classify the compensation vectors according to their modulus length and direction. The initial number of clustering centers is set as 6, and the Euclidean distance is used to measure the similarity between vectors. The clustering results will be used as the basis for the compensation feature dimension vectors in the subsequent steps. After the clustering analysis is completed, the centroid of each class is extracted as the compensation feature dimension vector set. These feature dimension vectors will be used to reconstruct the compensation vectors. The reconstruction of the compensation vectors is completed by linear interpolation. The compensation vectors of each class are weighted and averaged to obtain the standard initial compensation vector V0 = {v 01 , v 02 ,..., v 0m}, and ensure that the modulus of all vectors is between 0 and 5 millimeters, and the direction error does not exceed 2°.

[0039] Preferably, step S6 includes the following steps: Step S61: Perform geometric transformation on the variable contour data according to the gap compensation parameter to generate compensated contour parameters; construct a machining simulation environment based on the compensated contour parameters, where the environmental simulation accuracy is controlled at the 0.01 mm grid level, the temperature field fluctuation range is maintained within ±3°C, and the simulation step size is 0.1 s; Step S62: Perform secondary machining simulation on the variable contour data according to the machining simulation environment to generate secondary machining simulation data. During the simulation process, the loading speed range is controlled at 1 - 10 mm / s, the material elastic modulus is set within a floating range of ±5% of the actual material value, and the friction coefficient is limited to 0.1 - 0.3; Step S63: Determine the secondary machining evolution sequence based on the secondary machining simulation data; extract the evolved result shape of the workpiece according to the secondary machining evolution sequence; Step S64: Perform real-time shape gap comparison between the evolved result shape of the workpiece and the target bending data, and perform dynamic fine-tuning of the machining parameters according to the real-time comparison data.

[0040] In this embodiment, according to the positions and shapes of each point in the contour data, affine transformation processing is performed through set compensation parameters. The compensation parameters include the rotation angle, scaling factor, and offset. These compensation parameters are applied through matrix operations. The specific steps are as follows: For each data point, first perform position offset (translation), then adjust the size of the contour according to the compensation parameters (scaling), and finally adjust the direction of the contour according to the set rotation angle. After the compensated data undergoes geometric transformation, compensated contour parameters are generated. These compensated contour parameters are used as inputs for subsequent simulation environment construction. The accuracy control of the simulation environment is set at the 0.01 mm grid level, which means that the side length of each grid in the simulation process is 0.01 mm, accurate to the micron level. In addition, the fluctuation range of the temperature field in the simulation is controlled within ±3°C to ensure that the temperature change in the simulation environment conforms to the actual machining situation. The simulation step size is set at 0.1 s, and the time interval between each calculation step is 0.1 second. The compensated contour data is imported into the machining simulation environment, and the set simulation loading speed range is between 1 mm / s and 10 mm / s. This range adapts to the deformation characteristics under different machining scenarios. By setting the material elastic modulus within a ±5% floating range of the actual material value, the physical properties of the material are ensured to conform to the actual situation. For example, if the material is aluminum alloy, its elastic modulus can be set between 66.5 GPa and 73.5 GPa. The friction coefficient range is set between 0.1 and 0.3 to adapt to the material deformation behavior under different surface friction conditions. Under these conditions, secondary machining simulation is performed to obtain simulation data, and the deformation characteristics in each machining step are recorded. By analyzing the simulation data of each machining step, the deformation amount and stress distribution at each moment are extracted. Using time series analysis techniques, the order of morphological evolution at each moment is established. Each item in the evolution sequence reflects the changes experienced by the material during the machining process, including shape changes, stress distribution, and evolution of the material state. By processing these sequences, the evolution data at each moment during the machining process is obtained, forming a complete machining evolution sequence. The evolved result shape of the workpiece is extracted from the secondary machining evolution sequence. Through curve fitting and interpolation algorithms, the shape data at each moment is reconstructed to obtain the actual morphology of the workpiece at each moment during the entire machining process. The interpolation technique is used to fill in the discontinuous points to ensure the smoothness of the evolved result shape. By extracting the data points in the evolution sequence, the final workpiece morphology data is formed for comparison with the target bending data. When performing real-time shape gap comparison between the evolved result shape of the workpiece and the target bending data, a point cloud matching algorithm is used to compare the evolved shape of the workpiece with the target shape point by point. By calculating the Euclidean distance between the two, the gap value of each point is determined. If the shape gap of a certain part exceeds the set threshold (e.g., 0.05 mm), the machining parameters are dynamically adjusted. According to the size of the gap, the loading speed, machining force, or other key parameters are adjusted.

[0041] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be embraced within the present invention.

[0042] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A real-time drawing verification method for the workpiece drawing of a bending machine numerical control system, characterized in that, Including the following steps: Step S1: Obtain the original drawing of the bent workpiece; identify the geometric element features of the original drawing of the bent workpiece, and perform boundary feature cutting based on the geometric element features to generate feature region data; Based on the preset workpiece material data, calculate the load safety margin of the feature region data to obtain the safety threshold boundary; Step S2: Decompose the safety threshold boundary in the principal stress direction and construct a continuous stress flow field; Based on the continuous stress flow field, reconstruct the geometric constraints of the geometric element features and identify the variable contour data; Step S3: Obtain the target bending data; Based on the target bending data, plan the bending trajectory of the variable contour data to obtain the planned bending sequence; predict the pre-processed bending shape according to the planned bending sequence; Step S4: Project the target bending data and the pre-processed bending shape at corresponding positions, and perform gap comparison analysis to generate the real-time processing gap; based on the real-time processing gap, perform elastic springback detection on the variable contour data to obtain the elastic springback data of the bent workpiece; Step S5: Perform springback compensation based on the elastic springback data of the bent workpiece to generate the bending springback compensation data; map the compensation parameters to the original drawing of the bent workpiece according to the bending springback compensation data to obtain the gap compensation parameters; Step S6: Perform secondary processing simulation on the variable contour data according to the gap compensation parameters to generate secondary processing simulation data; perform real-time comparison based on the secondary processing simulation data and the target bending data, and perform fine-tuning of dynamic processing parameters according to the real-time comparison data.

2. The real-time drawing verification method for the workpiece drawing of the numerical control system of a bending machine according to claim 1, characterized in that, The identification of the geometric element features of the original drawing of the bent workpiece in Step S1 and the boundary feature cutting based on the geometric element features include: Identify the geometric element features of the original drawing of the bent workpiece; Construct a topological relationship according to the geometric element features, and draw a connection topology diagram based on the topological relationship. Among them, the minimum distance between nodes in the connection topology diagram shall not be less than 0.1 mm, and the maximum connectivity error in the topological path shall not exceed 0.05 mm; Identify the drawing boundary contour based on the connection topology diagram; Perform entity feature cutting on the drawing boundary contour, perform area cutting according to the minimum unit segmentation size of 1 mm², and control the aspect ratio of the cutting area within 1:5 to 5:1 to generate feature region data.

3. The real-time drawing verification method for the machining part drawing of the NC system of a bending machine according to claim 1, characterized in that, The calculation of the load safety margin of the feature region data based on the preset workpiece material data in Step S1 includes: Associate the feature region data with the material type based on the preset workpiece material data to generate material distribution data; Project the anisotropy index of the material distribution data, and construct a directional strength matrix based on the projected data; Map the material attributes of the feature region data according to the directional strength matrix to obtain a set of material parameters; Reconstruct the material microstructure according to the set of material parameters, and perform constitutive relationship conversion based on the material microstructure to generate the material stress tensor; Calculate the critical load distribution based on the material stress tensor, and perform safety margin limitation based on the critical load distribution to obtain the safety threshold boundary.

4. The real-time drawing verification method for the workpiece drawing of the numerical control system of the bending machine according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Identify the surface features of the safety threshold boundary, and control the feature control region based on the surface feature recognition; perform mesh generation based on the feature control region to obtain finite element meshes; Step S22: Decompose the principal stress direction based on the finite element network to obtain the principal stress direction vector; construct a continuous stress flow field according to the principal stress direction vector; Step S23: Perform density gradient mapping based on the continuous stress flow field, and extract stress hot zone data based on the density gradient; perform material density distribution analysis according to the stress hot zone data to obtain the material density distribution; Step S24: Identify the unit stress concentration points based on the material density distribution; perform local shape correction on the unit stress concentration points to obtain the corrected stress concentration points; Step S25: Adjust the stress distribution evenly for the geometric element features based on the corrected stress concentration points, and reconstruct the geometric constraints based on the stress distribution evenly adjusted data to generate adjustment limit conditions; Step S26: Perform unrestricted contour recognition on the geometric element features according to the adjustment limit conditions to obtain variable contour data.

5. The real-time drawing verification method for the workpiece drawing of the numerical control system of the bending machine according to claim 1, characterized in that The bend trajectory planning for the variable contour data based on the target bend data in Step S3 includes: Perform boundary constraint mapping on the target bend data to obtain boundary constraint data; Perform bend angle serialization processing based on the boundary constraint data to obtain angle serialization data; Perform multi-axis collaborative path mapping on the variable contour data according to the angle serialization data to generate a candidate bend trajectory; Perform bend simulation based on the candidate bend trajectory, and perform deformation field mapping based on the simulated bend data to generate deformation topology data; Detect the bend conflict structure in the deformation topology data; Avoid the conflict trajectory for the candidate bend trajectory based on the bend conflict structure, and perform bend action planning according to the conflict avoidance trajectory to obtain the planned bend sequence.

6. The real-time drawing verification method for the workpiece drawing of the CNC system of a bending machine according to claim 1, characterized in that The prediction of the pre-processed bend shape according to the planned bend sequence in Step S3 includes: Reconstruct the variable bend part structure according to the variable contour data; Perform path torque mapping on the variable bend part structure based on the planned bend sequence to obtain bend force guidance data; Perform displacement curvature folding on the bend force guidance data to generate configuration trend data; Speculate on the geometric changes of the variable bend part structure based on the configuration trend data to obtain the pre-processed bend shape.

7. The real-time drawing verification method for the machining part drawing of the numerical control system of the bending machine according to claim 1, characterized in that, The corresponding position projection of the target bend data and the pre-processed bend shape and the gap comparison analysis in Step S4 include: Extract the target feature control points of the target bend data; Extract the actual feature control points of the pre-processed bend shape; Construct a bidirectional projection matrix based on the target feature control points and the actual feature control points; Perform corresponding position projection according to the bidirectional projection matrix to obtain corresponding position projection data; Perform distance metric calculation based on the corresponding position projection data to generate a shape deviation vector; Perform directional analysis on the shape deviation vector to obtain the shape deviation trend; Identify the critical deviation points based on the form deviation trend; Determine the regionalized gap according to the critical deviation points; integrate and statistically analyze the regionalized gap to determine the real-time processing gap.

8. The real-time drawing verification method for the workpiece drawing of the CNC system of a bending machine according to claim 1, characterized in that, The elastic springback detection of the variable contour data based on the real-time processing gap in Step S4 includes: Determine the change gap direction based on the real-time processing gap; Perform material property parameterization on the variable contour data to obtain material property parameters; Conduct collision simulation based on the material property parameters, and perform elastic response analysis based on the simulated collision data to obtain material elastic response data; Quantify the springback influence value according to the real-time machining gap and the direction of the change gap; Derive the response strength of the material elastic response data based on the springback influence value to obtain the elastic springback strength; Determine the elastic springback direction based on the direction of the change gap; Perform data fusion on the elastic springback strength and the elastic springback direction to generate the elastic springback data of the bent part.

9. The real-time drawing verification method for the workpiece drawing of the numerical control system of a bending machine according to claim 1, wherein, Step S5 includes the following steps: Step S51: Perform principal component decomposition on the elastic springback data of the bent part to obtain the principal springback direction set; construct a springback vector field based on the principal springback direction set; Step S52: Perform direction transformation calculation on the elastic springback data of the bent part according to the springback vector field to generate an initial compensation vector; Step S53: Perform critical point compensation simulation based on the initial compensation vector to generate local compensation data; perform global coordinated compensation according to the local compensation data to generate bent springback compensation data; Step S54: Perform feature point mapping on the original drawing of the bent workpiece according to the bent springback compensation data to obtain feature corresponding data; Step S55: Convert the bent springback compensation data into a machining parameter correction value based on the feature corresponding data to obtain a gap compensation parameter.

10. The real-time drawing verification method for the workpiece drawing of the CNC system of a bending machine according to claim 1, characterized in that Step S6 includes the following steps: Step S61: Perform geometric transformation on the variable contour data according to the gap compensation parameter to generate compensated contour parameters; construct a machining simulation environment based on the compensated contour parameters, where the environmental simulation accuracy is controlled at the 0.01 mm grid level, the temperature field fluctuation range is maintained within ±3°C, and the simulation step size is 0.1 s; Step S62: Perform secondary machining simulation on the variable contour data according to the machining simulation environment to generate secondary machining simulation data, where the loading speed range during the simulation is controlled at 1 - 10 mm / s, the material elastic modulus is set within a ±5% floating range of the actual material value, and the friction coefficient is limited to 0.1 - 0.3; Step S63: Determine the secondary machining evolution sequence based on the secondary machining simulation data; extract the evolved result shape of the workpiece according to the secondary machining evolution sequence; Step S64: Perform real-time shape gap comparison between the evolved result shape of the workpiece and the target bending data, and perform dynamic adjustment of machining parameters according to the real-time comparison data.

Images