A plate shape control parameter optimization method based on digital twinning

Abstract

The present application relates to the field of simulation technology, in particular to a plate shape control parameter optimization method based on digital twinning, a micro-performance digital twinning body of the material is constructed, a physical neutral layer offset is calculated through virtual straightening, a non-uniform differential curvature target is derived and reverse iteration is performed, and global optimal parameters of the straightening machine are output. The virtual simulation of shear crack propagation of the digital twinning body is driven by using a fracture mechanics model, burr and edge collapse data are calculated, and the shear cutter angle and surface prestress parameters are adjusted. A geometric topography digital twinning body is constructed by fusing the laser point cloud and visual image of the production line, a defect topology fingerprint is generated, and the two digital twinning bodies are coupled by running the moving target point interception algorithm, and the optimal transverse trajectory and local reduction of the stress relief roller are planned. Through multi-physical field virtual simulation and closed-loop verification, the present application realizes the collaborative optimization of global straightening, dynamic shearing and local repair in the production of high-strength steel plates.

Description

Technical Field

[0001] This invention relates to the field of simulation technology, specifically to a method for optimizing plate shape control parameters based on digital twins. Background Technology

[0002] As automotive body panels develop towards ultra-wide and ultra-high strength, panel shape control has evolved into a complex modeling problem involving highly nonlinear material constitutives and multi-physics coupling.

[0003] Existing technologies for parameter design and optimization of such sheet materials often rely on simplified analytical models or empirical lookup tables, lacking accurate computer simulations of complex mechanical behaviors such as microscopic anisotropy and the Bauschinger effect. Especially in the realm of simulation technology, existing modeling methods struggle to accurately reproduce crack evolution and stress wave transmission during high-speed dynamic shearing in a virtual environment, failing to effectively predict shear burrs and edge collapse defects. Furthermore, when faced with micron-level local surface morphology defects, traditional methods lack path planning and simulation verification capabilities based on high-dimensional geometric topology features, making it impossible to calculate precise local stress release strategies before physical action occurs, thus failing to meet the stringent requirements of high-end sheet materials for overall stress balance and surface quality.

[0004] To address the technical challenge of collaborative optimization modeling of multi-dimensional process parameters, such as asymmetric straightening, dynamic shearing, and local stress release in the heterogeneous residual stress field of ultra-high strength steel, a plate shape control parameter optimization method based on digital twins is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for optimizing plate shape control parameters based on digital twins, which aims to achieve accurate optimization of process parameters through multi-physics virtual simulation.

[0006] A method for optimizing plate shape control parameters based on digital twins, comprising:

[0007] A digital twin of the material's microscopic properties is constructed and a virtual straightening simulation is performed. The offset of the physical neutral layer is calculated, and a non-uniform differential curvature target that can equivalently fit the residual stress tensor field is derived. Based on the target, reverse iteration is performed to output the global optimal parameter combination of the straightener's asymmetric bending roll force, roll cross angle, and overall reduction.

[0008] Using a fracture mechanics model to drive a digital twin of the material's microscopic properties for virtual simulation, the crack propagation process of the plate during shearing is simulated, and defect characteristic data including burr height and edge collapse are calculated. Based on this, the angle of the subsequent shearing blade and the surface prestress parameters of the previous process are adjusted.

[0009] By integrating laser point cloud data and visual images from the production line, a geometric digital twin is constructed and topological feature analysis is performed to generate a defect topological fingerprint containing the coordinates and energy level of the stress concentration area. Based on the defect topological fingerprint, a moving target interception algorithm is run, and the two digital twins are coupled to simulate the stress relief roller, virtual plate surface, and bottom support mechanism. The optimal lateral movement trajectory is planned and the local pressure is calculated to verify the local defect repair effect.

[0010] Preferably, the steps of constructing a digital twin of the material's microscopic properties and performing virtual straightening pre-simulation, calculating the offset of the physical neutral layer, and deriving the non-uniform differential curvature target of the equivalent fitting residual stress tensor field specifically include: real-time acquisition of signals from a multispectral electromagnetic sensor array, establishing a magnetostriction and stress statistical coupling model based on the material thickness direction, mapping and reconstructing the magnetoelastic detection data of the physical plate into a residual stress tensor field, and constructing a digital twin of the material's microscopic properties with microscopic anisotropic characteristics;

[0011] A reduced-order constitutive model incorporating isotropic hardening, kinematic hardening, and the Bauschinger effect is loaded into the digital twin of the material's microstructure properties. This model drives a virtual straightening machine to rapidly simulate the complete virtual straightening process involving multiple loading-unloading-reverse loading cycles. By tracking the dynamic drift trajectory of the neutral layer, the offset of the physical neutral layer is accurately calculated, and the target distribution of spatially non-uniform differential curvature that can maximize the offset of the residual stress tensor field is derived.

[0012] Preferably, the step of performing reverse iteration based on the target to output the globally optimal parameter combination of the asymmetric bending roll force, roll cross angle, and overall reduction of the straightener includes:

[0013] Using the non-uniform differential curvature objective as the inverse optimization constraint, a physical field proxy model containing the straightener roll gap geometry, contact nonlinearity, and dynamic response of the servo hydraulic system is constructed in the material microscopic property digital twin. An inverse iterative optimization algorithm based on sequential quadratic programming is run. By continuously adjusting the parameters of the asymmetric bending roll force vector of the upper working roll, the cross angle of the upper and lower roll systems, and the overall reduction, the deviation of the residual stress tensor field of the virtual plate exit section from the ideal straight state is gradually converged to within the preset tolerance. The parameter combination of the straightener asymmetric bending roll force, roll cross angle, and overall reduction that satisfies the global convergence solution is output.

[0014] Preferably, the step of using a fracture mechanics model to drive a digital twin of the material's microscopic properties for virtual simulation, simulating the crack propagation process of the plate during shearing, calculating defect feature data including burr height and edge collapse, and adjusting the angle of the subsequent shearing blade and the surface prestress parameters of the previous process accordingly specifically includes: loading a fracture mechanics model into the digital twin of the material's microscopic properties to perform high-speed shearing transient virtual simulation, simulating the crack initiation and propagation process of the plate during shearing, and extracting defect feature data such as burr height, edge collapse, and cross-sectional roughness in real time;

[0015] Based on the defect characteristic data, a correlation database between shearing quality and process parameters is established. Through feedback control strategy, the shear blade tilt angle, blade side clearance, and surface prestress loading parameters of subsequent shearing cycles are adjusted so that the burr height and edge collapse of the virtual shearing section converge to within a set threshold. The optimal combination of shear blade angle and surface prestress parameters is output. The parameter combination includes the upper shear blade tilt angle, the lower shear blade tilt angle, the shear blade side clearance, the prestress on the plate inlet side surface, the prestress on the plate outlet side surface, the lateral offset of the prestress loading roller group, and the prestress application advance.

[0016] Preferably, the step of fusing laser point cloud data and visual images from the production line to construct a geometrical digital twin and perform topological feature analysis to generate a defect topological fingerprint containing the coordinates and energy level of the stress concentration region specifically includes: real-time acquisition of three-dimensional point cloud data and texture images of the board surface, mapping to generate a real-time geometrical digital twin of the board with geometric coordinates and surface features; performing surface reconstruction and noise filtering on the geometrical digital twin, and then performing multi-scale geometric deviation analysis to identify and classify defects including edge waves, middle waves, local concavity and convexity, and warping; based on the geometric curvature abrupt change and surface texture anomaly of the defect region, calculating the stress concentration equivalent and energy level corresponding to each defect to obtain a defect topological fingerprint containing the defect spatial coordinates, defect type, boundary contour, and stress concentration energy level.

[0017] Preferably, the step of running a moving target interception algorithm based on the defect topology fingerprint, coupling two digital twins, simulating stress relief rollers, virtual board surface and bottom support mechanism, planning the optimal lateral movement trajectory and calculating local pressure, and verifying the local defect repair effect includes: according to the defect topology fingerprint, treating stress concentration defect areas with energy levels exceeding the threshold as dynamic moving target points that move with the longitudinal movement of the board.

[0018] Using the real-time running speed of the plate as input, the future position sequence of each target point in the coordinate system of the straightener is predicted in real time; using the stress relief roller as a motion actuator, a time-optimal moving target interception algorithm is run to generate the coupled motion trajectory of the stress relief roller in the lateral and vertical directions, so that the roller surface prioritizes tracking and covering the current target point with the highest energy level at each moment.

[0019] The rolling contact process between the stress relief roller, the virtual plate surface, and the bottom support mechanism is simulated in a digital twin that is coupled in real time with the digital twin of the material's microscopic properties and the digital twin of its geometric morphology. The local plastic deformation and stress relief in the contact area are calculated. Based on the stress relief effect obtained from the simulation, the local reduction and holding time of each interception point are adjusted in reverse iteratively until the residual stress concentration energy level in the defect area drops below the set threshold.

[0020] After traversing all target points, a complete lateral motion trajectory curve of the stress relief roller and a corresponding local pressure reduction timing parameter table are obtained. Closed-loop verification is then performed in the coupled digital twin to confirm that the target local defects meet the repair criteria.

[0021] Preferably, the method further includes: establishing a real-time comparison mechanism between physical production line quality data and digital twin prediction data; using time series analysis to identify deviations caused by equipment wear or thermal drift; updating the baseline parameters of the digital twin based on the deviation results; calculating global parameter bias values; and realizing adaptive tracking and parameter correction of the aging state of physical equipment by the digital twin model.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] 1. By constructing a digital twin of the material's microscopic properties, the straightening loading-unloading process is simulated in a virtual environment to accurately calculate the physical neutral layer drift caused by the Bauschinger effect and derive the non-uniform differential curvature target. This mechanism-based inverse iterative algorithm can output a globally optimal solution that includes asymmetric bending roll force, roll cross angle, and overall reduction, effectively solving the macroscopic shape defects in high-strength steel caused by springback prediction deviations, and significantly improving the physical self-consistency of parameter settings and the first-piece yield.

[0024] 2. A fracture mechanics model is introduced for transient virtual simulation of shearing. Crack propagation behavior is pre-simulated in digital space, and burr and edge collapse characteristics are quantified, establishing a correlation between quality and parameters. Based on this, the system adjusts the shear blade angle and the surface prestress parameters from the previous process, pre-setting a stress state favorable to cross-sectional quality within the plate. This strategy effectively suppresses the generation of dynamic shearing defects and utilizes feedforward compensation to achieve synergistic efficiency and closed-loop control of the "straightening-shearing" process.

[0025] 3. By fusing multi-source sensing data to construct a digital twin of the geometric topography, and utilizing topological analysis to transform complex surface defects into quantified "topological fingerprints" and energy levels, a moving target interception algorithm is run to precisely plan the lateral trajectory and local reduction of the stress relief roller in a virtual environment, simulating the point-to-point elimination process for high-potential wells. This simulation verification method ensures the thorough repair of localized micro-waves and residual stress concentrations, significantly improving the surface flatness of the sheet material. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the steps of a plate shape control parameter optimization method based on digital twins according to the present invention.

[0027] Figure 2 This is a schematic diagram of the coupling process of two digital twins in this invention;

[0028] Figure 3 This is a schematic diagram illustrating the process of performing defect repair using a geometric digital twin according to the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figures 1 to 3 This invention provides a method for optimizing plate shape control parameters based on digital twins, referring to... Figure 1 The following is a flowchart of the steps of this invention, and the technical solution is as follows:

[0031] A method for optimizing plate shape control parameters based on digital twins, comprising:

[0032] A digital twin of the material's microscopic properties is constructed and a virtual straightening simulation is performed. The offset of the physical neutral layer is calculated, and a non-uniform differential curvature target that can equivalently fit the residual stress tensor field is derived. Based on the target, reverse iteration is performed to output the global optimal parameter combination of the straightener's asymmetric bending roll force, roll cross angle, and overall reduction.

[0033] Using a fracture mechanics model to drive a digital twin of the material's microscopic properties for virtual simulation, the crack propagation process of the plate during shearing is simulated, and defect characteristic data including burr height and edge collapse are calculated. Based on this, the angle of the subsequent shearing blade and the surface prestress parameters of the previous process are adjusted.

[0034] By integrating laser point cloud data and visual images from the production line, a geometric digital twin is constructed and topological feature analysis is performed to generate a defect topological fingerprint containing the coordinates and energy level of the stress concentration area. Based on the defect topological fingerprint, a moving target interception algorithm is run, and the two digital twins are coupled to simulate the stress relief roller, virtual plate surface, and bottom support mechanism. The optimal lateral movement trajectory is planned and the local pressure is calculated to verify the local defect repair effect.

[0035] Example 1

[0036] This embodiment is mainly applied to the finishing and shape control process of high-strength steel plate production line, especially for the shape straightening scenario of 1180 MPa duplex steel plate. It elaborates on a method for optimizing shape control parameters based on digital twins. By constructing a high-precision digital twin of microscopic performance and a digital twin of geometric shape, digital closed-loop control of straightening, shearing and local repair processes is realized.

[0037] First, at the inspection station before the sheet enters the straightening machine, the signals of the multispectral electromagnetic sensor array are collected in real time. Specifically, the system activates the 128-channel multispectral electromagnetic sensor array arranged laterally on the upper and lower surfaces of the sheet, and sets the frequency scanning range of the excitation magnetic field to 10.50 Hz to 1000.50 Hz. When the sheet passes through the inspection area at a running speed of 2.00 m / s, the sensor array simultaneously collects Barkhausen noise signals and hysteresis loop data at a sampling rate of 5000.00 Hz.

[0038] Furthermore, a magnetostriction and stress statistical coupling model based on the material thickness direction is established; the system performs frequency domain decomposition on the collected multispectral signals, extracts the energy values ​​of characteristic frequency bands, and combines the skin effect depth in the thickness direction of the plate to analyze the magnetic domain motion characteristics of different depth layers.

[0039] For establishing the magneto-force coupling model, the system adopts a weighted mapping framework based on crystal texture. Specifically, the system reconstructs the stress through the coupling model based on the Barkhausen noise amplitude matrix of the multi-sensor probes acquired in real time at multiple excitation frequencies.

[0040] The principal stress components along the rolling direction and in the transverse direction are characterized as weighted summation functions of noise amplitude, frequency weighting coefficient, and skin effect depth, respectively; in particular, the stress calculation term along the rolling direction also includes a coupling term related to the normal stress, which is adjusted by the Bauschinger effect coupling coefficient.

[0041] The frequency weighting coefficient is determined based on the crystal texture orientation distribution function measured by an X-ray diffractometer; the skin effect depth is jointly determined by the material resistivity, excitation angular frequency, and permeability (including vacuum permeability and relative permeability); and the Bauschinger effect coupling coefficient is obtained by reverse calibration based on the material's historical strain state.

[0042] Based on the above coupling relationship, the multi-layer sampling data of each grid node along the thickness direction is mapped in layers to ensure that the virtual stress tensor field can accurately reflect the stress gradient change from the surface to the core of the physical plate.

[0043] Inversion from surface signals to internal stress fields:

[0044] The skin effect-based depth layer detection utilizes the physical property that high-frequency electromagnetic waves penetrate shallowly while low-frequency electromagnetic waves penetrate deeply to control the sensor to perform multi-frequency scanning of the board material. The system defines the signal under high-frequency excitation as the stress response of the very surface layer of the board material, and the signal under low-frequency excitation as the cumulative stress response of the entire thickness, including the surface layer to the core, thereby establishing a correspondence between signal frequency and detection depth.

[0045] Deep signal analysis based on interlayer difference employs a layer-by-layer peeling calculation logic. The cumulative signal across the entire thickness obtained from low-frequency detection is considered a sum, from which the contribution component of the surface signal obtained from high-frequency detection is subtracted. Through this differential calculation, surface interference is logically eliminated, separating independent stress response characteristics belonging only to the deep or core layer of the plate, transforming the mixed magnetic signal into discrete stress values ​​distributed along the thickness direction.

[0046] Based on the physical constraint correction of mechanical self-equilibrium, and according to the rigid mechanics law that the resultant force of internal forces and the resultant bending moment of the plate in a free state are zero, a verification model is established. The system uses the surface stress data with extremely high measurement accuracy as a benchmark, and combines the above-mentioned equilibrium law to perform trend locking and numerical correction on the calculated deep stress data (i.e., if the surface is under tension, the core must be under compression), forcibly ensuring that the reconstructed three-dimensional stress field satisfies self-consistent equilibrium in mechanics, thereby obtaining an accurate internal stress distribution without damaging the plate.

[0047] Specifically, the magnetoelastic test data of the physical sheet material is mapped and reconstructed into a residual stress tensor field. The system generates a finite element mesh in virtual space with the same dimensions as the physical sheet material, and the mesh element size is set to 5.00 mm x 5.00 mm. Based on the magnetic-force coupling calibration database for the current sheet material grade (such as 1180 MPa duplex steel), which is established by linking static tensile test and magnetoelastic test, the system includes the magnetic-stress conversion coefficients of the material in the stress range of 0-800 MPa. The system maps the analytical micro-stress components along the rolling direction, transverse direction, and normal direction to each corresponding mesh node. Thus, a digital twin of the material's micro-properties with micro-anisotropic characteristics is constructed. The initial maximum residual compressive stress value recorded inside the digital twin is 450.25 MPa, and the maximum residual tensile stress value is 380.60 MPa.

[0048] Furthermore, a reduced-order constitutive model is loaded into the digital twin of the material's microscopic properties; specifically, the constitutive model sets the Young's modulus of the material to 206.00 GPa and the Poisson's ratio to 0.29; the model introduces a hybrid hardening criterion that includes isotropic hardening, kinematic hardening and Bauschinger effect, wherein the initial yield strength is set to 850.50 MPa, the saturation value of the back stress evolution is 210.20 MPa, and the nonlinear hardening parameter is 12.50.

[0049] Specifically, the system rapidly simulates the complete virtual straightening process of driving the virtual straightener through multiple loading-unloading-reverse loading cycles. The system constructs a virtual straightener model containing 13 straightening rollers in a digital twin environment, with the roller diameter set to 160.00 mm and the roller spacing set to 180.00 mm. During the simulation, the virtual sheet material sequentially undergoes a high curvature bend at the inlet side and a low curvature reverse bend at the outlet side.

[0050] Furthermore, the physical neutral layer offset is accurately calculated by tracking the dynamic drift trajectory of the neutral layer;

[0051] The following neutral layer positioning criteria are adopted: For each cross-section of the virtual plate, the node locations that satisfy the following conditions are searched in the finite element mesh. :

[0052] ;

[0053] in Let h represent the normal stress distribution along the rolling direction, h be the plate thickness, and z be the coordinate along the thickness direction. The system determines this distribution using a bisection search within each time step. It records the trajectory of the entire process of multiple loading-unloading-reverse loading.

[0054] Define the physical neutral layer offset as: Where n is the total number of passes in the virtual straightening process. The position of the neutral layer in the i-th pass.

[0055] The system calculates the stress integral value in real time for each pass based on a time step of 0.002 seconds. When the absolute value of this integral first drops below 20% of the value of the previous time step, it determines the stress at that moment. This refers to the location of the neutral layer in this passage;

[0056] Based on the aforementioned neutral layer offset, the system constructs a non-uniform differential curvature target. For the i-th differential unit along the length direction of the plate, its ideal reverse bending curvature is... It should meet the following requirements:

[0057] ;

[0058] in This represents the residual stress distribution within the element after virtual straightening, where E is Young's modulus. The curvature value is the unit length; this curvature value ensures that during the subsequent physical straightening process, the elastic curvature generated by the loading cancels out the plastic inversion generated by the release of the initial residual stress, thus achieving the final straightening effect.

[0059] The system monitors the geometric position of zero tangential stress on the cross-section of the plate in real time with a time step of 0.002 seconds. Simulation data shows that during the second bending pass, the physical neutral layer shifted 0.55 mm towards the inward arc. During the eighth reverse loading pass, the position of the physical neutral layer changed, and the distance relative to the geometric center plane decreased to 0.18 mm. The system calculates the final physical neutral layer shift as 0.42 mm by weighted averaging of the instantaneous shifts throughout the process. (According to materials science theory, the neutral layer drift caused by the Bauschinger effect is usually positively correlated with the work hardening index, the number of cyclic bending passes, and the plate thickness. For 1180 MPa duplex steel, a shift of 0.42 mm during 13 straightening passes is within a reasonable range from a materials science perspective.)

[0060] Based on the calculated physical neutral layer offset and residual stress tensor field, the system derives a target distribution of spatially non-uniform differential curvature that can maximize the offset of the residual stress tensor field. Specifically, the system discretizes the plate into 200 differential units along its length. For the residual stress distribution within each unit, combined with a neutral layer correction value of 0.42 mm, the system calculates the ideal reverse bending curvature required for the unit to achieve zero stress. For example, the target differential curvature derived in the head region of the plate is 0.028 cdm, and the target differential curvature derived in the middle region of the plate is 0.015 cdm.

[0061] Furthermore, based on the stated objective, a reverse iteration is performed to output the globally optimal parameter combination for the straightener. The system uses the derived non-uniform differential curvature objective as a constraint condition and inputs it into the reverse solution algorithm for the straightening process. The iteration convergence threshold is set to 0.05. After 150 iterations, the system outputs the following globally optimal parameter combination: the asymmetric bending roll force on the inlet side is set to 2200.50 kN, and the asymmetric bending roll force on the outlet side is set to 1150.25 kN; the roll crossing angle is set to 0.66 degrees; and the overall reduction is set to decrease linearly from 12.50 mm at the inlet to 0.85 mm at the outlet.

[0062] Furthermore, a fracture mechanics model is used to drive the virtual simulation of the material's microscopic properties using a digital twin. After virtual straightening is completed, the system simulates the crack propagation process of the plate at the moment of shearing. The gap of the virtual shear blade is set to 8.00% of the plate thickness, and the shearing speed is 1.20 meters per second.

[0063] Specifically, the defect feature data, including burr height and collapse amount, were calculated; the simulation results show that, under the current stress state, the burr height at the virtual plate fracture is 0.08 mm and the collapse amount is 0.15 mm.

[0064] Furthermore, the angle of the subsequent shearing blade and the surface prestress parameters of the previous process are adjusted accordingly. Based on the defect characteristic data, the system adjusts the shearing blade angle of the next shearing action on the physical production line to 1.50 degrees and reduces the surface tension roller prestress value in the previous process by 15.50 MPa to suppress the unstable propagation of shear cracks.

[0065] Furthermore, by integrating laser point cloud data and visual images from the production line, a digital twin of geometric shape is constructed; the system calls a high-precision 3D laser profilometer and an 8K line array camera to obtain depth and texture information of the board surface and generate a geometric model with a point cloud density of 10 points per square millimeter.

[0066] Specifically, topological feature analysis was performed to generate a defect topological fingerprint containing the coordinates and energy level of the stress concentration area. The analysis results identified a wave defect located 1500.00 mm from the head on the left edge of the plate. The corresponding stress concentration energy level was marked as level 5, and the coordinates were accurate to (1500.20, 20.50).

[0067] Furthermore, based on the defect topological fingerprint, a moving target interception algorithm is run to couple two digital twins; the system maps the stress data of the material microstructure digital twin onto the geometric shape digital twin to simulate the contact process of the stress relief roller, the virtual plate surface, and the bottom support mechanism.

[0068] Specifically, the optimal lateral motion trajectory is planned and the local reduction amount is calculated; the system plans the stress relief roller to move to the defect coordinates at a speed of 0.50 m / s, sets the local reduction amount to 1.25 mm, and rolls along the defect topology; finally, the local defect repair effect is verified in the digital twin, and after confirming that the flatness error of the repaired virtual board surface is reduced to within 0.30 mm, the motion trajectory command is sent to the physical actuator.

[0069] By establishing a magnetostrictive coupling model and loading a higher-order constitutive model containing the Bauschinger effect, the invisible heterogeneous residual stress field inside high-strength steel and its neutral layer drift trajectory can be accurately restored. This makes the derivation of the non-uniform differential curvature target based on the real micromechanical behavior of the material, and improves the accuracy of predicting the springback characteristics of high-strength steel.

[0070] Based on the previous step which derived the target distribution of spatial non-uniform differential curvature, this step aims to accurately solve for the optimal control command of the actuator through a high-precision physical field proxy model and optimization algorithm.

[0071] The steps for combining the globally optimal parameters of the asymmetric bending roll force, roll crossing angle, and overall reduction of the output straightener described in this embodiment specifically include the following process:

[0072] First, using the non-uniform differential curvature objective as the inverse optimization constraint, a physical field proxy model is constructed in the digital twin of the material's microscopic properties. Specifically, this model is no longer a single mechanical model, but integrates the straightener roll gap geometry model, the contact nonlinear model, and the dynamic response model of the servo hydraulic system.

[0073] Furthermore, when constructing the roll gap geometric model, the system discretizes the spatial position of the roll system of the virtual straightener and sets the discretization accuracy of the roll surface profile to 0.01 mm to accurately describe the elastic flattening and contact geometry relationship between the work roll, intermediate roll and support roll; when constructing the contact nonlinear model, the Coulomb friction coefficient between the plate and the roll surface is set to 0.12, and Hertz contact theory is introduced to calculate the local contact stress.

[0074] Specifically, when constructing the dynamic response model of the servo hydraulic system, the system is based on the characteristics of the hydraulic cylinder of the physical straightener, the bulk modulus of the hydraulic oil is 1.40 gigapascals, the frequency response characteristic of the servo valve is set to 45.00 Hz, and a system lag time constant of 0.05 seconds is introduced; through the coupling of the above three sub-models, a reduced-order physical field proxy model that can quickly predict the deformation state of the plate under different process parameters is constructed.

[0075] Furthermore, the inverse iterative optimization algorithm based on sequential quadratic programming (SQP) is run, and the steps include:

[0076] An objective function is established to minimize the residual stress. This function consists of two parts: first, the average equivalent stress of all finite element elements at the virtual plate outlet section, which characterizes the residual stress level; and second, the total deviation of each decision variable from its initial value, weighted by a balance factor to ensure the stability of the optimization results.

[0077] The system defines boundary conditions for optimization, including the physical equipment travel range of asymmetric bending roll force, roll crossing angle, and overall reduction. It also sets maximum allowable thresholds for sheet exit straightness and exit residual stress as quality constraints. Decision variables are initialized using production experience values, and initial parameters for the trust region radius and line search step size are set. Within each iteration cycle, the system updates the decision variables by calculating the gradient vector and Hessian matrix of the objective function (using the BFGS correction algorithm) and solving a quadratic programming subproblem. The convergence of the objective function is monitored in real time. The algorithm terminates when the change in the objective function value between two consecutive iterations is less than a preset small threshold, or when the number of iterations exceeds a set upper limit, and outputs the current optimal parameter combination.

[0078] The system sets the "deviation between the residual stress tensor field of the virtual plate exit section and the ideal flat state (i.e., zero stress state)" as the objective function, and sets the limit capabilities of the physical equipment (such as the maximum bending roller force of 3000.00 kN and the maximum reduction of 20.00 mm) as the boundary constraints.

[0079] A weighted mapping algorithm combining crystal orientation distribution function (ODF) is used to solve the problem of accurate reconstruction of one-dimensional magnetoelastic signal into residual stress tensor field in strongly anisotropic materials.

[0080] When constructing a digital twin of a material's microstructure properties, linear magnetoelastic coefficients should not be used for conversion alone. It is recommended to introduce crystal texture data measured by X-ray diffraction to construct a nonlinear coupling matrix that includes magnetic domain wall movement resistance and lattice distortion energy. This matrix decouples the scalar voltage signal collected by the sensor array into principal stress components along the rolling direction and transverse direction. Combined with skin effect depth correction, the two-dimensional surface data is voxelized and extrapolated into a three-dimensional tensor field with gradient along the thickness direction, thus serving as the initial stress state boundary. This overcomes the stress detection distortion problem caused by neglecting the anisotropy of the microstructure of high-strength steel in existing technologies, significantly improving the characterization accuracy of the digital twin for the distribution of "dark energy" inside the material. It also ensures that the subsequent virtual straightening pre-simulation is based on a physically real energy benchmark, thereby greatly improving the reliability of the neutral layer offset calculation.

[0081] Specifically, the system initializes a set of initial parameters and then enters an iterative loop; in each iteration, the algorithm automatically adjusts variables in three dimensions:

[0082] Asymmetric bending force vector of upper working roll: Adjust the hydraulic cylinder pressure on the operating side and the transmission side respectively to deal with the asymmetric internal stress in the transverse direction of the plate.

[0083] Upper and lower roller cross angle: Adjust the slight angle between the axes of the upper and lower working rollers in the horizontal plane to control the difference in elongation between the edge and the center of the sheet material;

[0084] Overall reduction parameter: Adjust the inlet and outlet roll gap opening.

[0085] Furthermore, by continuously adjusting the above parameters, the proxy model is driven to perform calculations, so that the deviation between the residual stress tensor field of the virtual plate outlet section and the ideal straight state gradually converges to within the preset tolerance; the convergence criterion set by the system is: the maximum residual stress deviation of the section is less than 5.50 MPa, and the straightness I unit is less than 0.50.

[0086] Specifically, after 45 iterations and 0.85 seconds, the gradient of the objective function drops to 1.00 x 10^-4, satisfying the convergence condition. At this point, the system locks the parameter values ​​of the last iteration and outputs the globally optimal parameter combination that satisfies the globally convergent solution.

[0087] The specific data are as follows: The asymmetric bending roll force of the straightener output by the system is 1850.50 kN applied on the drive side and 1920.25 kN applied on the operating side, exhibiting obvious asymmetric loading characteristics to counteract the wedge stress of the plate; the roll cross angle output by the system is 0.66 degrees, which precisely matches the transverse shape defect characteristics of the current plate and effectively eliminates edge waviness; the overall reduction output by the system is: the inlet roll gap is set to 12.50 mm and the outlet roll gap is set to 0.80 mm, forming a consistent transition from large deformation to micro-shaping; finally, this parameter combination is directly transmitted to the basic automation system of the production line (L1 level) to complete the precise setting of the physical straightener.

[0088] The physical field proxy model is mainly used for preliminary optimization and rapid prediction of process parameters. When process conditions such as plate thickness, width, material, and running speed change, the proxy model remains effective only if the change in these conditions does not exceed ±20% of the design working point. If this range is exceeded, the system should trigger the benchmark model recalibration procedure.

[0089] By utilizing the physical field proxy model and the sequential quadratic programming (SQP) algorithm, this invention transforms the complex nonlinear parameter optimization problem into a mathematically convergent calculation, achieving a rapid balance between equipment stiffness, contact nonlinearity, and process objectives. This ensures that the output asymmetric bending roller and cross angle parameters are not only theoretically optimal, but also fully consider the execution capability and dynamic response limitations of the physical equipment.

[0090] Furthermore, the specific execution flow of the steps for adjusting the shear blade angle and surface prestress parameters is as follows:

[0091] First, a fracture mechanics model is loaded into a digital twin of the material's microstructure properties;

[0092] Specifically, the system adopts the GTN microscopic damage model, in which key indicators such as the pore nucleation volume fraction, critical pore volume fraction, and isotropic hardening parameters are all obtained by backfit calibration based on the specific material of the plate.

[0093] For high-strength steel (such as 1180 MPa duplex steel), the system sets specific parameter thresholds based on the results of static tensile and shear tests, determines the initial strain node at which the material begins to nucleate pores, and the critical value of the pore volume fraction when the material reaches the failure state, and calibrates the correlation parameters used to calculate the relationship between effective stress and pore influence factor.

[0094] The system locally refines the mesh in the predetermined shearing region of the virtual sheet material, refining the characteristic length of the hexahedral mesh unit; and arranges a sufficient number of mesh layers along the thickness direction in the shearing region to capture gradient changes within the shear band.

[0095] In the virtual shearing simulation, the system simulates the pressing and shearing process of the cutting tool on the sheet metal, setting a time step in microseconds. By monitoring the evolution of the pore volume fraction of each unit inside the virtual sheet metal in real time, when the monitored value first reaches the set critical failure value, the unit is determined to have failed and removed from the model. This process continues until a complete crack channel penetrating the thickness direction of the sheet metal is established, at which point the simulation stops. The system automatically measures the virtual fracture morphology after the simulation is completed and calculates the following defect characteristic data: burr height is the distance measured from the surface of the virtual sheet metal along the vertical direction to the outer edge of the fracture contour; edge collapse is the difference in the change of the sheet metal width on both sides of the cutting line before and after virtual shearing; and surface roughness is calculated by taking the arithmetic mean deviation of the virtual crack contour line.

[0096] Based on the aforementioned defect characteristic data, the system establishes a multidimensional relational database and uses multiple linear regression or Gaussian process regression to fit the following mapping relationship:

[0097] A functional relationship between burr height and process parameters was established, which was characterized as a weighted combination of shear blade angle, blade side clearance, and prestress on the plate surface. A functional relationship between edge collapse and process parameters and historical state was established, which was characterized as a weighted combination of shear blade angle, blade side clearance, surface prestress, and historical strain of the plate from the straightening process. The weight coefficients of each item in the above functional relationship were calculated by least squares fitting of historical production data and multiple simulation results.

[0098] The parameters of subsequent shearing cycles are adjusted through a feedback control strategy. The system adopts a gradient descent method under constraints to solve for the optimal combination of shearing blade angle, blade side clearance and surface prestress under the multi-objective constraints of minimizing burr height and edge collapse.

[0099] Specifically, the system calls the GTN micro-damage model, sets the pore nucleation volume fraction to 0.04, and the critical pore volume fraction to 0.15; in order to capture the micro-crack features at the moment of shearing, the system performs local mesh refinement in the predetermined shearing area of ​​the virtual plate, refining the feature length of the hexahedral mesh unit to 0.05 mm.

[0100] Furthermore, a high-speed shear transient virtual simulation was conducted to simulate the crack initiation and propagation process of the plate during shearing. The virtual shearing speed was set to 1.50 meters per second, and the simulation time step was set to 2.00 microseconds. After the simulation started, the system simulated the entire process of the upper shear blade contacting the plate surface, pressing in the material, crack initiation, and crack penetration through the plate thickness.

[0101] Specifically, defect feature data is extracted in real time. After the simulation, the system automatically measures the virtual fracture morphology and extracts the current shearing defect data: burr height is 0.12 mm, edge collapse is 0.25 mm, and surface roughness is 12.50 micrometers. The above data indicates that the shearing quality under the current process parameters does not meet the preset finishing grade standard (i.e., the burr height should be less than 0.05 mm).

[0102] Furthermore, based on the defect feature data, a correlation database between shearing quality and process parameters is established; the system utilizes historical production data and multiple simulation results to construct a multidimensional mapping relationship table, which quantifies the influence weight of each 0.01 mm change in shearing gap on burr height, and the suppression effect of each 1.00 MPa change in surface prestress on edge collapse.

[0103] Furthermore, the stress triaxiality control logic aimed at improving the fracture quality of shear bands utilizes a specific physical path to suppress shear section defects by fine-tuning surface prestress parameters:

[0104] When simulating crack propagation behavior using fracture mechanics models, the digital twin should focus on calculating the triaxial evolution of stress within the shear zone. The algorithm aims to pre-place a high-pressure stress layer on the plate surface by adjusting the surface prestress parameters of the previous process (such as the reduction ratio of the final straightening pass), thereby increasing the hydrostatic pressure value in the shear zone. The system automatically optimizes the prestress distribution that can suppress micropore aggregation and nucleation and minimize shear fracture energy, rather than simply pursuing geometric flatness. This strategy explains how "straightening" serves "shearing" from the perspective of microscopic fracture mechanism, realizing cross-process collaboration by actively managing the quality of the downstream section using the upstream stress field.

[0105] Specifically, the process parameters of subsequent shearing cycles are adjusted through a feedback control strategy; the system adopts an adaptive step size search algorithm with the goal of minimizing defect feature data; for the burr height of 0.12 mm, the algorithm determines that the lateral clearance of the shearing blade needs to be reduced and the tilt angle of the shearing blade needs to be increased; for the edge collapse of 0.25 mm, the algorithm determines that the pre-stress on the surface of the plate needs to be increased to limit plastic rheology.

[0106] Furthermore, the system performs verification iterations on the adjusted parameters in the digital twin to bring the burr height and collapse amount of the virtual shear section within the set threshold. After 12 virtual iterations, when the virtual burr height is reduced to 0.03 mm and the collapse amount is reduced to 0.08 mm, the iteration is considered to have converged.

[0107] Specifically, the optimal combination of shear blade angle and surface prestress parameters is output; this parameter combination includes the following seven specific values, which are then transmitted to the production line actuator:

[0108] The tilt angle of the upper shear blade is set to 1.50 degrees to optimize the stress distribution of the shear section.

[0109] The tilt angle of the lower shear blade is set to 0.00 degrees to maintain a horizontal reference.

[0110] The lateral clearance of the shearing blade is adjusted to 0.18 mm, which is approximately 9.00% of the sheet thickness.

[0111] The pre-tension stress on the inlet side surface of the plate is set to 25.50 MPa and is applied through the inlet tension roller;

[0112] The pre-tension stress on the exit side surface of the plate is set to 10.20 MPa to keep the plate taut during shearing.

[0113] The lateral offset of the prestressed loading roller group is set to 5.00 mm to generate a local bending moment near the shear line to counteract the shear overturning moment;

[0114] The prestressing advance is set to 0.20 seconds, meaning that a stable prestressing field is established by the servo system 0.20 seconds before the shear blade contacts the plate.

[0115] By introducing high-speed shear transient virtual simulation technology, the originally lagging quality inspection (burrs, edge collapse) is transformed into forward process prediction and boundary condition back-inference. By adjusting the surface prestress parameters, a favorable stress state is pre-set before the plate enters the shear zone, which effectively suppresses the uncontrolled propagation of cracks and realizes cross-process quality feedforward control of "straightening serving shear".

[0116] Furthermore, the specific steps for constructing a geometrical digital twin and performing topological feature analysis by fusing production line laser point cloud data and visual images are as follows:

[0117] First, the system acquires 3D point cloud data and visual texture images of the board surface in real time. Specifically, the system activates four sets of high-precision laser profilometers and two 8K line scan industrial cameras spanning the production line roller conveyor. The lateral detection resolution of the laser profilometers is set to 0.05 mm, the depth detection accuracy to 0.01 mm, and the sampling frequency to 8000.00 Hz. The line scan camera's row frequency is simultaneously set to 25000.00 Hz. As the board passes under the detection beam at a speed of 2.50 m / s, the system simultaneously acquires the depth point cloud flow and grayscale texture flow of the board surface.

[0118] Furthermore, the system maps and generates a real-time geometric digital twin of the board material with geometric coordinates and surface features. The system uses a timestamp alignment algorithm to spatially register the laser point cloud data with the visual image. Specifically, the system constructs a virtual board segment reference surface with a length of 3000.00 mm and a width of 1500.00 mm in virtual space, maps the massive amount of collected point cloud data onto this reference surface, generates a high-density triangular mesh model, and covers the mesh surface with the corresponding texture image as a material map, thereby constructing a geometric digital twin that is completely consistent with the geometric features of the physical board material.

[0119] Furthermore, the geometric digital twin undergoes surface reconstruction and noise filtering. The system applies the moving least squares method to smooth the original triangular mesh, setting the sampling radius to 2.00 mm to filter out high-frequency noise signals generated by production line vibration. Subsequently, using the chord length parameterization method, the vertices of the three-dimensional discrete mesh are mapped to the two-dimensional parameter domain with the reference plane as the projection object, establishing the correspondence between physical coordinates and parameter coordinates. Then, node vectors are generated according to the preset smoothness, and the basis function weights of each data point are calculated using the B-spline recursive algorithm. Based on this, an overdetermined linear equation system is constructed with the control points to be solved as unknowns and the basis functions as the coefficient matrix. The least squares method is used to solve for the coordinates of the globally optimal control points that minimize the overall fitting error. Finally, the normal deviation between the generated surface and the original data is calculated. If the deviation exceeds the limit, node insertion and iterative recalculation are automatically performed locally until a mathematically complete analytical surface that meets the accuracy requirements is obtained. Through the above steps, the smoothed discrete mesh is reconstructed into a continuous NURBS surface (non-uniform rational B-spline surface).

[0120] Specifically, multi-scale geometric deviation analysis is performed to identify and classify various defects; the system sets the standard reference plane as an absolutely flat surface (height Z equals 0.00 mm); the reconstructed NURBS surface is compared with the standard reference plane using Boolean difference operations to extract the height deviation matrix; the system performs classification and identification based on a pre-set geometric feature library.

[0121] When periodic sinusoidal undulations are detected in the edge area of ​​the board material, and the peak height is greater than 2.50 mm and the wavelength is less than 300.00 mm, it is determined to be an edge wave;

[0122] When there are long wave undulations with a height greater than 3.00 mm in the center area of ​​the width of the board, it is judged as a medium wave;

[0123] When there are isolated protrusions in a local area with a diameter of less than 50.00 mm and a height change of more than 0.50 mm, it is judged as a local concavity and convexity.

[0124] When the absolute value of the vertical deviation of the four corners of the board relative to the center plane exceeds 5.00 mm, it is judged as warping.

[0125] Furthermore, based on the abrupt changes in geometric curvature and abnormal surface texture of the defect area, the system calculates the equivalent stress concentration and energy level corresponding to each defect; the system performs Gaussian curvature calculation on the identified defect areas; for example, for a local concave-convex defect, the system calculates that the maximum principal curvature at its apex is 0.08 millimeters; at the same time, the system analyzes the visual texture of the area, and if the grayscale gradient change detected exceeds 30.00 (8-bit grayscale depth), it indicates the presence of surface slip lines or oxide peeling, and the stress weight needs to be increased.

[0126] Specifically, based on the calculation results, the system quantifies the stress concentration level into energy levels from 1 to 10; for the aforementioned local uneven defects, combined with their geometric curvature and texture characteristics, the system calculates their stress concentration equivalent to 280.50 MPa and classifies them as energy level 4.

[0127] Furthermore, a defect topological fingerprint is obtained, containing the defect's spatial coordinates, defect type, boundary profile, and stress concentration energy level. The system encapsulates the above analysis results into a structured data packet, namely, the "defect topological fingerprint." For example, the topological fingerprint data generated for the detected edge wave defect is as follows:

[0128] The defect type is operational side wave;

[0129] The spatial coordinates of the defect are located at the center point, which is 1250.50 mm in the length direction and 50.20 mm in the width direction of the plate.

[0130] The boundary profile is an elliptical region enclosed by a series of coordinate points, with a major axis of 200.00 mm and a minor axis of 80.00 mm.

[0131] The stress concentration energy level is 6 (corresponding to the high residual stress zone).

[0132] The topological fingerprint is then sent to the subsequent moving target interception algorithm module to guide the path planning of the local straightening mechanism.

[0133] By fusing three-dimensional point clouds and two-dimensional textures to construct a digital twin of geometric shape, this invention can quantify complex surface defects into a "topological fingerprint" containing precise coordinates and energy levels, filter out environmental noise interference, and identify minute undulations and warps that are difficult to detect with the naked eye, providing reliable mathematical targets for subsequent point-to-point precise repair.

[0134] For localized high-energy stress concentration areas (such as localized rib waves or dead wrinkles) that were not completely eliminated in the previous process, local stress release and repair are based on a moving target interception algorithm. The process of moving target interception and local repair of stress release rollers based on defect topology fingerprints is described as follows:

[0135] First, based on the defect topology fingerprint generated in the previous steps, stress concentration defect regions with energy levels exceeding a threshold are screened out. Specifically, the system sets the energy level threshold to level 3. In the current plate sample, the system identifies two high-energy defect regions to be processed: the first is a local wave located 50.20 mm from the edge of the plate on the operating side, with an energy level of 6; the second is a stress spot located 300.50 mm from the edge of the plate on the transmission side, with an energy level of 4. The system marks these two defect regions as dynamically moving target points that move with the plate at a longitudinal speed of 2.50 m / s.

[0136] Furthermore, using the real-time running speed of the plate as input, the system predicts the future position sequence of each target point in the coordinate system of the straightener in real time; the system establishes a local coordinate system with the initial position of the stress relief roller as the origin, and calculates that the first high-energy target point will reach the transverse axis position of the stress relief roller in 0.85 seconds, and the second target point will reach it in 1.45 seconds.

[0137] Specifically, the stress relief roller is used as the motion actuator, and a time-optimal moving target interception algorithm is run. The system plans the optimal interception path based on the maximum lateral movement speed (set to 1.00 m / s) and maximum acceleration (set to 2.50 m / s²) of the stress relief roller. The algorithm prioritizes the highest energy level 6 defects as the primary interception target, generates a coupled lateral and vertical motion trajectory, and ensures that the roller surface can accurately move to the lateral coordinate of 50.20 mm in 0.85 seconds and simultaneously complete the vertical pressing action.

[0138] The time-optimal moving target interception algorithm includes the following steps:

[0139] A two-dimensional tracking model was established for the stress relief roller and the defect. By combining the real-time speed of the sheet material and the local movement component of the defect relative to the substrate, the future trajectory coordinates of the target point at different times were deduced. A state-space model containing position and velocity was constructed using Kalman filtering. Sensing data was processed using state transition and measurement matrices, and the predicted values ​​were corrected in real time using Kalman gain to eliminate observation errors. Time-optimal control was solved under the constraint of the maximum acceleration of the servo system. A three-stage strategy of "pre-acceleration-uniform speed-deceleration" was adopted to ensure that the stress relief roller is synchronized with the target point at the moment of contact and the total motion time is minimized. Segmented time-varying control was implemented according to the energy level distribution of the defect. The pressing process was divided into three proportional steps: initial contact, high-energy zone and tail transition. A high-order interpolation algorithm was used to ensure the smooth and continuous change of the pressing amount in each stage and to avoid damage to the sheet material.

[0140] Furthermore, by combining the spatiotemporal synchronization control logic of Kalman filtering, the dynamic response and contact mode of the moving target interception algorithm on high-speed moving plates are optimized.

[0141] To plan the lateral motion trajectory of the stress relief roller, a Kalman filter algorithm is introduced to predict the displacement deviation of the defect target point under microsecond-level time steps as it moves at high speed with the sheet material. The trajectory generated by the algorithm is not a simple point-to-point straight line, but a "flexible following curve" that matches not only the spatial position but also the phase velocity of the local waveform of the sheet material. When calculating the local compression, a progressive loading strategy is adopted, and the effectiveness of the progressive loading strategy is verified in virtual simulation. Comparative tests show that when using sudden full compression, the maximum contact stress reaches 1800 MPa, and a 0.05 mm scratch mark is produced on the sheet surface. However, when using progressive loading (0.50 times initially, 1.00 times in the middle stage, and 0.75 times in the final stage), the maximum contact stress drops to 1200 MPa, with no obvious scratches, and the stress relief effect is improved due to more sufficient plastic rheology. This ensures that the rollers maintain a rolling contact with the virtual panel with zero relative speed difference at the moment of pressing, avoiding sliding scratches; while ensuring that local residual stress is fully released, it protects the surface smoothness of the high-surface-grade automotive panel to the greatest extent and avoids secondary scratch defects introduced by repair operations.

[0142] Reference Figure 2 This is a schematic diagram of the coupling process of the two digital twins in this invention;

[0143] In space, a unified coordinate system is established based on the sheet metal itself. The surface point cloud data of the geometric digital twin is mapped to the physical boundary of the mesh, and the internal residual stress data of the material microscopic performance digital twin is filled into the internal attributes of the mesh, thereby constructing an enhanced voxel model that combines external morphology and internal energy. At the same time, virtual delay queue technology is used to eliminate the time lag caused by the physical workstation spacing. The caching and precise release of upstream microscopic data are controlled according to the real-time running speed of the sheet metal to ensure that it is strictly aligned with the physical slices collected downstream on the time axis. Finally, data interaction is completed through standardized multidimensional state vectors. The three-dimensional coordinates and waveform features of defects provided by the geometric twin are encapsulated and fused with the local stress tensor and stiffness modulus provided by the microscopic twin. This provides holographic data support, including precise geometric positioning and internal energy level, for the moving target interception algorithm to plan the repair path.

[0144] Furthermore, in an environment where the digital twin of material microstructure properties and the digital twin of geometric morphology are coupled in real time, the rolling contact process between the stress relief roller, the virtual plate surface, and the bottom support mechanism is simulated; the system constructs a virtual stress relief roller model with a diameter of 80.00 mm and sets the spacing between the bottom support rollers to 120.00 mm.

[0145] Specifically, refer to Figure 3 This is a schematic diagram of the process of performing defect repair using a geometrical digital twin according to the present invention. The local plastic deformation and stress release in the contact area are calculated. In the initial simulation, the system attempts to apply a local compression of 0.80 mm. The simulation results show that at the instant of rolling contact, the equivalent plastic strain in the defect area reaches 0.002, the peak residual stress decreases from 450.00 MPa to 280.00 MPa, and the corresponding energy level drops to level 4, which does not meet the repair standard (it needs to be reduced to below level 2).

[0146] Furthermore, based on the stress release effect obtained from the simulation, the local reduction and holding time of each interception point are adjusted in reverse iteration; the system automatically increases the reduction parameter in step size of 0.10 mm; after 4 iterations, when the local reduction is adjusted to 1.25 mm and the effective contact length in the defect core area reaches 15.00 mm, the simulation shows that the residual stress peak value is reduced to 120.50 MPa and the energy level is reduced to level 1, which meets the set threshold requirements.

[0147] Furthermore, after traversing all target points, the complete lateral motion trajectory curve of the stress relief roller and the corresponding local reduction timing parameter table are obtained; for the above two target points, the trajectory instructions generated by the system are as follows:

[0148] 0.00 seconds to 0.85 seconds: The stress relief roller accelerates laterally from the standby position and reaches the lateral coordinate of 50.20 mm at 0.85 seconds, simultaneously performing a pressing action of 1.25 mm;

[0149] 0.85 seconds to 1.45 seconds: The stress relief roller lifts up and moves laterally to the drive side, reaching the lateral coordinate -300.50 mm (relative to the center line) at 1.45 seconds, and performs a 0.95 mm pressing action (the pressing amount calculated for level 4 defects).

[0150] 1.45 seconds later: Reset to standby zero position.

[0151] Specifically, closed-loop verification is performed in the coupled digital twin; the system loads the generated complete trajectory instructions to conduct a full-process virtual simulation; the verification results show that after local fixed-point repair of this specific trajectory, the overall flatness error of the virtual board surface decreases from the initial 2.50I units to 0.35I units, and no new secondary scratches or stress concentrations are introduced, confirming that the target local defect meets the repair standard; finally, the trajectory data is packaged and sent to the production line PLC controller for execution.

[0152] By employing a moving target interception algorithm and rolling contact simulation, this invention solves the problem of spatiotemporal synchronization in which dynamic defects on continuous production lines are difficult to accurately locate and repair. By pre-simulating the local pressing path and stress release effect in digital space, it achieves point-to-point repair, which not only eliminates local high-energy potential wells but also avoids plate thinning or secondary damage caused by excessive pressing.

[0153] During long-term continuous production, the accuracy of digital twin models decays over time due to physical factors such as mechanical wear and thermal expansion. The adaptive tracking and parameter correction steps of digital twin models are described in detail below:

[0154] First, a real-time comparison mechanism is established between physical production line quality data and digital twin prediction data. Specifically, the system synchronously collects the measured data of the physical production line outlet flatness meter and the predicted flatness data output by the digital twin at a cycle of 1.00 minutes.

[0155] Furthermore, a deviation dataset was constructed; within a 4.00-hour production window of continuous operation, the system recorded 240 sets of comparison samples; for example, at the 120th minute, the physically detected residual wave height in the middle of the plate was 0.45 mm, while at the same moment the wave height predicted by the digital twin was 0.25 mm, and the system recorded the instantaneous deviation value at this time as 0.20 mm.

[0156] Furthermore, time series analysis is used to identify deviations caused by equipment wear and / or thermal drift; the system applies a sliding window mean filtering algorithm to the above deviation dataset, with the window length set to 30 sample points;

[0157] Specifically, system analysis revealed a clear linear monotonically increasing trend in the deviation data. Data showed that at the 10.00-minute mark of production, the average deviation was only 0.02 mm (within the range of random noise); however, by the 200.00-minute mark, the average deviation had cumulatively drifted to 0.18 mm. Through cross-correlation analysis with historical data from the equipment's temperature sensor, the system calculated a Pearson correlation coefficient of 0.88 between this deviation trend and the hydraulic station oil temperature change. Based on this, it was determined that the deviation was caused by zero-point drift of the servo actuator due to thermal drift in the hydraulic system, rather than a sudden malfunction. Simultaneously, the system monitored that the estimated value of the roll surface roughness coefficient slowly decreased with increasing cumulative steel throughput, identifying it as a slight characteristic of roll wear.

[0158] Furthermore, the system applies adaptive filtering and trend analysis to the deviation dataset. First, a Kalman smoother is used for smoothing, with process noise set to 0.01 mm and measurement noise set to 0.02 mm. Then, the smoothed deviation sequence is subjected to first-order and second-order differences to calculate its linear trend term and periodic term.

[0159] The following criteria are used to distinguish the sources of deviation:

[0160] If the deviation exhibits a linear monotonic trend (standard deviation of the first difference < 0.002 mm, absolute mean of the second difference < 0.001 mm), and the Pearson correlation coefficient with hydraulic oil temperature change > 0.80, it is determined to be thermal drift; the corresponding correction factor is the deviation change per degree Celsius / equipment temperature change rate.

[0161] If the deviation changes abruptly at a certain moment (single change > 0.05 mm), and the cross-correlation coefficient with the estimated value of the roll surface roughness is > 0.70, it is judged as roll wear; the corresponding correction factor is the wear depth per ton of steel passed (calculated by integrating the dynamic trajectory and the reduction).

[0162] Based on the above determination, thermal drift correction parameters and wear correction parameters are calculated independently and applied to the virtual hydraulic system parameters and the geometric parameters of the virtual work roll, respectively, to achieve hierarchical adaptive correction.

[0163] Furthermore, based on the deviation results, the baseline parameters of the digital twin are updated; the system initiates a model parameter correction procedure based on the identified thermal drift characteristics.

[0164] Specifically, in the digital twin, the system sets the zero-position offset parameter correction of the virtual hydraulic cylinder to a temperature rise drift rate of 0.05 mm / h; for roll wear, the system corrects the diameter parameter of the virtual work roll from the initial 160.00 mm to 159.92 mm, and lowers the friction coefficient of the virtual roll surface from 0.12 to 0.11, so as to reflect the actual wear and smoothness of the physical roll after long-term operation.

[0165] Furthermore, the global parameter bias value is calculated to enable the digital twin model to adaptively track the aging state of the physical equipment. Based on the corrected digital twin, the system re-runs the parameter inverse solution algorithm to calculate the compensation amount required under the current equipment state. Specifically, the system outputs a set of global bias instructions: for thermal drift, a compensation bias of 0.15 mm is added to the original overall reduction amount; for stiffness changes caused by roll wear, the basic setting value of the asymmetric bending roll force is uniformly increased by 50.50 kN.

[0166] Finally, the above bias values ​​were superimposed on the main control loop and verified within the following 10.00 minutes. The verification data showed that after adaptive correction, the average deviation between the physical measured value and the digital twin predicted value fell back to within 0.03 mm, successfully eliminating the systematic errors caused by equipment aging and environmental changes.

[0167] By establishing a virtual-real data comparison mechanism based on time series analysis, the digital twin system is endowed with the ability to adaptively evolve in response to physical equipment wear and environmental thermal drift. This enables the optimization model to automatically correct the baseline parameters and bias values ​​as the production cycle progresses, overcoming the drawback of traditional control models gradually failing due to equipment aging and ensuring the stability of parameter optimization during long-term operation.

[0168] This embodiment constructs a digital twin of the material's microstructure and geometric morphology, reconstructs the virtual stress field, calculates the neutral layer offset through virtual straightening pre-simulation, and solves the optimal straightening parameters through reverse iteration, fine-tunes the shearing process using a fracture mechanics model, and runs a moving target interception algorithm based on defect topology fingerprinting to plan local repair trajectories, and finally combines equipment wear and thermal drift analysis to adaptively correct model parameters and achieve full-process digital closed-loop control.

[0169] Example 2

[0170] This implementation method is mainly applied to the exit section of the continuous annealing production line for third-generation automotive high-strength steel (such as QP1180 quenched and distributed steel). Compared with ordinary high-strength steel, this type of material contains metastable residual austenite, has an extremely high work hardening rate and complex micro-stress evolution behavior, and the plate shape defects often manifest as a composite "W"-shaped wave with the superposition of edge waves and rib waves at the quarter position.

[0171] This embodiment elaborates on the steps of "construction of digital twin of material microstructure properties" and "local repair of moving target interception", and the specific implementation steps are as follows:

[0172] Step 1: Constructing a highly sensitive digital twin of the microscopic properties of materials

[0173] First, taking advantage of the unique magnetic-force coupling characteristics of QP1180 steel, the system activates a high-density multispectral electromagnetic sensor array. Specifically, the system arranges 256 high-sensitivity probes along the width of the plate, with the sensor spacing reduced to 6.50 mm to capture fine stress stripes. The excitation signal is set as a multi-frequency composite waveform, including a 1500.00 Hz high-frequency component for detecting the surface and a 15.00 Hz low-frequency component for penetrating and detecting the core.

[0174] Furthermore, when establishing the coupling model, the system introduces a residual austenite phase transformation correction factor; since the residual austenite is a non-ferromagnetic phase, its content fluctuations will interfere with stress detection; the system combines the residual austenite volume fraction measured in real time by the X-ray diffractometer at the production line entrance, which is about 12.5%, to perform benchmark compensation on the magnetoelastic detection data.

[0175] Specifically, the magnetoelastic detection data of the physical plate is mapped and reconstructed into a residual stress tensor field; the system generates a fine mesh model with a thickness of 1.60 mm in virtual space; by analyzing the compensated multispectral signal, the system reconstructs the residual stress field inside the plate that presents a "saddle-shaped" distribution; the data shows that in this composite wave defect region, the peak value of the residual compressive stress along the rolling direction is as high as 680.50 MPa, and there is a sharp abrupt change in the gradient at one-quarter of the plate width, with the stress gradient value reaching 55.00 MPa per millimeter.

[0176] Furthermore, the virtual straightening machine was driven to perform a virtual straightening pre-simulation. After loading a high-precision constitutive model with a follow-up hardening modulus of 2200.00 MPa, the simulation showed that due to the material's extremely high work hardening capacity, the physical neutral layer underwent significant dynamic drift during the severe bending process. During the fourth bending pass, the physical neutral layer deviated from the geometric center plane by 0.38 mm. Based on this, the non-uniform differential curvature target calculated by the system indicated that a local reverse bending with a curvature of 0.045 cuboids needed to be applied at one-quarter of the plate width. This far exceeded the adjustment capability of the conventional straightening machine, thus triggering the subsequent local repair process.

[0177] Step 2: Repair of Local Composite Defects Based on Moving Target Interception

[0178] Based on the above analysis, the system determined that overall straightening alone could not completely eliminate high-energy composite defects, and therefore initiated a local repair procedure.

[0179] First, based on the defect topological fingerprint, the high-energy stress concentration area is located. The system identifies a discontinuous rib wave with a major axis of 450.00 mm located 350.00 mm from the edge of the plate on the operating side. Its topological fingerprint shows a stress concentration energy level of 8, which is the highest risk level. Moreover, the defect area moves downstream with the plate at a high speed of 3.20 m / s.

[0180] Furthermore, a time-optimal moving target interception algorithm is run; considering the high production line speed and the fact that the defect location is within the width of the board, the system plans for the stress relief roller to execute a highly dynamic response "oblique cut-in" interception strategy.

[0181] Specifically, the system generates motion trajectories that are coupled laterally and vertically:

[0182] The pre-acceleration stage is 1.20 seconds before the defect arrives, during which the stress relief roller is moved laterally to a standby position 350.00 mm away from the edge of the plate, and the pre-pressure of the hydraulic servo system is established.

[0183] The dynamic interception stage involves the stress relief roller rapidly pressing down to a depth of 1.55 mm when the front end of the defect enters the repair zone. Since the defect is accompanied by a slight lateral S-shaped oscillation (oscillation amplitude of about 15.00 mm), the algorithm drives the stress relief roller to perform lateral fine-tuning at a speed of 0.30 m / s while pressing down, keeping the center of the roller surface always aligned with the high-energy core of the defect.

[0184] The variable compression reduction control is based on real-time simulation guidance from a digital twin of the material's microscopic properties. The system calculates that the stress release requirements at the head, middle, and tail of the defect are different. Therefore, during the interception process, the local compression reduction is not constant, but dynamically adjusted according to a time-varying sequence of 1.55 mm, 1.80 mm, and 1.40 mm to match the stress distribution profile inside the defect.

[0185] Furthermore, the repair effect was verified in a digital twin. Simulation results showed that after the precise interception of the dynamic variable pressure reduction, due to the introduction of local plastic deformation, the peak residual compressive stress in the original rib wave region dropped sharply from 680.50 MPa to 150.20 MPa, and the energy level dropped to level 2. After confirmation, the complex instruction set containing the lateral following trajectory and time-varying pressure reduction was sent to the physical actuator, realizing the precise elimination of composite defects in the high-speed flowing QP1180 steel plate.

[0186] This embodiment addresses composite plate defects in QP1180 high-strength steel by constructing a highly sensitive digital twin of the material's microscopic properties. Through multi-frequency composite excitation and residual austenite phase transformation correction, a virtual stress field with drastically abrupt gradient changes is reconstructed. For "W"-shaped waves, neutral layer shift is calculated and local repair is triggered. A moving target interception algorithm is used, and a stress release roller is planned to execute a lateral following and time-varying reduction strategy to dynamically intercept high-energy defects moving at high speeds, effectively eliminating complex composite defects.

[0187] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing plate shape control parameters based on digital twins, characterized in that, include: The process involves constructing a digital twin of the material's microscopic properties and performing virtual straightening simulations. This includes calculating the offset of the physical neutral layer and deriving the non-uniform differential curvature target for the equivalent fitting residual stress tensor field. Specifically, the steps include: real-time acquisition of signals from a multispectral electromagnetic sensor array; establishing a magnetostrictive and stress statistical coupling model based on the material thickness direction; mapping and reconstructing the magnetoelastic detection data of the physical sheet material into a residual stress tensor field; constructing a digital twin of the material's microscopic properties with microscopic anisotropy; loading a reduced-order constitutive model containing isotropic hardening, kinematic hardening, and the Bauschinger effect into the digital twin; driving a virtual straightener to rapidly simulate a complete virtual straightening process involving multiple loading-unloading-reverse loading passes; accurately calculating the offset of the physical neutral layer by tracking the dynamic drift trajectory of the neutral layer; and deriving the spatial non-uniform differential curvature target distribution that maximizes the offset of the residual stress tensor field; performing reverse iteration based on the target to output the globally optimal parameter combination for the straightener's asymmetric bending roll force, roll cross angle, and overall reduction. Using a fracture mechanics model to drive a digital twin of the material's microscopic properties for virtual simulation, the crack propagation process of the plate during shearing is simulated, and defect characteristic data including burr height and edge collapse are calculated. Based on this, the angle of the subsequent shearing blade and the surface prestress parameters of the previous process are adjusted. By integrating laser point cloud data and visual images from the production line, a geometric digital twin is constructed and topological feature analysis is performed to generate a defect topological fingerprint containing the coordinates of the stress concentration area and the energy level. Based on the defect topological fingerprint, a moving target interception algorithm is run, and two digital twins are coupled to simulate the stress relief roller, virtual plate surface and bottom support mechanism. The optimal lateral movement trajectory is planned and the local pressure is calculated to verify the local defect repair effect. The method further includes: establishing a real-time comparison mechanism between physical production line quality data and digital twin prediction data; using time series analysis to identify deviations caused by equipment wear and / or thermal drift; updating the baseline parameters of the digital twin based on the deviation results; calculating global parameter bias values; and realizing adaptive tracking and parameter correction of the aging state of physical equipment by the digital twin model.

2. The method for optimizing plate shape control parameters based on digital twins according to claim 1, characterized in that, The step of performing reverse iteration based on the target to output the globally optimal parameter combination of the asymmetric bending roll force, roll cross angle, and overall reduction of the straightener includes: Using the non-uniform differential curvature objective as the inverse optimization constraint, a physical field proxy model containing the straightener roll gap geometry, contact nonlinearity, and dynamic response of the servo hydraulic system is constructed in the material microscopic property digital twin. An inverse iterative optimization algorithm based on sequential quadratic programming is run. By continuously adjusting the parameters of the asymmetric bending roll force vector of the upper working roll, the cross angle of the upper and lower roll systems, and the overall reduction, the deviation of the residual stress tensor field of the virtual plate exit section from the ideal straight state is gradually converged to within the preset tolerance. The parameter combination of the straightener asymmetric bending roll force, roll cross angle, and overall reduction that satisfies the global convergence solution is output.

3. The method for optimizing plate shape control parameters based on digital twins according to claim 1, characterized in that, The steps of using a fracture mechanics model to drive a digital twin of the material's microscopic properties for virtual simulation, simulating the crack propagation process of the sheet metal during shearing, calculating defect characteristic data including burr height and edge collapse, and adjusting the subsequent shearing blade angle and the surface prestress parameters of the previous process accordingly, specifically include: A fracture mechanics model is loaded into a digital twin of material microstructure properties to perform high-speed shear transient virtual simulation, simulating the crack initiation and propagation process of the plate during shearing, and extracting defect feature data such as burr height, edge collapse amount and cross-sectional roughness in real time. Based on the defect characteristic data, a correlation database between shearing quality and process parameters is established. Through feedback control strategy, the shear blade tilt angle, blade side clearance, and surface prestress loading parameters of subsequent shearing cycles are adjusted so that the burr height and edge collapse of the virtual shearing section converge to within a set threshold. The optimal combination of shear blade angle and surface prestress parameters is output. The parameter combination includes the upper shear blade tilt angle, the lower shear blade tilt angle, the shear blade side clearance, the prestress on the plate inlet side surface, the prestress on the plate outlet side surface, the lateral offset of the prestress loading roller group, and the prestress application advance.

4. The method for optimizing plate shape control parameters based on digital twins according to claim 1, characterized in that, The steps of fusing laser point cloud data and visual images from the production line to construct a geometrical digital twin and perform topological feature analysis to generate a defect topological fingerprint containing the coordinates of stress concentration regions and energy levels specifically include: Real-time acquisition of 3D point cloud data and texture images of the board surface is used to generate a real-time geometrical digital twin of the board with geometric coordinates and surface features. After surface reconstruction and noise filtering of the geometrical digital twin, multi-scale geometric deviation analysis is performed to identify and classify defects including edge waves, middle waves, local concavity and convexity, and warping. Based on the geometric curvature abrupt change and surface texture anomaly of the defect area, the stress concentration equivalent and energy level corresponding to each defect are calculated to obtain a defect topological fingerprint that includes the defect spatial coordinates, defect type, boundary contour, and stress concentration energy level.

5. The method for optimizing plate shape control parameters based on digital twins according to claim 1, characterized in that, Based on the defect topological fingerprint, a moving target interception algorithm is run, coupling two digital twins to simulate the stress relief roller, virtual plate surface, and bottom support mechanism. The optimal lateral movement trajectory is planned, and the local pressure is calculated. Verification of the local defect repair effect includes: Based on the defect topological fingerprint, stress concentration defect regions with energy levels exceeding the threshold are regarded as dynamically moving target points that move with the longitudinal movement of the plate. Using the real-time running speed of the plate as input, the future position sequence of each target point in the coordinate system of the straightener is predicted in real time; using the stress relief roller as a motion actuator, a time-optimal moving target interception algorithm is run to generate the coupled motion trajectory of the stress relief roller in the lateral and vertical directions, so that the roller surface prioritizes tracking and covering the current target point with the highest energy level at each moment. The rolling contact process between the stress relief roller, the virtual plate surface, and the bottom support mechanism is simulated in a digital twin that is coupled in real time with the digital twin of the material's microscopic properties and the digital twin of its geometric morphology. The local plastic deformation and stress relief in the contact area are calculated. Based on the stress relief effect obtained from the simulation, the local reduction and holding time of each interception point are adjusted in reverse iteratively until the residual stress concentration energy level in the defect area drops below the set threshold. After traversing all target points, the complete lateral motion trajectory curve of the stress relief roller and the corresponding local pressure reduction timing parameter table are obtained. Closed-loop verification is then performed in the coupled digital twin to confirm that the target local defects meet the repair criteria.

Images