Digital twin control method for shape of ultra-thin strip based on multi-scale surrogate model

Abstract

This invention discloses a digital twin control method for ultra-thin strip rolling shape based on a multi-scale surrogate model, belonging to the field of metal rolling shape control technology. The method includes: firstly, real-time acquisition of the operating parameters of the physical rolling mill, inputting them into a pre-trained macroscopic surrogate model to predict post-rolling macroscopic stress and node positions; then, extraction of boundary conditions corresponding to mesoscopic representative volume elements, inputting them into a mesoscopic surrogate model to predict internal grain stress; subsequently, fusion of macroscopic and mesoscopic stress data to calculate a comprehensive stress uniformity index S; when S is below a preset threshold, optimizing the operating parameters in virtual space with the goal of minimizing index deviation, generating an optimal control strategy, and issuing it to the rolling mill for execution. This invention effectively suppresses shape defects caused by microstructural inhomogeneities and achieves feedforward precise intelligent control of the ultra-thin strip rolling process.

Description

Technical Field

[0001] This invention belongs to the field of metal rolling processing shape control technology, and particularly relates to a digital twin control method for ultra-thin strip rolling shape based on a multi-scale surrogate model. Background Technology

[0002] In the rolling production process, strip shape control is a core element in ensuring product quality. Existing technologies primarily rely on physical models based on macroscopic mechanical principles, combined with online measurement feedback systems to achieve strip shape regulation. Specifically, this system typically collects real-time straightness data of the rolled strip using a strip shape meter installed at the mill exit. When shape deviations (such as edge waviness or center waviness) are detected, the control system adjusts the mill's process parameters, such as roll tilt, bending force, rolling speed, and tension, in real time according to a preset PID (proportional-integral-derivative) algorithm. This method effectively corrects macroscopic geometric deformations and ensures product strip shape quality when processing conventional strip steel with thicknesses in the millimeter or sub-millimeter range. The technology is relatively mature and widely used.

[0003] However, as products become increasingly thinner, the control methods based on the macroscopic continuum assumption exhibit significant limitations when strip thickness reaches the micrometer level. First, traditional models treat the material as a homogeneous, isotropic continuum, neglecting the significant impact of the material's internal mesoscopic structure (such as grain orientation, size, morphology, and grain boundary interactions) on macroscopic mechanical behavior at extremely thin scales. In this case, the strip thickness direction contains only a few grains, and the uneven stress distribution at the mesoscopic scale becomes the deep physical root cause of macroscopic strip shape defects, which existing macroscopic models cannot perceive or compensate for. Second, feedback control relying on the exit strip shape gauge suffers from response lag. Adjustments often depend on operator experience for trial-and-error parameter corrections, lacking precise, quantitative control standards and theoretical basis, resulting in low control accuracy and poor stability. Furthermore, the stress state within the rolling deformation zone is complex, and critical residual stresses are difficult to observe directly online, making the process like a "black box." Although finite element simulation can perform offline analysis and optimization, its enormous computational cost cannot meet the stringent timeliness requirements of online real-time control. Therefore, existing technologies cannot accurately describe and control the relationship between mesoscopic structure evolution and macroscopic quality during ultra-thin strip rolling, which restricts further improvement in the shape quality and production stability of ultra-thin strip products. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a digital twin control method for ultra-thin strip rolling profile based on a multi-scale surrogate model, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, this invention provides a digital twin control method for ultra-thin strip rolling profile based on a multi-scale surrogate model, comprising: Real-time acquisition of actual operating parameters of the physical rolling mill; The actual working condition parameters are input into a pre-trained macroscopic proxy model to predict the macroscopic node positions and macroscopic stress data of the rolled strip. Based on the macroscopic node location data, the boundary conditions corresponding to the pre-established mesoscopic representative volume element RVE model are extracted. The boundary conditions are input into a pre-trained mesoscopic surrogate model to predict the stress data of the mesoscopic nodes inside the RVE. Based on the macroscopic stress data and the mesoscopic nodal stress data, the comprehensive stress uniformity index S is calculated; Determine whether the comprehensive stress uniformity index S is less than a preset stress difference threshold. If so, then minimize. With the objective of obtaining the optimal control strategy, the operating parameters are virtually optimized based on the macroscopic proxy model and the mesoscopic proxy model. The optimal control strategy is sent to the control system of the physical rolling mill to adjust the rolling process parameters.

[0006] Preferably, the calculation of the comprehensive stress uniformity index S specifically includes: Calculate the macroscopic stress uniformity value based on the macroscopic stress data. ; Calculate the mesoscopic stress uniformity value based on the aforementioned mesoscopic nodal stress data. ; Through formula The comprehensive stress uniformity index S is calculated, where α is a preset macroscopic stress weighting coefficient.

[0007] Preferably, the macroscopic stress uniformity value The calculation steps include: Real-time macroscopic stress data from N characteristic monitoring points on the rolling sheet are collected, and their average value and standard deviation are calculated to obtain the macroscopic stress variation coefficient. ; Through formula The macroscopic stress uniformity value was calculated. .

[0008] Preferably, the mesoscopic stress uniformity value The calculation steps and the macroscopic stress uniformity value The calculation steps are logically consistent.

[0009] Preferably, the construction steps of the macroscopic proxy model and the mesoscopic proxy model include an offline phase, in which: A high-precision finite element full-size model was established, and multiple sets of representative working condition parameter combinations were generated through experimental design methods to conduct batch simulations in order to obtain training data. Based on the training data, the macroscopic proxy model and the mesoscopic proxy model are trained respectively.

[0010] Preferably, the method determines that the comprehensive stress uniformity index S is less than a preset stress difference threshold. Then, the optimization algorithm is started to perform the virtual optimization.

[0011] Preferably, the optimal control strategy includes quantitative adjustment instructions for at least one of rolling speed, rolling force, and rolling tension, corresponding to the numerical range of the comprehensive stress uniformity index S.

[0012] Preferably, the mesoscopic proxy model is established based on crystal plastic constitutive relations.

[0013] Preferably, the macroscopic proxy model and the mesoscopic proxy model are integrated into the virtual twin module of the digital twin system, and the virtual twin module interacts with the physical rolling mill in real time through the data interaction and control module.

[0014] Preferably, the operating parameters include rolling speed, rolling force, and strip tension.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention, through a technical scheme of "calculating the comprehensive stress uniformity index S based on the macroscopic stress data and the mesoscopic nodal stress data" and subsequent optimization control steps, shifts the control objective from traditional macroscopic geometric correction to the level of "mesoscopic nodal stress equilibrium" within the material. By constructing and calculating the quantitative index S that integrates macroscopic and mesoscopic stress information in real time, the control system can directly perceive and respond to the fundamental cause of uneven stress distribution due to mesoscopic factors such as grain orientation and grain boundary interactions. This proactively optimizes the stress state at the source, significantly reducing the generation of macroscopic defects such as edge waves and mid-waves.

[0016] This invention, through a digital twin system integrating macroscopic and mesoscopic proxy models, enables real-time and rapid prediction of future stress distribution based on current operating conditions before actual shape deviations occur during the physical rolling process. It then actively optimizes parameters in virtual space based on the prediction results (S-value). This forms a feedforward-feedback composite intelligent control paradigm based on a mechanistic model, replacing the lag adjustment relying on exit detection and manual trial and error, significantly improving the timeliness, accuracy, and stability of control.

[0017] This invention utilizes two key features: inputting data into a pre-trained macroscopic surrogate model and inputting data into a pre-trained mesoscopic surrogate model. A lightweight surrogate model, trained using offline high-fidelity simulation data, replaces the computationally time-consuming full-scale finite element model or crystal plastic finite element model for online prediction. These surrogate models, while maintaining prediction accuracy, achieve computational efficiency that meets the stringent time requirements of online real-time control cycles, thus enabling the application of deep analysis based on multi-scale physical mechanisms in real-time industrial control scenarios.

[0018] This invention achieves a more uniform mesoscopic stress distribution (i.e., increases the S-value) through proactive optimization, which not only facilitates better flatness but also reduces the risk of subsequent processing cracking caused by localized stress concentration. Simultaneously, the entire closed-loop control process based on digital twins enables real-time interaction and decision-making between physical production and the virtual model, giving the rolling process self-sensing, self-predicting, and self-optimizing capabilities, significantly improving the intelligence level of the production system and the scientific rigor of process optimization. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is an overall flowchart of the digital twin system according to an embodiment of the present invention; Figure 2 This is a flowchart of digital twin data processing according to an embodiment of the present invention; Figure 3 This is an example diagram of the visualization interface of the multi-scale rolling digital twin system according to an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0022] Example 1 like Figure 1 As shown, this embodiment provides a digital twin system, which mainly includes three parts: a physical entity module, a virtual twin module, and a data interaction and control module.

[0023] Physical entity module: that is, the actual ultra-thin strip mill, including rolls, motors, sensors (such as speed, rolling force, tension sensors) and actuators.

[0024] Virtual twin module: This is the core of the invention and is deployed on an industrial computer or in the cloud. It includes a macroscopic proxy model library, a mesoscopic RVE model library, and a solver for model scheduling and computation.

[0025] Data Interaction and Control Module: Responsible for the data connection between the physical and virtual worlds. It collects real-time operating data of the physical rolling mill and transmits it to the virtual twin module; at the same time, it receives the optimal control commands calculated by the virtual module and sends them to the physical rolling mill.

[0026] like Figure 2 As shown in the flowchart of digital twin data processing, the specific working principle is as follows: The macroscopic surrogate model first completes the macroscopic stress field prediction based on the received global geometric feature data. Then, the system extracts the corresponding macroscopic response parameters of the RVE boundary region as boundary driving conditions and maps and transmits them to the microscopic surrogate model, thereby realizing cross-scale joint simulation prediction from the overall stress distribution to the local RVE refined stress.

[0027] This embodiment provides a digital twin control method for ultra-thin strip rolling profile based on a multi-scale surrogate model, including: Step 1: Real-time acquisition of actual operating parameters of the physical rolling mill; Furthermore, the operating parameters include rolling speed, rolling force, and strip tension.

[0028] Step 2: Input the actual working condition parameters into the pre-trained macroscopic surrogate model to predict the macroscopic node positions and macroscopic stress data of the rolled strip. Specifically, the construction of the macroscopic proxy model (offline stage) involves establishing a high-precision finite element (FEM) full-size model: using ABAQUS finite element software, a refined three-dimensional ultra-thin strip rolling model including the roll system and strip is established. This model can accurately simulate the macroscopic deformation behavior of strip steel under different working conditions.

[0029] Design Experimental Sample Points (DoE): Within a common process parameter space, such as a rolling speed range of 1 m / s to 5 m / s and a rolling force range of 150 kN to 300 kN, a series of representative working condition combinations are generated using methods such as Latin hypercube sampling.

[0030] Batch simulation and data acquisition: Run the FEM model to perform batch simulation calculations for the above working conditions. After each simulation, extract key result data, including the spatial coordinates of nodes, stress data, mesh type, etc.

[0031] Training and validating the surrogate model: The collected data is organized into "input-output" pairs. The input is the operating parameters, and the output is the position coordinate matrix of all nodes of the rolled strip. A neural network and Gaussian process regression algorithm are used to train the macroscopic surrogate model. This allows it to fit the following relationship: ,in Includes post-tie node coordinates and stress. The operating parameters include rolling speed, strip tension, and rolling force. The initial position node coordinates are given, where the coordinates in the rolling direction are based on the plate length cycle of offline simulation in the actual production mileage.

[0032] This model mainly realizes the continuous proxy of macroscopic discrete simulation results, which meets the requirements of real-time simulation and continuous chemical condition parameter changes.

[0033] Step 3: Based on the macroscopic node location data, extract the boundary conditions corresponding to the pre-established mesoscopic representative volumetric unit RVE model; Step 4: Input the boundary conditions into the pre-trained mesoscopic surrogate model to predict the mesoscopic nodal stress data inside the RVE; Furthermore, the construction steps of the macroscopic proxy model and the mesoscopic proxy model include an offline phase, in which: A high-precision finite element full-size model was established, and multiple sets of representative working condition parameter combinations were generated through experimental design methods to conduct batch simulations in order to obtain training data. Based on the training data, the macroscopic proxy model and the mesoscopic proxy model are trained respectively.

[0034] Furthermore, the mesoscopic proxy model is established based on crystal plastic constitutive relations.

[0035] Furthermore, the macroscopic proxy model and the mesoscopic proxy model are integrated into the virtual twin module of the digital twin system, and the virtual twin module interacts with the physical rolling mill in real time through the data interaction and control module.

[0036] Specifically, the mesoscopic RVE model is constructed (offline stage) by establishing a representative volumetric unit (RVE): a 5×5mm region is selected in the post-rolling region (the stable deformation zone of the strip), and an RVE model based on crystal plasticity is constructed. Polycrystalline geometries that conform to the true statistical characteristics of the corresponding material state are generated using techniques such as Voronoi diagrams.

[0037] Assigning a plastic constitutive model to the crystal: Each grain in the RVE model is assigned a plastic constitutive relation based on dislocation slip theory. This constitutive model can describe the anisotropic elastoplastic deformation behavior of the grains under stress.

[0038] Apply displacement boundary conditions and calculate: Apply different displacement combinations (corresponding to macroscopic deformation) to the boundary of the RVE model, and solve for the stress tensor of all nodes inside the RVE under this displacement using CP-FEM.

[0039] Training the mesoscopic surrogate model: RVE simulation is performed based on a simulation case using macroscopic sampling conditions. The position matrix of the RVE boundary nodes and the stress matrix data of all internal nodes are extracted. The mesoscopic surrogate model is trained using the same surrogate model technique to construct the mapping relationship.

[0040] Step 5: Calculate the comprehensive stress uniformity index S based on the macroscopic stress data and the mesoscopic nodal stress data; Furthermore, the calculation of the comprehensive stress uniformity index S specifically includes: Calculate the macroscopic stress uniformity value based on the macroscopic stress data. ; Calculate the mesoscopic stress uniformity value based on the aforementioned mesoscopic nodal stress data. ; Through formula The comprehensive stress uniformity index S is calculated, where α is a preset macroscopic stress weighting coefficient.

[0041] Furthermore, the macroscopic stress uniformity value The calculation steps include: Real-time macroscopic stress data from N characteristic monitoring points on the rolling sheet are collected, and their average value and standard deviation are calculated to obtain the macroscopic stress variation coefficient. ; Through formula The macroscopic stress uniformity value was calculated. .

[0042] Furthermore, the mesoscopic stress uniformity value The calculation steps and the macroscopic stress uniformity value The calculation steps are logically consistent.

[0043] Specifically, the method for constructing the stress uniformity index integrates macroscopic stress data of the rolled plate and microscopic stress data of grains within the RVE (Resilience Variation). Through normalization and weighted fusion, a comprehensive stress uniformity index S with a value ranging from 0 to 1 is obtained. A lower S value indicates a more uneven stress distribution in the rolled plate; S=1 indicates that the macroscopic and microscopic stress distributions of the rolled plate have reached an ideal state. Step 5 involves the following steps: Step 501: First, collect real-time macroscopic stress data from N feature monitoring points on the rolling plate using dimensionality reduction. Calculate the average value of macroscopic stress. with standard deviation Furthermore, the macroscopic stress variation coefficient was obtained. Through formula Obtain the macroscopic stress uniformity value , It changes continuously as the measured value increases, and It falls within the interval [0, 1]. hour (Macroscopic stress is completely uniform) When it increases Monotonically decreasing, meaning the macroscopic stress dispersion increases; Step 502: Collect real-time micro-stress data of M grains within the RVE. Using the same calculation method as the macroscopic stress uniformity value, the grain stress variation coefficient was obtained. and stress uniformity value This ensures consistency in calculation logic with the macroscopic stress uniformity value. Step 503: Combining the macroscopic stress weighting coefficient α corresponding to different materials, and through weighted fusion... Calculate the overall stress uniformity index S.

[0044] The operation comparison table for uniformity index is shown in Table 1.

[0045] Table 1 Step 6: Determine whether the comprehensive stress uniformity index S is less than the preset stress difference threshold. If so, then minimize. With the objective of obtaining the optimal control strategy, the operating parameters are virtually optimized based on the macroscopic proxy model and the mesoscopic proxy model. Step 7: Send the optimal control strategy to the control system of the physical rolling mill to adjust the rolling process parameters.

[0046] Furthermore, the method determines that the comprehensive stress uniformity index S is less than a preset stress difference threshold. Then, the optimization algorithm is started to perform the virtual optimization.

[0047] Furthermore, the optimal control strategy includes quantitative adjustment instructions for at least one of the rolling speed, rolling force, and rolling tension, corresponding to the numerical range of the comprehensive stress uniformity index S.

[0048] Specifically, the online operation and control (online phase) of a digital twin system includes: (1) Data acquisition: The system collects the current actual working conditions from the physical rolling mill in real time.

[0049] (2) Macroscopic deformation prediction: Input the actual working condition data and the current node coordinates into the macroscopic surrogate model. In this process, the predicted node location matrix and stress matrix of the rolled strip surface are calculated.

[0050] (3) Micro-stress prediction: Extract the boundary node positions corresponding to the RVE region from the macro-data matrix. Input them into the meso-scale model to calculate the stress matrix of all nodes inside the RVE.

[0051] (4) Optimization and decision-making: Set an ideal stress difference threshold. If the current Then the optimization algorithm is started to minimize With the objective of achieving this, virtual calculations are performed by virtually adjusting the input operating conditions. Combined with a parameter control lookup table, the optimal control strategy is output.

[0052] (5) Control command issuance: The calculated optimal working condition is used as a control command and issued to the control system of the rolling mill through the data interaction module to adjust the actual running speed of the rolling mill.

[0053] This embodiment forms a feedforward closed-loop control system based on micromechanical prediction, thereby achieving precise and intelligent control of the ultra-thin strip rolling process.

[0054] like Figure 3 As shown, this is an example of a visualization interface for a multi-scale rolling digital twin system. The upper view displays the stress evolution diagram of representative mesoscopic volumetric elements at key locations, while the lower view shows the plate stress diagram and explicit grain distribution characteristics generated based on a macroscopic surrogate model. By extracting macroscopic field data to drive the mesoscopic model, real-time monitoring of cross-scale digital twins is achieved.

[0055] Technical effects of this embodiment: This embodiment effectively suppresses local non-uniform deformation caused by grain anisotropy and grain boundary interactions by actively controlling the balance of mesoscopic stress, thereby improving the shape quality of ultra-thin strips.

[0056] This embodiment helps improve the flatness of the product by homogenizing the mesoscopic stress distribution, reducing the risk of deformation and cracking in subsequent processing, and improving the mechanical properties of the product.

[0057] In this embodiment, the digital twin system can monitor and predict the rolling state in real time, quickly calculate new optimal process parameter combinations, enhance the robustness and flexibility of the production system, and realize intelligent and adaptive optimization of the rolling process.

[0058] This embodiment, through extensive virtual experiments and process optimization in a digital twin system, can provide a reference for physical experiments, shorten the process development cycle of new materials and new specifications, and further reduce R&D and production costs.

[0059] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A digital twin control method for ultra-thin strip rolling profile based on a multi-scale surrogate model, characterized in that, Includes the following steps: Real-time acquisition of actual operating parameters of the physical rolling mill; The actual working condition parameters are input into a pre-trained macroscopic proxy model to predict the macroscopic node positions and macroscopic stress data of the rolled strip. Based on the macroscopic node location data, the boundary conditions corresponding to the pre-established mesoscopic representative volume element RVE model are extracted. The boundary conditions are input into a pre-trained mesoscopic surrogate model to predict the stress data of the mesoscopic nodes inside the RVE. Based on the macroscopic stress data and the mesoscopic nodal stress data, the comprehensive stress uniformity index S is calculated; Determine whether the comprehensive stress uniformity index S is less than a preset stress difference threshold. If so, then minimize. With the objective of obtaining the optimal control strategy, the operating parameters are virtually optimized based on the macroscopic proxy model and the mesoscopic proxy model. The optimal control strategy is sent to the control system of the physical rolling mill to adjust the rolling process parameters.

2. The control method according to claim 1, characterized in that, The calculation of the comprehensive stress uniformity index S is specifically as follows: Calculate the macroscopic stress uniformity value based on the macroscopic stress data. ; Calculate the mesoscopic stress uniformity value based on the aforementioned mesoscopic nodal stress data. ; Through formula The comprehensive stress uniformity index S is calculated, where α is a preset macroscopic stress weighting coefficient.

3. The control method according to claim 2, characterized in that, The macroscopic stress uniformity value The calculation steps include: Real-time macroscopic stress data from N characteristic monitoring points on the rolling sheet are collected, and their average value and standard deviation are calculated to obtain the macroscopic stress variation coefficient. ; Through formula The macroscopic stress uniformity value was calculated. .

4. The control method according to claim 2, characterized in that, The mesoscopic stress uniformity value The calculation steps and the macroscopic stress uniformity value The calculation steps are logically consistent.

5. The control method according to claim 1, characterized in that, The construction steps of the macroscopic proxy model and the mesoscopic proxy model include an offline phase, in which: A high-precision finite element full-size model was established, and multiple sets of representative working condition parameter combinations were generated through experimental design methods to conduct batch simulations in order to obtain training data. Based on the training data, the macroscopic proxy model and the mesoscopic proxy model are trained respectively.

6. The control method according to claim 1, characterized in that, The method determines that the comprehensive stress uniformity index S is less than a preset stress difference threshold. Then, the optimization algorithm is started to perform the virtual optimization.

7. The control method according to claim 1, characterized in that, The optimal control strategy includes quantitative adjustment instructions for at least one of the rolling speed, rolling force, and rolling tension, corresponding to the numerical range of the comprehensive stress uniformity index S.

8. The control method according to claim 1, characterized in that, The mesoscopic proxy model is established based on the crystal plastic constitutive relation.

9. The control method according to claim 1, characterized in that, The macroscopic proxy model and the mesoscopic proxy model are integrated into the virtual twin module of the digital twin system. The virtual twin module interacts with the physical rolling mill in real time through the data interaction and control module.

10. The control method according to claim 1, characterized in that, The operating parameters include rolling speed, rolling force, and strip tension.

Images