Method for optimizing roughing pass system of multi-billet continuous casting square billet based on finite element software

Abstract

The application discloses a multi-billet type continuous casting square billet rough rolling pass system optimization method based on finite element software, and relates to the technical field of steel rolling and casting. The method comprises the following steps: based on the principle of equal metal second flow, a digital model of volume flow dynamic balance during rolling of different section billets is established to quantize the square billet rough rolling deformation law; a pass modeling for multi-billet type switching is carried out by using a finite element simulation model, a collaborative computing architecture is built, and finite element calculation scheduling tasks are dynamically deployed; based on multi-physical field coupling analysis, stress and strain characteristics and size evolution law of the billet under multi-dimensional parameter conditions are predicted, process optimization targets are generated, and the demand of different section billets is met; virtual and real data real-time mapping is established, a process knowledge base is built, and regular learning and updating are carried out. By using the finite element software, the rough rolling deformation law of different section billets can be more accurately simulated, the shape and size of the rolled material can be more accurately predicted and controlled, and the accuracy and reliability of the optimization result are improved.

Description

Technical Field

[0001] The invention relates to the technical field of steel rolling and casting, in particular to a method for optimizing a rough rolling pass system of a multi-bill continuous casting square billet based on finite element software. Background Art

[0002] In the field of casting technology, continuous billet casting is a common method for producing steel. The main process involves cooling liquid steel through a crystallizer to form a solid billet, which is then processed through roughing and finishing rolling to produce steel of various specifications. In steel rolling technology, optimizing the roughing pass system is a key step in improving steel quality and production efficiency, directly affecting the steel's dimensional accuracy, surface quality, and internal structure.

[0003] In the field of computer-aided design, finite element software is widely used to simulate and optimize various manufacturing processes, including casting and rolling. Existing methods for optimizing the roughing pass system for continuous casting billets are primarily based on experience and experimental data, adjusting the pass parameters and reduction to meet the requirements of different cross-sectional billet shapes.

[0004] In addition, some researchers have used finite element software to simulate the rough rolling process to optimize the pass design. However, these methods often require a lot of computing resources and time, and the optimization results may not meet the requirements of all cross-section billets.

[0005] There are some problems and limitations in the existing optimization methods for the rough rolling pass system of continuous casting square billets. First, methods based on experience and experimental data are often unable to accurately predict the deformation laws of billets with different cross-sections, resulting in insufficient accuracy and reliability of the optimization results. Secondly, although finite element software can provide more accurate simulation results, its computational efficiency is low, and for complex pass systems, it requires large computing resources. In addition, existing methods are often unable to fully consider different pass parameters, the stress and strain characteristics of the corners and the surface to the core of the rolled material under different reductions, and the size and area of ​​the rolled section after deformation, resulting in the optimization results may not meet the needs of multi-blank continuous casting square billets. Summary of the Invention

[0006] Based on this, it is necessary to provide an optimization method for the rough rolling pass system of multi-bill continuous casting square billets based on finite element software to address the above technical problems.

[0007] The present invention provides a method for optimizing a rough rolling pass system of a multi-bill continuous casting billet based on finite element software, comprising:

[0008] S1. Based on the principle of equal metal flow rate per second, a digital model of the dynamic balance of volume flow during rolling of different cross-section billets is established to quantify the deformation law of billet rough rolling;

[0009] S2. Use the finite element simulation model to model the hole pattern for multi-blank switching, build a collaborative computing architecture, and dynamically deploy finite element calculation scheduling tasks;

[0010] S3. Based on multi-physics field coupling analysis, predict the stress-strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions, generate process optimization targets, and meet the requirements of blanks with different cross-sections;

[0011] S4. Establish real-time mapping between virtual and real data, build a process knowledge base and regularly study and update it.

[0012] Furthermore, based on the principle of equal metal flow rate per second, a digital model of the dynamic balance of volume flow during rolling of different cross-section billets was established to quantify the deformation laws of billet rough rolling, including:

[0013] S11. Set the geometric features of various billet sections, build a parametric template library, and establish a mapping relationship between billet geometric parameters and hole system;

[0014] S12. Based on the principle of equal metal flow rate per second, a dynamic balance equation for volume flow in multi-pass rolling is established, a mathematical model for the relationship between width expansion and reduction is constructed, and a shape factor correction term is introduced to quantify the metal flow differences of different billet shapes;

[0015] S13. Combine historical rolling data to establish a statistical relationship model among rolling force, torque and hole parameters, and integrate and summarize the deformation laws of rolling under different cross-section conditions.

[0016] Furthermore, the finite element simulation model is used to model the hole pattern of multi-blank switching, and a collaborative computing architecture is built to dynamically deploy finite element calculation scheduling tasks including:

[0017] S21, extracting geometric parameters of the billet, inputting them into a finite element simulation model, generating a hole profile curve through a non-uniform rational B-spline, matching an initial hole parameter combination in a billet rule library, and generating a parametric model of the billet hole;

[0018] S22. Build a cognitive grid in the finite element simulation model and optimize the grid topology;

[0019] S23. Divide the rolling process into several stages, match them with corresponding parallel acceleration strategies, and build a quantum-classical-edge collaborative computing architecture to collaboratively optimize computing resources;

[0020] S24. Allocate the finite element calculation task to the optimal calculation node according to the finite element calculation load.

[0021] Furthermore, building a cognitive grid in the finite element simulation model and optimizing the grid topology include:

[0022] S221, after the geometric parameters of the billet are input, an optimized mesh topology is generated, and the mesh is divided into a corner area, a surface area, and a core area according to the positions of the corners, surface, and core of the billet;

[0023] S222. Generate a seven-layer progressively finer mesh in the corner area, use anisotropic elements in the surface area to capture friction effects, and configure a coarse mesh in the core area.

[0024] S223. Deploy a stress gradient sensitivity detector in the grid. When the stress gradient sensitivity is greater than a preset sensitivity threshold, trigger local grid encryption and insert high-density monitoring points in the potential defect area.

[0025] Furthermore, the rolling process is divided into several stages, matched with corresponding parallel acceleration strategies, and a quantum-classical-edge collaborative computing architecture is built to collaboratively optimize computing resources, including:

[0026] S231, dividing the billet rolling process into a biting stage, a steady-state stage, and a casting stage;

[0027] S232. Configure a multi-stage parallel acceleration strategy. During the biting phase, assign the finite element calculation task to the GPU's stream multiprocessor. A single stream multiprocessor handles the calculation of one roll contact point. During the steady-state phase, enable multi-process services to simultaneously calculate the temperature-deformation coupling field of multiple passes. During the casting phase, use the GPU memory to directly store deformation history data.

[0028] S233. Build a quantum-classical-edge collaborative computing architecture that integrates quantum computing, classical computing, and edge computing functions to achieve collaborative optimization of heterogeneous computing resources.

[0029] Furthermore, the collaborative computing architecture includes a quantum computing layer, a classical computing layer, and an edge computing layer;

[0030] Among them, the quantum computing layer converts the hole parameter optimization problem into the Ising model and defines the Hami variables; the classical computing layer is deployed in a high-performance computing cluster, and each node is used for calculations in a specific stage; the edge computing layer deploys a deep neural network model at the rolling mill site to predict rolling force and width expansion in real time.

[0031] Furthermore, based on multi-physics field coupling analysis, the stress-strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions are predicted, and the process optimization objectives are generated, including:

[0032] S31. Construct a fully coupled model of temperature field, stress field and microstructure field, and solve it through distributed coupling to show the influence, distribution and transformation of each physical field;

[0033] S32. Based on the regional division standard of the billet shape, quantify the differential deformation characteristics of the billet at the corners, surface and core, and evaluate the rough rolling defects of the billet;

[0034] S33. Based on the pre-established material constitutive relationship adapted to multiple working conditions, a quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established;

[0035] S34. Set the blank size accuracy, strain uniformity and surface integrity as process optimization goals, establish the optimization objective function and constraint conditions, and solve to obtain the optimal process parameter combination.

[0036] Furthermore, based on the regional division standard of the billet shape, the differential deformation characteristics of the billet at the corners, surface and core are quantified, and the rough rolling defects of the billet are evaluated, including:

[0037] S321. Divide the cross section into a corner sensitive area, a surface transition area, and a core stable area according to the locations of the corners, surface, and core of the billet, and select differentiated monitoring indicators for each area.

[0038] S322. Based on the cognitive grid of the finite element simulation model, extract the equivalent strain distribution data of each region, calculate the local gradient vector, and set the ratio of the maximum strain gradient to the average strain as the strain gradient index through normalization processing;

[0039] S323, uniformly selecting N monitoring points in each area, recording the temperature value in the steady-state rolling stage, calculating the temperature range and average temperature of each area, and setting the ratio of the temperature range to the average temperature as the temperature uniformity coefficient;

[0040] S324. Match similar strain gradient indices and temperature uniformity coefficients in the historical database, and evaluate the process defects in the current billet rolling through threshold judgment.

[0041] Furthermore, based on the pre-established material constitutive relationship that adapts to multiple working conditions, the quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established, including:

[0042] S331, extracting process parameters related to the cross-sectional dimensions of the billet and defining the parameter range;

[0043] S332. Load the dynamic constitutive equation into the finite element simulation model, implement thermomechanical coupling calculation through the temperature-strain rate coupling field, and use orthogonal experiments to generate several sets of process parameter combinations. Perform full-process rolling simulation for each set of process parameters and output simulation data.

[0044] S333. Standardize and perform principal component analysis on the simulation data, screen the influencing indicators, and construct a quantitative relationship equation between the process parameters and the simulation data.

[0045] Furthermore, the optimization objective function is:

[0046] ;

[0047] Where, f1 represents dimensional accuracy; f2 represents strain uniformity; f3 represents surface integrity; represents the simulation size of the i-th cross-section position; represents the target size of the i-th section position; represents the equivalent effect standard deviation; n represents the total number of cross-sectional locations; represents the equivalent effect mean; ε surface represents the average equivalent strain of the surface transition zone; ε core represents the average equivalent strain of the core stable area; Ra represents the surface roughness; N defect Represents the number of defects per meter; ω1 represents the roughness weight coefficient; ω2 represents the defect number weight coefficient.

[0048] The beneficial effects of the present invention are:

[0049] 1. By adopting finite element software, the rough rolling deformation law of different cross-section billets can be simulated more accurately, thereby more accurately predicting and controlling the shape and size of the rolled material, and improving the accuracy and reliability of the optimization results; by identifying the stress and strain characteristics of the corners and the surface to the core of the rolled material under different pass parameters and reduction, as well as the cross-sectional size and area after deformation, the rough rolling pass system of multi-blank continuous casting square billets is fully optimized, meeting the needs of various cross-sectional billets and improving production efficiency; compared with existing methods, the calculation efficiency is higher, and even for complex pass systems, simulation and optimization can be completed with reasonable computing resources, reducing optimization costs, while having higher accuracy, stronger adaptability and higher efficiency, which can better meet the needs of modern steel production.

[0050] 2. By building an intelligent computing architecture and hybrid modeling technology, we achieved breakthroughs in both computational efficiency and optimization accuracy. In terms of computational efficiency, we employed GPU parallel computing technology to decompose the rolling process into three stages (bite, steady state, and steel throwing), reducing the simulation time per pass from 4-6 hours with traditional methods to less than 30 minutes. This also improved the efficiency of searching in high-dimensional parameter spaces, shortening the optimization cycle. In terms of precision control, we embedded the metal flow conservation equation into the machine learning model through a physical information neural network, addressing the lack of physical rationality in traditional data-driven models.

[0051] 3. Through multi-physics field coupling analysis and dynamic management of the process knowledge base, the quality consistency and process controllability of multi-bill rolling have been comprehensively improved. First, based on a fully coupled model of temperature field, stress field, and microstructure field, cross-scale predictions from macroscopic deformation to microstructure are achieved, accurately quantifying the differentiated deformation characteristics of corners, surfaces, and cores. Second, the constructed process knowledge base stores data from over 2,000 working conditions. Combined with a similarity matching algorithm, it provides historically optimized parameter recommendations for new billets, significantly reducing the amount of experimental data required for developing new specifications. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0053] Figure 1 4 is a flow chart of a method for optimizing a rough rolling pass system of a multi-bill continuous casting billet based on finite element software according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] See also Figure 1 , provides a multi-bill continuous casting billet rough rolling pass system optimization method based on finite element software, the method includes:

[0056] S1. Based on the principle of equal metal flow rate per second, a digital model of the dynamic balance of volume flow during rolling of billets with different cross-sections is established to quantify the deformation law of billet rough rolling.

[0057] In the description of the present invention, based on the principle of equal metal flow rate per second, a digital model of dynamic balance of volume flow rate during rolling of different cross-section billets is established to quantify the deformation law of rough rolling of square billets, including:

[0058] S11. Set the geometric features of various billet sections, build a parametric template library, and establish a mapping relationship between the billet geometric parameters and the hole system.

[0059] Specifically, the core principle of establishing a parametric template library is to convert the geometric features of different billets into quantifiable digital parameters through mathematical methods, so as to solve the problem of insufficient accuracy caused by reliance on empirical formulas in traditional rolling process design.

[0060] First, common billet shapes (such as square, rectangular, and round billets) must be geometrically decomposed to extract key dimensional parameters (side length, fillet radius, aspect ratio, etc.), and a standardized parameter system must be established. For example, for a square billet cross-section, parameters such as the initial side length (H0), corner fillet radius (R), and aspect ratio (α = W / H) must be defined. Each parameter must be within a reasonable range (e.g., a square billet side length tolerance of ±1.5mm).

[0061] Based on this, a parametric template library was constructed, storing the characteristic parameters of each billet and their allowable tolerances in JSON format. This implementation involved developing a parameter association algorithm to establish a mapping matrix between billet geometry parameters and pass system parameters (such as roll opening W, roll gap S, and roll crown C).

[0062] Taking a steel mill rolling Φ60mm round steel as an example, its parametric template is defined as: the initial square billet is 160×160mm (R=10mm), the target round billet is Φ60±0.3mm, and the corresponding hole system parameters include the elliptical hole opening W=1.18×160+3.5×Δh (Δh is the pass reduction). Through this mapping relationship, the hole design parameters for each pass can be automatically generated, which improves process design efficiency by 70%.

[0063] S12. Based on the principle of equal metal flow rate per second, a dynamic balance equation for volume flow in multi-pass rolling is established, a mathematical model of the relationship between width expansion and reduction is constructed, and a shape factor correction term is introduced to quantify the metal flow differences of different square billet shapes.

[0064] Specifically, based on the principle of equal metal second flow rate (Q=Av=Const, where: A represents the cross-sectional area of ​​the piece; v represents the average linear speed of the piece; Q represents the metal second flow rate; Const represents the constant flow rate value, which is not a fixed value, but means that the flow rates at different positions / passes in the same rolling process must be equal), it is necessary to establish a dynamic balance equation for the volume flow rate of multi-pass rolling.

[0065] First, the basic equation is derived based on the relationship between the inlet and outlet cross-sectional areas:

[0066] ;

[0067] where v i =(H i-1 / H i )×v i-1 ×(1+Δb / B i-1 ), Δb is the width expansion, and B is the billet width.

[0068] In order to achieve accurate calculation of the expansion, it is necessary to improve the traditional Siebel formula, introduce the temperature compensation factor and the shape factor correction term λ (i.e., shape factor), and establish the expansion-reduction correlation model:

[0069] ;

[0070] Where T is the rolling temperature (°C), and the shape factor λ is determined by the type of pass conversion (1.25 for square to elliptical and 0.85 for elliptical to circular). Taking a workshop rolling Φ50mm bearing steel as an example, the elliptical cross-section size at the second pass entrance is 220×130mm, the target round billet is Φ50mm, Δh=18mm, T=980°C (0.0025 is the temperature term coefficient obtained from the experiment, the unit of the coefficient itself is °C) -1 , eliminating the temperature dimension of the term e), the calculated value is Δb = 0.35 × 18^0.8 × e^(0.0025 × 980) × 0.85 ≈ 32.1 mm. The measured width expansion is 31.8 mm, with an error of only 0.9%. This model enables precise control of metal flow, reducing second-by-second flow rate fluctuations from ±8% to ±1.5%, significantly reducing process defects such as steel stretching and steel piling.

[0071] S13. Combine historical rolling data to establish a statistical relationship model among rolling force, torque and hole parameters, and integrate and summarize the deformation laws of rolling under different cross-section conditions.

[0072] Specifically, the core purpose of building a statistical model based on historical data is to overcome the limitations of theoretical models and reveal the hidden patterns among rolling parameters through data mining. First, data cleaning and feature engineering are performed on parameters such as rolling force (P), torque (M), temperature (T), and reduction (Δh) to extract key features such as strain rate and aspect ratio.

[0073] The rolling force prediction model was constructed using the multivariate nonlinear regression method, and the dynamic correction coefficient was obtained by fitting the experimental data. The torque model was established as M=0.5PDμ(1+tanα), and the temperature dependence of the friction coefficient μ was fitted by an exponential function: μ=0.45e -0.0018T By integrating theoretical models with data-driven methods, a knowledge base covering deformation laws of different cross-sections is ultimately formed to support intelligent optimization decisions for process parameters.

[0074] S2. Use the finite element simulation model to model the hole pattern of multi-blank switching, build a collaborative computing architecture, and dynamically deploy finite element calculation scheduling tasks.

[0075] In the description of the present invention, the finite element simulation model is used to model the hole pattern of multi-blank switching, and a collaborative computing architecture is built to dynamically deploy finite element calculation scheduling tasks, including:

[0076] S21. Extract the geometric parameters of the billet and input them into the finite element simulation model. Generate the hole profile curve through non-uniform rational B-spline. Match the initial hole parameter combination in the billet rule library to generate the billet hole parametric modeling.

[0077] Specifically, the core of step S21 is to convert the billet's geometric parameters into a computable die profile curve, enabling digital modeling for rapid switching between multiple die types. By extracting geometric parameters such as the billet's side length, fillet radius, and aspect ratio, the mathematical properties of non-uniform rational B-splines (NURBS), such as control point weighting and node vector definition, are utilized to generate a smooth and continuous die profile curve.

[0078] During implementation, billet geometry parameters are extracted from mechanical drawings (e.g., CAD drawings) or 3D scan data. For example, for a 160×160mm billet, the corner radius R = 10mm and the width-to-thickness ratio are 1.0. Five control points are set according to the pass design requirements, and the weight coefficients are set to 1.0, 0.8, 1.2, 0.9, and 1.0 based on the rolling force distribution to generate an elliptical pass profile curve. Finally, the K-nearest neighbor (KNN) algorithm is used to match similar parameter combinations within the billet rule library. For example, when a 160mm billet is input, the elliptical pass with an opening of W = 220mm and a reduction of Δh = 35mm, which is a historically successful example, is automatically retrieved.

[0079] S22. Build a cognitive grid in the finite element simulation model and optimize the grid topology.

[0080] In the description of the present invention, building a cognitive grid in a finite element simulation model and optimizing the grid topology includes:

[0081] S221. After the geometric parameters of the billet are input, an optimized mesh topology is generated, and the mesh is divided into a corner area, a surface area, and a core area according to the positions of the corners, the surface, and the core of the billet.

[0082] Specifically, the following aspects can be referred to when defining the three-layer monitoring area:

[0083] 1. Corner sensitive area (5mm from the edge);

[0084] 2. Surface transition zone (5-15mm depth);

[0085] 3. Central stable zone (>15mm depth).

[0086] S222. Generate seven layers of progressively denser grids in the corner areas, use anisotropic elements in the surface areas to capture friction effects, and configure a coarse grid in the core area.

[0087] Specifically, after completing the area division, a customized mesh strategy needs to be implemented according to the physical characteristics of each area. For the corner area, a seven-layer progressive encryption mesh technology is used. When expanding from the corner point to the outer layer, the unit size increases in geometric series. The side length of the first layer unit is 0.2mm, and the size of each subsequent layer increases by 1.5 times to ensure the calculation accuracy of the stress concentration area. Anisotropic hexahedral units are deployed in the surface area, and the unit aspect ratio along the rolling direction (X-axis) is set to 5:1 to accurately capture the friction shear effect between the roll and the billet. The core area is configured with a uniform coarse grid of size 2.0mm to reduce the computational load by reducing the number of nodes.

[0088] For example, in a rolling simulation, this strategy increased the number of elements in the corner region from 15,000 in a traditional uniform mesh to 52,000, while the total mesh size only increased from 1.2 million to 1.35 million. This improved accuracy while minimizing the computational overhead. The anisotropic elements in the surface region reduced the friction stress prediction error from 18% to 6.5%, validating the effectiveness of the meshing strategy.

[0089] S223. Deploy a stress gradient sensitivity detector in the grid. When the stress gradient sensitivity is greater than a preset sensitivity threshold, trigger local grid encryption and insert high-density monitoring points in the potential defect area.

[0090] Specifically, in order to dynamically optimize the mesh density, a stress gradient sensitivity detection mechanism needs to be embedded in the solution process. This detector calculates the von Mises stress gradient value between cells in real time.

[0091] When the gradient value detected in a certain area exceeds a preset threshold (e.g., 200 MPa / mm), local mesh refinement is automatically triggered, reducing the element size to half of its original value and inserting a network of monitoring points with a density four times that of conventional elements in that area. For example, midway through the rolling process, the system detected a sudden increase in the stress gradient to 280 MPa / mm at a point 8 mm from a corner. The system immediately implemented a three-level mesh refinement in that area, refining the element size from 0.5 mm to 0.125 mm. This resulted in a correction of the predicted peak stress value from 435 MPa to 489 MPa, reducing the error from the measured value of 502 MPa to 2.6%. Furthermore, high-density monitoring points were embedded in areas with a historical high incidence of defects (e.g., the transition zone between the core and the surface). Using a 0.1 mm spacing element network, the system captured signs of microcrack initiation, providing early warning of potential defects three simulation steps in advance and providing data support for dynamic adjustment of process parameters.

[0092] S23. Divide the rolling process into several stages, match the corresponding parallel acceleration strategy, and build a quantum-classical-edge collaborative computing architecture to collaboratively optimize computing resources.

[0093] In the description of this invention, the rolling process is divided into several stages, matched with corresponding parallel acceleration strategies, and a quantum-classical-edge collaborative computing architecture is built. The collaborative optimization of computing resources includes:

[0094] S231. The billet rolling process is divided into a biting stage, a steady-state stage and a casting stage.

[0095] Specifically, the stage division standards can refer to the following aspects:

[0096] 1. Biting stage: the front end of the workpiece contacts the roller until it fills the die (time 0-0.2s);

[0097] 2. Steady-state stage: the entire length of the rolled piece passes through the rollers (time 0.2~1.5s);

[0098] 3. Steel throwing stage: The tail of the rolled piece leaves the roll and cools freely (time 1.5~2.0s).

[0099] S232. Configure a multi-stage parallel acceleration strategy. During the biting stage, the finite element calculation task is assigned to the GPU's stream multiprocessor. A single stream multiprocessor handles the calculation of one roll contact point. During the steady-state stage, start the multi-process service to simultaneously calculate the temperature-deformation coupling field of multiple passes. During the steel throwing stage, use the GPU video memory to directly store the deformation history data.

[0100] Specifically, when configuring the parallel acceleration strategy, heterogeneous computing resources must be deployed based on the computational characteristics of each stage. The bite phase utilizes 108 streaming multiprocessors (SMs), each assigned a single roll contact point calculation task. For example, if the roll has 12 contact zones, 12 SMs are activated for concurrent computation. Each SM loads a local mesh (approximately 5,000 elements) within a 10mm radius around the contact point. Sharing memory accelerates data exchange, reducing single-step computation time from 15ms on the CPU to 0.8ms.

[0101] During the steady-state phase, multiple processes are activated, each corresponding to the temperature-deformation coupled field calculation for a single pass. Using the MPI parallel architecture, eight passes can be processed simultaneously on a 64-core CPU cluster. For example, in a six-pass rolling process, processes 1 through 6 calculate the steady-state field for each pass, while processes 7 and 8 serve as hot backups, automatically switching when the residual error of a process exceeds a threshold.

[0102] During the steel throwing stage, deformation history data (such as strain tensor and temperature gradient) are directly stored in the GPU memory, which can cache a large amount of historical state data, significantly reduce data reading latency, and achieve fast rollback and incremental calculation.

[0103] S233. Build a quantum-classical-edge collaborative computing architecture that integrates quantum computing, classical computing, and edge computing functions to achieve collaborative optimization of heterogeneous computing resources.

[0104] In this invention, the collaborative computing architecture comprises a quantum computing layer, a classical computing layer, and an edge computing layer. The quantum computing layer transforms the pass parameter optimization problem into an Ising model and defines Hami variables. The classical computing layer is deployed in a high-performance computing cluster, with each node dedicated to a specific computational stage. The edge computing layer deploys a deep neural network model at the rolling mill site to predict rolling force and width spread in real time.

[0105] Specifically, the quantum computing layer first maps the discrete and combined variables in the hole parameter optimization problem (such as hole geometry parameters, mold configuration, etc.) into spin variables in the Ising model; by constructing the Hamiltonian, it represents the objective function and constraints (such as metal second flow balance, expansion-reduction correlation, etc.), so that the optimal solution corresponds to the state with the lowest energy of the Hamiltonian.

[0106] The classical computing layer divides the overall optimization process into multiple computational stages (such as finite element simulation preprocessing, postprocessing, and iterative optimization), each of which is executed by dedicated nodes in the cluster. The classical computing layer receives the optimization candidate solutions output by the quantum computing layer and, combined with the finite element simulation results, performs more refined parameter adjustments and multi-physics coupling calculations.

[0107] The edge computing layer deploys edge computing devices at the rolling mill site and uses pre-trained deep neural network models to predict real-time sensor data (such as temperature, rolling force, roll speed, and roll width). The prediction results are used to adjust process parameters in real time, and the on-site status data is fed back to the classic layer for subsequent optimization and verification.

[0108] S24. Allocate the finite element calculation task to the optimal calculation node according to the finite element calculation load.

[0109] Specifically, each finite element task submission must include resource requirements, including the number of CPU cores required (e.g., 16 cores), memory capacity (e.g., 64GB), GPU acceleration requirements (e.g., NVIDIA A100), storage type (SSD), estimated computation time (e.g., 2 hours), and priority (high / medium / low). Each node is monitored for CPU utilization (e.g., Node A currently has 15% utilization), available memory (e.g., Node B has 96GB of free memory), GPU status (e.g., Node C's V100 GPU has 30% utilization), storage IOPS (e.g., Node D's SSD has 150k IOPS), and network latency (e.g., Node E has 0.5ms latency to the storage cluster). High-priority tasks (e.g., real-time quality control simulation) are prioritized for matching with the highest-scoring available nodes, while reserving some node resources for urgent tasks.

[0110] For example, when two tasks are submitted simultaneously—Task X (high priority, requiring 16 cores / 64GB) and Task Y (medium priority, requiring 32 cores / 128GB), the scheduler assigns Node A (32 free cores / 128GB memory) to Task X, while Task Y waits or is assigned to the combination of Nodes B+C.

[0111] S3. Based on multi-physical field coupling analysis, the stress-strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions are predicted, and process optimization targets are generated to meet the requirements of blanks with different cross-sections.

[0112] In the description of the present invention, based on multi-physics field coupling analysis, the stress-strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions are predicted, and the process optimization goals are generated, including:

[0113] S31. Construct a fully coupled model of temperature field, stress field and microstructure field, and solve it through distributed coupling to show the influence, distribution and transformation of each physical field.

[0114] Specifically, multi-physics coupling requires integrating heat conduction, mechanical deformation, and microstructural evolution mechanisms, fully coupling the temperature, stress, and microstructural fields. The temperature field influences the material's fluidity and plasticity, which in turn influences the stress-strain distribution. Simultaneously, the deformation history affects the microscopic grain structure and precipitation phases. Partial differential equations are used to describe the interactions between these fields, constructing a set of coupled equations to solve. For example, the overall behavior can be described using equations for energy conservation, momentum conservation, and the kinetics of microstructural evolution.

[0115] A distributed solution approach is employed, utilizing the finite element method (FEM) to spatially discretize the temperature and stress fields. Microstructure evolution models (such as grain growth and phase transformation models) are also introduced. Multi-physics simulation software (such as ANSYS, ABAQUS, and COMSOL) is employed to determine the spatial and temporal distribution of each physical field during the rolling process through step-by-step or fully coupled solution techniques.

[0116] Ultimately, the data provides visualization of temperature, stress, strain, and microstructural distributions, demonstrating the mutual influence and evolution trends of various physical fields. The field distribution and variation patterns in key areas (such as corners, surfaces, and centers) under different rolling conditions are analyzed, providing data support for subsequent quantitative analysis.

[0117] S32. Based on the regional division standard of the billet shape, quantify the differential deformation characteristics of the billet at the corners, surface and core, and evaluate the rough rolling defects of the billet.

[0118] In the description of the present invention, based on the regional division standard of the billet shape, the differential deformation characteristics of the billet at the corners, surface and core are quantified, and the rough rolling defects of the billet are evaluated, including:

[0119] S321. According to the locations of the corners, surface and core of the billet, the cross section is divided into a corner sensitive area, a surface transition area and a core stable area, and differentiated monitoring indicators are selected for each area.

[0120] Specifically, examples of the corner sensitive area, surface transition area, and core stable area are described as follows:

[0121] 1. Corner sensitive area: 5mm from the edge, mesh density increased to 0.3mm, monitoring stress concentration factor Kt;

[0122] 2. Surface transition zone: 5-15mm depth, analyze the risk of iron oxide scale peeling and surface roughness evolution;

[0123] 3. Central stable zone: >15mm depth, used to evaluate the expansion trend of central porosity and segregation defects.

[0124] S322. Based on the cognitive grid of the finite element simulation model, extract the equivalent strain distribution data of each region, calculate the local gradient vector, and set the ratio of the maximum strain gradient to the average strain as the strain gradient index through normalization processing.

[0125] S323. Evenly select N monitoring points in each area, record the temperature value in the steady-state rolling stage, calculate the temperature range and average temperature of each area, and set the ratio of the temperature range to the average temperature as the temperature uniformity coefficient.

[0126] S324. Match similar strain gradient indices and temperature uniformity coefficients in the historical database, and evaluate the process defects in the current billet rolling through threshold judgment.

[0127] S33. Based on the pre-constructed material constitutive relationship that adapts to multiple working conditions, a quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established.

[0128] In the description of the present invention, based on the pre-established material constitutive relationship adapted to multiple working conditions, the quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established, including:

[0129] S331. Extract process parameters related to the cross-sectional dimensions of the billet and define the parameter range.

[0130] Specifically, the core process parameters include the following aspects:

[0131] 1. Deformation parameters: reduction (Δh), pass distribution ratio, and widening coefficient;

[0132] 2. Temperature parameters: rolling temperature (T), temperature gradient (ΔT);

[0133] 3. Friction parameters: roller surface roughness (Ra), lubrication conditions;

[0134] 4. Equipment parameters: roll diameter (D), pass geometry (fillet radius R, opening degree).

[0135] Set the parameter reasonable range through historical data analysis and process specification, for example:

[0136] 1. Reduction Δh: 5%-30% of billet height;

[0137] 2. Rolling temperature T: 1050-1200℃ (austenite zone);

[0138] 3. Friction coefficient μ: 0.2-0.7 (dry friction to lubricated state).

[0139] S332, load the dynamic constitutive equation in the finite element simulation model, realize the thermal-mechanical coupling calculation through the temperature-strain rate coupling field, and generate several sets of process parameter combinations by orthogonal test, and execute the full-process rolling simulation for each set of process parameters, and output the simulation data.

[0140] Specifically, the pre-constructed material constitutive relationship suitable for multiple working conditions is introduced in the finite element simulation model, and the constitutive model can reflect the plastic flow, strain hardening and softening characteristics of the material under different temperature and strain rate conditions. Ensure that the constitutive equation can dynamically respond to the changes of temperature and strain rate in the rolling process, realize the thermal-mechanical coupling calculation, that is, consider the interaction between temperature field and mechanical field. Using the multi-physical field coupling method, the temperature field and stress-strain field are solved at the same time in the simulation, so that the influence of temperature change on material performance and deformation behavior can be accurately reflected.

[0141] According to the key process parameters and their value ranges determined in S331, an orthogonal test design method is used to construct several sets of process parameter combinations, ensuring that each parameter level is evenly covered in the test scheme. For each set of process parameter combination, run the full-process rolling simulation, from the initial state to the final rolling state, output the related simulation data such as the square billet cross-sectional size, local stress-strain distribution, etc.

[0142] S333, standardize and principal component analysis the simulation data, screen the influence indicators, and construct the quantitative relationship equation between the process parameters and the simulation data.

[0143] Specifically, each process parameter and the output simulation data (such as the cross-sectional size of each region, stress-strain value, etc.) need to be standardized to eliminate differences in different dimensions and numerical levels, and ensure that the data is compared and analyzed on the same scale. Use principal component analysis method to reduce the dimension of the standardized simulation data, select the key indicators that have the greatest influence on the square billet cross-sectional size, and eliminate redundant and noise information. Analyze the contribution rate of each principal component to determine which combination of parameters plays a leading role in the final cross-sectional size change.

[0144] Based on the selected key influencing indicators, multiple regression analysis, nonlinear fitting, or other statistical modeling methods are used to establish a quantitative mathematical relationship equation between process parameters and simulation output data (such as cross-sectional dimensions). The consistency between the model prediction results and the simulation data is verified, and the model parameters are further adjusted to ensure that the quantitative relationship equation can accurately reflect the evolution of the cross-sectional dimensions of the billet under different process parameter combinations. For example, a multiple nonlinear regression equation is constructed based on the selected principal components:

[0145] ;

[0146] Where W is the outlet width (mm); W0 is the initial width (mm); k1 is the reduction-temperature coefficient (mm 2 / ℃); k2 is the speed coefficient (mm); k3 is the diameter coefficient (mm -1 );v ref Indicates the reference speed (m / s).

[0147] The model achieved a prediction error of ±0.28mm (1.1% relative error) on the validation set, a 62% reduction compared to traditional linear models. For example, when inputted with T = 950°C, Δh = 28mm, v = 1.6m / s, and D = 600mm, the predicted outlet width was 52.1mm, while the measured value was 51.8mm, a deviation of only 0.3mm.

[0148] Through this quantitative relationship equation, process personnel can quickly reverse-solve the parameter combination required to achieve the target size. For example, to achieve Φ50±0.2mm round steel, the recommended optimization parameters are T=1020℃, Δh=22mm, v=1.3m / s, and D=580mm. Actual production verification shows that the qualified rate of finished products has increased from 88% to 96%.

[0149] S34. Set the blank size accuracy, strain uniformity and surface integrity as process optimization goals, establish the optimization objective function and constraint conditions, and solve to obtain the optimal process parameter combination.

[0150] Specifically, the process optimization objectives and optimal process parameter combinations in the above steps are essentially based on multi-physical field coupling analysis. By quantifying the complex relationship between material deformation laws and process parameters, scientific and precise control of the rolling process is ultimately achieved.

[0151] The optimization goal is the quantitative direction of process improvement, which includes the following core indicators:

[0152] 1. Billet size accuracy: The deviation range between the actual size of the final rolled product and the designed size (such as diameter tolerance ±0.2mm) needs to be minimized through process parameter control.

[0153] 2. Strain uniformity: The equivalent strain difference coefficient of different areas (corners, surface, and core) inside the rolled piece (for example, the strain at the corner is 0.8, the strain at the core is 0.5, and the difference coefficient is 0.3). The goal is to reduce local strain concentration and avoid cracking or uneven structure.

[0154] 3. Surface integrity: indicators such as surface roughness of rolled parts (Ra≤6.3μm), microcrack density (such as ≤2 / cm²), etc., are controlled by coordinated control of temperature and reduction to avoid scale intrusion or mechanical damage.

[0155] The optimal parameter combination refers to the set of process parameters that achieves the best overall performance of the above optimization objectives while meeting equipment capabilities, energy consumption constraints, and material properties through a multi-objective optimization algorithm. It usually includes:

[0156] 1. Temperature parameters: rolling temperature of each pass (e.g. first pass 1050℃ → last pass 920℃);

[0157] 2. Deformation parameters: pass reduction (e.g. Δh = 25mm → 18mm → 12mm);

[0158] 3. Speed ​​parameters: roller linear speed (e.g. 1.2m / s→1.5m / s→1.8m / s);

[0159] 4. Equipment parameters: roller crown (e.g. 0.1mm), cooling water flow (e.g. 40m³ / h).

[0160] The optimization method is to solve the inefficiency problem of traditional empirical trial and error method through multidisciplinary model coupling and data-driven optimization.

[0161] In the description of the present invention, the optimization objective function is:

[0162] ;

[0163] Where, f1 represents dimensional accuracy; f2 represents strain uniformity; f3 represents surface integrity; represents the simulation size of the i-th cross-section position; represents the target size of the i-th section position; represents the equivalent effect standard deviation; n represents the total number of cross-sectional locations; represents the equivalent effect mean; ε surface represents the average equivalent strain of the surface transition zone; ε core represents the average equivalent strain of the core stable area; Ra represents the surface roughness; N defect Indicates the number of defects per meter; ω1 indicates the roughness weight coefficient; ω2 indicates the defect number weight coefficient, and ω1+ω2=1. Both the roughness weight coefficient and the defect number weight coefficient have units. Ra and N are respectivelydefect Convert to dimensionless data.

[0164] In the description of the present invention, constraints include the following aspects:

[0165] 1. Second flow conservation constraint:

[0166] ;

[0167] Where A0 represents the rolling inlet cross-sectional area; v0 represents the inlet rolling speed; A1 represents the rolling outlet cross-sectional area; v1 represents the outlet rolling speed.

[0168] 2. Equipment safety constraints:

[0169] ;

[0170] Where M 轧制 Indicates the actual rolling torque; M 额定 Indicates the rated load moment of the rolling mill.

[0171] 3. Temperature gradient constraint:

[0172] ;

[0173] Where, Represents the temperature gradient in the corner sensitive area.

[0174] 4. Tissue uniformity constraints:

[0175] ;

[0176] Where, X DRX represents the dynamic recrystallization volume fraction.

[0177] S4. Establish real-time mapping between virtual and real data, build a process knowledge base and regularly study and update it.

[0178] Specifically, building a system for real-time mapping of virtual and real data and dynamically updating the process knowledge base requires designing a closed-loop data flow architecture that spans physical and digital spaces. An Industrial Internet of Things (IIoT) sensor network deployed on the rolling mill production line collects real-time physical parameters such as rolling speed (±0.05 m / s accuracy), rolling force (±25 kN), and temperature distribution (±5°C). Every 100 milliseconds, a data packet containing a 128-dimensional feature vector is generated and transmitted via a 5G private network to an edge computing node for data alignment. After receiving the physical data, the synchronously running digital twin uses dynamic constitutive equations in the virtual space to modify simulation parameters in real time. The strain rate sensitivity coefficient, C, is dynamically updated based on online data, with an update frequency set to every 30 seconds. In one case, when a sudden 15% increase in rolling force was detected, the digital twin completed parameter inversion within 5 seconds, identifying that the roll eccentricity exceeded the standard by 0.12 mm and triggering correction instructions, reducing the dimensional fluctuation of the rolled product from ±0.8 mm to ±0.3 mm.

[0179] The process knowledge base is constructed using a graph neural network (GNN) and knowledge graph fusion architecture. Historical process data (100,000 rolling records), simulation results (5TB of finite metadata), and expert experience rules (2,000 if-then logic) are encoded into a multidimensional feature vector. Knowledge nodes include three types of entities: process parameters (temperature, reduction, etc.), material responses (strain field, recrystallization rate, etc.), and equipment status (roller wear, motor load, etc.). The TransE algorithm learns the relationship weights between these entities. For example, the relationship weight for "rolling temperature-dynamic recrystallization fraction" reaches 0.92 after training, indicating a strong correlation. When a new process solution (such as a rolling parameter combination for Φ60mm round steel) is generated, the system automatically traverses similar nodes in the knowledge graph and recommends the baseline parameters of v = 1.6m / s and Δh = 25mm. By comparing 20 sets of historical similar cases, the predicted width expansion deviation is no more than 0.4mm.

[0180] The dynamic update mechanism employs an online incremental learning strategy and a dual-buffered storage area to achieve separate data-knowledge updates. Physical data flows through a sliding time window (60-second window size, 10-second step size) for feature extraction, triggering a local update of the knowledge base. For continuous parameters (such as the temperature compensation coefficient), exponential smoothing is used to update weights. For discrete rules (such as the defect warning threshold), hypothesis testing (p < 0.05) determines whether to include them in the knowledge base. During a rolling process optimization, the system analyzed 300 sets of new production data and found that when the rolling speed exceeded 1.8 m / s, the probability of surface cracks deviated by 18% from theoretical predictions. The system automatically adjusted the speed sensitivity coefficient in the crack warning model from 0.25 to 0.31, increasing warning accuracy from 82% to 94%. Furthermore, a knowledge decay mechanism is established to automatically downgrade process rules that have been unreferenced for more than their validity period (e.g., 6 months) to ensure the currency of the knowledge base.

[0181] In summary, with the help of the above-mentioned technical scheme of the present invention, by adopting finite element software, the rough rolling deformation law of different cross-section billets can be simulated more accurately, so as to more accurately predict and control the shape and size of the rolled material, and improve the accuracy and reliability of the optimization results; by identifying different hole parameters, stress and strain characteristics of the rolled material corners and the surface to the core under different reduction amounts, and the rolled cross-sectional size and area after deformation, the rough rolling hole system of multi-blank continuous casting square billets is fully optimized, meeting the needs of various different cross-section billets and improving production efficiency; compared with the existing methods, the calculation efficiency is higher, and for complex hole systems, simulation and optimization can also be completed under reasonable computing resources, reducing the optimization cost, while having higher accuracy, stronger adaptability and higher efficiency, which can better meet the needs of modern steel production.

[0182] By building an intelligent computing architecture and hybrid modeling technology, breakthroughs in both computational efficiency and optimization accuracy were achieved. In terms of computational efficiency, GPU parallel computing technology was used to decompose the rolling process into three stages (bite, steady state, and steel throwing), reducing the simulation time for a single pass from 4-6 hours with traditional methods to less than 30 minutes. This also improved the efficiency of searching in high-dimensional parameter spaces and shortened the optimization cycle. In terms of precision control, the metal flow conservation equation was rigidly embedded into the machine learning model through a physical information neural network, addressing the lack of physical rationality of traditional data-driven models.

[0183] Through the multi-physical field coupling analysis and the dynamic management of the process knowledge base, the quality consistency and process controllability of multi-billet rolling are comprehensively improved. First, based on the full coupling model of temperature field-stress field-microstructure field, the cross-scale prediction from macro deformation to microstructure is realized, and the differential deformation characteristics of the corner, surface and core are accurately quantified. Second, the process knowledge base constructed stores 2000+ working condition data, combined with the similarity matching algorithm, provides the historical optimal parameter recommendation for new billet, greatly reduces the experimental data required for new specification development.

[0184] It should be understood that although each step in the flowchart of the accompanying drawings is shown in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless explicitly stated herein, the execution of these steps is not strictly limited in sequence, and they can be executed in other sequences. Moreover, at least part of the steps in the flowchart of the accompanying drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

Claims

1. A method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software, characterized in that: include: S1. Based on the principle of equal metal flow rate per second, a digital model of the dynamic balance of volume flow during rolling of different cross-section billets is established to quantify the deformation law of billet rough rolling; S2. Use the finite element simulation model to model the hole pattern for multi-blank switching, build a collaborative computing architecture, and dynamically deploy finite element calculation scheduling tasks; S3. Based on multi-physics field coupling analysis, predict the stress-strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions, generate process optimization targets, and meet the requirements of blanks with different cross-sections; S4. Establish real-time mapping between virtual and real data, build a process knowledge base and regularly study and update it; Based on the principle of equal metal second flow rate, a digital model of dynamic balance of volume flow rate during rolling of different cross-section billets is established to quantify the deformation law of rough rolling of square billets, including: S11. Set the geometric features of various billet sections, build a parametric template library, and establish a mapping relationship between billet geometric parameters and hole system; S12. Based on the principle of equal metal flow rate per second, a dynamic balance equation for volume flow in multi-pass rolling is established, a mathematical model for the relationship between width expansion and reduction is constructed, and a shape factor correction term is introduced to quantify the metal flow differences of different billet shapes; S13. Combine historical rolling data to establish a statistical relationship model between rolling force, torque and pass parameters, and integrate and summarize the deformation laws of rolling under different cross-section billet conditions; The method of using the finite element simulation model to model the hole pattern of multi-blank switching, building a collaborative computing architecture, and dynamically deploying finite element calculation scheduling tasks includes: S21, extracting geometric parameters of the billet, inputting them into a finite element simulation model, generating a hole profile curve through a non-uniform rational B-spline, matching an initial hole parameter combination in a billet rule library, and generating a parametric model of the billet hole; S22. Build a cognitive grid in the finite element simulation model and optimize the grid topology; S23. Divide the rolling process into several stages, match corresponding parallel acceleration strategies, and build a quantum-classical-edge collaborative computing architecture to collaboratively optimize computing resources; S24. Allocate the finite element calculation task to the optimal calculation node according to the finite element calculation load; The multi-physics field coupling analysis is used to predict the stress and strain characteristics and dimensional evolution of the blank under multi-dimensional parameter conditions, and to generate process optimization goals including: S31. Construct a fully coupled model of temperature field, stress field and microstructure field, and solve it through distributed coupling to show the influence, distribution and transformation of each physical field; S32. Based on the regional division standard of the billet shape, quantify the differential deformation characteristics of the billet at the corners, surface and core, and evaluate the rough rolling defects of the billet; S33. Based on the pre-established material constitutive relationship adapted to multiple working conditions, a quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established; S34. Set the blank size accuracy, strain uniformity and surface integrity as process optimization goals, establish the optimization objective function and constraint conditions, and solve to obtain the optimal process parameter combination.

2. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 1, characterized in that: The step of building a cognitive grid in a finite element simulation model and optimizing the grid topology includes: S221, after the geometric parameters of the billet are input, an optimized mesh topology is generated, and the mesh is divided into a corner area, a surface area, and a core area according to the positions of the corners, surface, and core of the billet; S222. Generate a seven-layer progressively finer mesh in the corner area, use anisotropic elements in the surface area to capture friction effects, and configure a coarse mesh in the core area. S223. Deploy a stress gradient sensitivity detector in the grid. When the stress gradient sensitivity is greater than a preset sensitivity threshold, trigger local grid encryption and insert high-density monitoring points in the potential defect area.

3. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 1, characterized in that: The rolling process is divided into several stages, matched with corresponding parallel acceleration strategies, and a quantum-classical-edge collaborative computing architecture is built to collaboratively optimize computing resources, including: S231, dividing the billet rolling process into a biting stage, a steady-state stage, and a steel throwing stage; S232. Configure a multi-stage parallel acceleration strategy. During the biting phase, assign the finite element calculation task to the GPU's stream multiprocessor. A single stream multiprocessor handles the calculation of one roll contact point. During the steady-state phase, enable multi-process services to simultaneously calculate the temperature-deformation coupling field of multiple passes. During the casting phase, use the GPU memory to directly store deformation history data. S233. Build a quantum-classical-edge collaborative computing architecture that integrates quantum computing, classical computing, and edge computing functions to achieve collaborative optimization of heterogeneous computing resources.

4. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 3, characterized in that: The collaborative computing architecture includes a quantum computing layer, a classical computing layer, and an edge computing layer; Among them, the quantum computing layer converts the hole parameter optimization problem into the Ising model and defines the Hami variable; the classical computing layer is deployed in a high-performance computing cluster, and each node is used for calculation in a predetermined stage; the edge computing layer deploys a deep neural network model at the rolling mill site to predict the rolling force and width expansion in real time.

5. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 1, characterized in that: The regional division standard based on the billet shape quantifies the differential deformation characteristics of the billet at the corners, surface and core, and evaluates the rough rolling defects of the billet, including: S321. Divide the cross section into a corner sensitive area, a surface transition area, and a core stable area according to the locations of the corners, surface, and core of the billet, and select differentiated monitoring indicators for each area. S322. Based on the cognitive grid of the finite element simulation model, extract the equivalent strain distribution data of each region, calculate the local gradient vector, and set the ratio of the maximum strain gradient to the average strain as the strain gradient index through normalization processing; S323, uniformly selecting N monitoring points in each area, recording the temperature value in the steady-state rolling stage, calculating the temperature range and average temperature of each area, and setting the ratio of the temperature range to the average temperature as the temperature uniformity coefficient; S324. Match similar strain gradient indices and temperature uniformity coefficients in the historical database, and evaluate the process defects in the current billet rolling through threshold judgment.

6. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 1, characterized in that: The quantitative relationship between the cross-sectional dimensions of the billet after rolling and various process parameters is established based on the pre-established material constitutive relationship adapted to multiple working conditions, including: S331, extracting process parameters related to the cross-sectional dimensions of the billet and defining the parameter range; S332. Load the dynamic constitutive equation into the finite element simulation model, implement thermomechanical coupling calculation through the temperature-strain rate coupling field, and use orthogonal experiments to generate several sets of process parameter combinations. Perform full-process rolling simulation for each set of process parameters and output simulation data. S333. Standardize and perform principal component analysis on the simulation data, screen the influencing indicators, and construct a quantitative relationship equation between the process parameters and the simulation data.

7. The method for optimizing the rough rolling pass system of multi-bill continuous casting billets based on finite element software according to claim 1, characterized in that: The optimization objective function is: Where, f1 represents dimensional accuracy; f2 represents strain uniformity; f3 represents surface integrity; represents the simulation size of the i-th cross-section position; represents the target size of the i-th cross-sectional position; σ(ε) represents the equivalent standard deviation; n represents the total number of cross-sectional positions; represents the equivalent effect mean; ε surface represents the average equivalent strain of the surface transition zone; ε core represents the average equivalent strain of the central stable region; Ra represents surface roughness; N defect Represents the number of defects per meter; ω1 represents the roughness weight coefficient; ω2 represents the defect number weight coefficient.

Images