A method of extrusion process parameter optimization and related apparatus

Abstract

The application discloses an extrusion process parameter optimization method and related device, and relates to the technical field of metal forming process optimization and intelligent manufacturing. First, a deep neural network is used to process to-be-optimized process parameters, and an outer shape prediction result of a tubular part is obtained. A simulation result of the tubular part and the outer shape prediction result are input into an interpolation model, and an interpolated uniform regular grid is obtained. Then, the interpolated uniform regular grid is input into an intrinsic field quantity prediction model, and a field quantity prediction result is obtained. Finally, the to-be-optimized process parameters and the field quantity prediction result are input into a fast simulation proxy model, and an optimized process parameter set of the tubular part is obtained. Through the fast simulation proxy model, the optimized process parameters of the tubular part can be quickly calculated, and the efficiency of optimizing the extrusion process is greatly improved.

Description

Technical Field

[0001] This application relates to the fields of metal forming process optimization and intelligent manufacturing technology, and in particular to a method and related apparatus for optimizing extrusion process parameters. Background Technology

[0002] Hot extrusion is a manufacturing process in which metal material is processed to above its recrystallization temperature and then forced through the orifice of an extrusion die under pressure to obtain long or hollow profiles with specific cross-sectional shapes and dimensions. Understandably, as a key technology in metal forming, the product quality of hot extrusion is significantly affected by process parameters. However, traditional methods for optimizing extrusion processes are often based on empiricism: trial and error. This approach has the following problems:

[0003] Relying on experience-based trial and error is inefficient: Traditional optimization methods often rely on empirical parameter adjustments, verifying parameter combinations through repeated trial runs. This results in long trial cycles, high costs, and low efficiency. Furthermore, the simulation of hot extrusion processes involves thermo-elastic-plastic deformation, as well as calculations related to materials, geometry, and contact nonlinearities. These calculations typically require process simulation software. However, the calculation and simulation of precise metal flow behavior are time-consuming, and the efficiency of process tuning is low, making it difficult to meet the needs of rapid optimization design of forging process parameters, especially when the number of parameters involved is very large and the range of parameter variations is wide.

[0004] This shows that the efficiency of optimizing the extrusion process in existing technologies is relatively low. Summary of the Invention

[0005] In view of the above problems, this application provides a method and related apparatus for optimizing extrusion process parameters to improve the efficiency of the extrusion process. The specific solution is as follows:

[0006] The first aspect of this application provides a method for optimizing extrusion process parameters, including:

[0007] The set of process parameters to be optimized for tubular parts is input into a deep neural network model to obtain the shape prediction results of tubular parts; the set of process parameters to be optimized is: the set of key sensitive process parameters obtained after sampling the simulation results of the extrusion process of tubular parts and performing parameter sensitivity analysis.

[0008] The simulation results and shape prediction results are input into the interpolation model to obtain the interpolated uniform regular grid. The interpolation model is used to: interpolate the field quantity data of different unstructured grids with different process parameters to the uniform regular grid to be interpolated, and obtain the interpolated uniform regular grid. The different unstructured grids with different process parameters are obtained by discretizing the simulation results.

[0009] The interpolated uniform grid is input into the intrinsic field quantity prediction model to obtain the field quantity prediction results;

[0010] The set of process parameters to be optimized and the field quantity prediction results are input into the fast simulation proxy model to obtain the optimized process parameter set of the tubular part. The fast simulation proxy model is used to process the process parameters to be optimized in the set of process parameters to be optimized through the nonlinear mapping relationship between process parameters and objective function to obtain the optimized process parameter set.

[0011] In one possible implementation, the shape prediction results include the displacement results of the punch end of the billet and the displacement results of the free end of the tubular part; the field data of different unstructured meshes with different process parameters are interpolated to the uniform regular mesh to be interpolated, resulting in the interpolated uniform regular mesh, including:

[0012] The grid coordinates of the uniform regular grid to be interpolated are updated using the displacement results of the punch end and the displacement results of the free end.

[0013] The field quantity data of each grid of different unstructured grids with different process parameters are interpolated to each grid of the uniform regular grid to be interpolated.

[0014] In one possible implementation, the intrinsic field quantity prediction model is a prediction model that integrates the intrinsic orthogonal decomposition model and the modal participation coefficient solution model. The interpolated uniform grid is input into the intrinsic field quantity prediction model to obtain the field quantity prediction results, including:

[0015] The intrinsic orthogonal decomposition model is used to process the field quantity data in the interpolated uniform regular grid into low-order linear modal field quantity data;

[0016] The modal participation coefficients required for predicting field quantity data are calculated using the intrinsic orthogonal decomposition model by solving the model using modal participation coefficients.

[0017] The intrinsic orthogonal decomposition model calls the modal participation coefficient calculation model to calculate the modal participation coefficients, and then solves for the field quantity prediction results.

[0018] In one possible implementation, the modal participation coefficient solution model includes a backpropagation neural network model;

[0019] The backpropagation neural network model is used to update the modal participation coefficients in the intrinsic orthogonal decomposition model, and also to predict scalar data for tubular components.

[0020] In one possible implementation, the fast simulation surrogate model is a nonlinear surrogate model between process parameters and the objective function, constructed based on the Kriging model.

[0021] The set of process parameters to be optimized and the field quantity prediction results are input into the fast simulation surrogate model to obtain the optimized process parameter set for the tubular component, including:

[0022] The field quantity prediction results are input into the Kriging model to obtain the nonlinear mapping relationship between the process parameters to be optimized and the objective function in the set of process parameters to be optimized.

[0023] The objective function values ​​of the process parameters to be optimized are calculated by combining the expected improvement criteria. The objective function values ​​are the optimized process parameters.

[0024] In one possible implementation, before inputting the set of process parameters to be optimized for the tubular component into the deep neural network model, the following steps are also included:

[0025] A 2D axisymmetric equivalent model of the tubular component was established using numerical simulation tools;

[0026] The process parameters input during the establishment of the 2D axisymmetric equivalent model are sampled to obtain a set of process parameters;

[0027] Parameter sensitivity analysis is performed on each process parameter in the process parameter set to obtain the set of process parameters to be optimized.

[0028] A second aspect of this application provides an extrusion process parameter optimization device, comprising:

[0029] The shape prediction unit is used to input the set of process parameters to be optimized for tubular parts into a deep neural network model to obtain the shape prediction results of tubular parts; the set of process parameters to be optimized is: the set of various key sensitive process parameters obtained after sampling the simulation results of the extrusion process of tubular parts and performing parameter sensitivity analysis;

[0030] The interpolation unit is used to input the simulation results and shape prediction results into the interpolation model to obtain the interpolated uniform regular grid. The interpolation model is used to: interpolate the field quantity data of different unstructured grids with different process parameters to the uniform regular grid to be interpolated, and obtain the interpolated uniform regular grid. The different unstructured grids with different process parameters are obtained by discretizing the simulation results.

[0031] The field quantity prediction unit is used to input the interpolated uniform regular grid into the intrinsic field quantity prediction model to obtain the field quantity prediction results.

[0032] The process parameter optimization unit is used to input the set of process parameters to be optimized and the field quantity prediction results into the fast simulation proxy model to obtain the optimized process parameter set of the tubular part. The fast simulation proxy model is used to process the process parameters to be optimized in the set of process parameters to be optimized through the nonlinear mapping relationship between process parameters and objective function to obtain the optimized process parameter set.

[0033] A third aspect of this application provides an extrusion process parameter optimization device, comprising at least one processor and a memory connected to the processor, wherein:

[0034] Memory is used to store computer programs;

[0035] The processor is used to execute computer programs so that the extrusion process parameter optimization equipment can implement any of the extrusion process parameter optimization methods described above.

[0036] A fourth aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement any of the extrusion process parameter optimization methods described above.

[0037] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement any of the extrusion process parameter optimization methods described above.

[0038] By employing the above technical solution, the extrusion process parameter optimization method and related apparatus provided in this application first use a deep neural network to process the process parameters to be optimized, obtaining the predicted shape of the tubular part. The simulation results and shape prediction results of the tubular part are then input into an interpolation model to obtain an interpolated uniform grid. Next, the interpolated uniform grid is input into an intrinsic field prediction model to obtain field prediction results. Finally, the process parameters to be optimized and the field prediction results are input into a rapid simulation proxy model to obtain the optimized process parameter set for the tubular part. This application, through a rapid simulation proxy model, can quickly calculate the optimized process parameters for the tubular part, greatly improving the efficiency of optimizing the extrusion process. Attached Figure Description

[0039] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0040] Figure 1 A schematic flowchart illustrating the extrusion process parameter optimization method provided in this application;

[0041] Figure 2 Example diagram of the tubular extrusion process provided in this application;

[0042] Figure 3 An example diagram of displacement in the extrusion process of the tubular parts provided in this application;

[0043] Figure 4Example diagram of the structure of the deep neural network model provided in this application;

[0044] Figure 5 Example diagrams of interpolation effects provided in this application;

[0045] Figure 6 A schematic diagram of an extrusion process parameter optimization device provided in this application;

[0046] Figure 7 A schematic diagram of the extrusion process parameter optimization equipment provided in this application. Detailed Implementation

[0047] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0048] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0049] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0050] Existing technologies generally employ traditional experimental methods to optimize extrusion process parameters, but this approach has the following problems:

[0051] High time and cost: Taking aluminum extrusion as an example, it is necessary to establish a relationship model between process parameters and forming quality through multiple rounds of trial and error. A single test can take several hours to several days, and material loss can reach 10%-15%. The length of the blank is limited during positive extrusion, resulting in a low yield.

[0052] Limited predictive capability: Taking the extrusion of nickel-based superalloys as an example, the influence of extrusion and die angle on the temperature field requires fitting with a large amount of experimental data, but it cannot accurately predict the dynamic recrystallization behavior. Defects such as shrinkage cavities and dead zone folding are prone to occur during the extrusion process, thus demonstrating that traditional experimental methods are difficult to quantify and model.

[0053] Significant environmental interference: Taking the extrusion of 5A06 aluminum alloy ribbed box body as an example, the temperature fluctuation of the mold will cause uneven filling of the bottom rib, which requires repeated adjustments to optimize.

[0054] In addition, existing optimization schemes based on simulation also have many problems:

[0055] Taking a simulation solution based on commercial software as an example, the main problems are as follows:

[0056] High computational resource consumption: The simulation of cladding extrusion of nickel-based powder superalloys requires dividing millions of mesh elements, and a single fully coupled field simulation can take several hours, and it requires high-performance computing clusters; the simulation of aluminum profile extrusion requires repeated adjustment of structural parameters such as diversion holes and feeding grooves, which extends the development cycle by 3-5 times.

[0057] Parameter boundaries rely on experience: Optimization algorithms built into commercial software, such as gradient descent, require manual pre-setting of parameters like extrusion speed and die clearance to search the space, making them prone to getting trapped in local optima. For example, excessively high extrusion speeds can lead to coarse grains, but traditional methods struggle to automatically identify critical thresholds.

[0058] Inefficient iterative process: In optimizing cold extrusion forming process parameters, orthogonal experiments require designing nine schemes, each requiring multiple simulation verifications, taking several weeks. Differences in metal flow velocity within the extrusion cylinder can lead to the incorrect merging of coarse grain areas in the cavity-sensitive region, affecting the effectiveness of process parameter adjustments.

[0059] Another existing method for optimizing extrusion process parameters is the forging optimization region segmentation method, which has the following problems:

[0060] The segmentation method based on mass / volume equivalence has the following problems:

[0061] Lack of correlation between flow velocity and stress fields: The spatiotemporal distribution characteristics of the material flow velocity field and stress field are not considered. For example, differences in metal flow velocity within the extrusion cylinder can lead to the incorrect merging of shrinkage cavity sensitive areas and grain coarsening, affecting the effectiveness of process parameter adjustments.

[0062] Insufficient dynamic adaptability: Fixed segmentation regions are difficult to match the transient characteristics of extrusion deformation. For example, in multi-stage extrusion processes, the stress distribution of preceding processes can significantly alter the optimization target area of ​​subsequent processes, which traditional methods cannot dynamically adjust.

[0063] Low precision in process parameter adjustment: Traditional methods can only allow process parameters to fluctuate up or down by 5% from the baseline value, while field-quantity coupling analysis shows that even a 0.1mm fine adjustment of key parameters can lead to significant changes in forming quality.

[0064] In summary, existing technologies for optimizing extrusion process parameters face numerous challenges. Traditional experimental approaches struggle to handle complex coupled scenarios, while simulation methods are limited by computational resources and reliance on experience. Furthermore, optimization region segmentation methods fail to integrate the spatiotemporal characteristics of field quantities, leading to defects such as shrinkage cavities, folds, and grain inhomogeneity. A common problem among existing process parameter optimization methods is low efficiency.

[0065] To address the aforementioned problems, this application provides a method and related apparatus for optimizing extrusion process parameters.

[0066] Optional, see Figure 1 This application provides a schematic flowchart of a method for optimizing extrusion process parameters.

[0067] like Figure 1 As shown, the method for optimizing extrusion process parameters includes the following steps:

[0068] Step 101: Input the set of process parameters to be optimized for the tubular part into the deep neural network model to obtain the shape prediction result of the tubular part.

[0069] The core focus of hot extrusion process parameters for tubular parts lies in the precise control of material rheological properties and multi-physics coupling mechanisms. Process optimization requires comprehensive consideration of the influence of the following key parameters:

[0070] Material mechanical parameters, including the elastic modulus, yield strength, and tensile strength, directly affect deformation resistance and flow stress. For example, the yield strength of 7075 aluminum alloy is 465-500 MPa, and its tensile strength is 500-600 MPa. Its mechanical properties vary significantly with deformation temperature and extrusion ratio.

[0071] Fracture criterion parameters: used to predict the fracture behavior of materials during deformation, such as the maximum shear stress theory and dynamic recrystallization model. For example, during the extrusion of GH3625 alloy tubing, the critical melting temperature of the Laves phase is >1150℃, and the crack propagation threshold is directly related to the stress concentration factor.

[0072] Constitutive parameters: Mathematical models describing the stress-strain-temperature coupled behavior of materials, such as the Arrhenius equation. For example, in the constitutive equation of ZK60 magnesium alloy, the stress exponent is 9.45 and the activation energy is 120 kJ / mol, and these parameters are determined through hot compression tests.

[0073] Temperature and boundary condition parameters, including billet temperature, mold preheating temperature, lubrication conditions, etc., affect heat conduction and friction. For example, when extruding TA2 titanium alloy, the temperature difference needs to be controlled to <40℃ when the phase transformation temperature is below 904.6 degrees Celsius, and the mold temperature of 300-600℃ can reduce the extrusion pressure.

[0074] Punch stroke parameters: These determine the metal filling capacity and the degree of deformation, and are strongly related to the extrusion ratio. For example, in two-stage extrusion, the punch stroke distribution needs to balance the stress distribution of the preceding filling and the subsequent stress, which can improve the shrinkage cavity suppression efficiency by 40%.

[0075] The above parameters will affect the forming quality through a multi-field coupling mechanism, such as temperature field, velocity field and stress field. For example, when TA2 alloy is extruded at 870℃, the combination of temperature gradient <15℃ / mm, die temperature 550℃ and extrusion speed 10mm / s can make the grain size 35-55μm and the tensile strength 520MPa.

[0076] The parameters mentioned above are the process parameters to be optimized. It can be understood that the set of process parameters to be optimized includes, but is not limited to, the parameters mentioned above.

[0077] Next, we will describe the process of obtaining the set of process parameters to be optimized.

[0078] Step 1: Use numerical simulation tools to establish a 2D axisymmetric equivalent model of the tubular component.

[0079] This process mainly involves simulating the hot extrusion process of tubular parts. Specifically, it can be CAE (Computer-Aided Engineering) simulation, which establishes a 2D axisymmetric equivalent model of the hot extrusion process of tubular parts and obtains the CAE result file.

[0080] The reason for establishing a 2D axisymmetric equivalent model of the hot extrusion process of tubular parts during the simulation is as follows:

[0081] Plastic forming is a nonlinear process, resulting in a large computational load and long simulation time. Considering that most forged parts have axisymmetric characteristics, 2D axisymmetric models can accurately reflect the geometric features of forged parts, reduce the number of computational nodes and elements, thereby reducing computational resource requirements and simulation time. This makes them suitable for applications requiring rapid iteration, preliminary design, and large simulation databases.

[0082] Numerical simulation tools can also be called finite element analysis software, specifically Deform or a self-developed solver.

[0083] Specifically, a 2D axisymmetric model simulation model of the tubular part is established using Deform or a self-developed solver to simulate the hot extrusion process of the tubular part. The dimensional changes of the tubular part in the hot extrusion process, automatic network drawing, and automatic batch calculation of the solver are realized through Python scripts to obtain CAE result files.

[0084] Step 2: Sample the process parameters input during the establishment of the 2D axisymmetric equivalent model to obtain the process parameter set.

[0085] In order to obtain the original data or dataset required for training the surrogate model for optimizing the extrusion process parameters in this application, it is necessary to sample the 2D axisymmetric simulation model of the hot extrusion process of tubular parts.

[0086] The sampling methods include, but are not limited to, the Sobol sequence method and LHS (Latin Hypercube Sampling).

[0087] Specifically, based on the actual design and production requirements of tubular parts, the key parameters such as blank size and boundary conditions are first parametrically modeled, and the sampling interval is scientifically allocated. Then, the Sobol sequence method or LHS method is used to sample the simulation input parameters with high efficiency to ensure that the sample space is fully covered and evenly distributed, thereby improving the statistical reliability and sensitivity analysis accuracy of the simulation results. The process parameter set is obtained through sampling.

[0088] The simulation input parameters can include punch speed, initial temperature field, billet size, extrusion ratio, distance between die mandrel and billet, etc. The simulation output results include data such as temperature field, strain field, and grain size.

[0089] Step 3: Perform parameter sensitivity analysis on each process parameter in the process parameter set to obtain the set of process parameters to be optimized.

[0090] Next, a Design of Experiments (DOE) analysis needs to be performed on each process parameter in the process parameter set. Specifically, a parameter sensitivity analysis is performed on each process parameter in the process parameter set to analyze the degree of sensitivity between input and output, thereby determining which parameters in the process parameter set are the key influencing parameters. The key influencing parameters in the process parameter set are then used as the set of process parameters to be optimized.

[0091] Specifically, the cloud platform is used to quickly perform DOE analysis on the process parameter set. The CAE result files are processed in batches using scripts. The data of physical fields such as equivalent strain and stress at all nodes of all time steps are processed and analyzed to identify sensitive parameters and select key parameters to form the process parameter set to be optimized.

[0092] In summary, the set of process parameters to be optimized is the set of key sensitive process parameters obtained after sampling the simulation results of the tubular extrusion process and performing parameter sensitivity analysis.

[0093] After obtaining the set of process parameters to be optimized, they are input into a pre-trained deep neural network model (DNN) to obtain the shape prediction results of the tubular parts.

[0094] It should be noted that the purpose of using the DNN surrogate model in this application is to predict the final shape parameters under different process parameters, mainly including the displacement results of the punch end and the displacement results of the free end of the tubular part.

[0095] For example, see Figure 2 Example diagram of the tubular extrusion process provided in this application.

[0096] After heating the metal billet to a suitable temperature, it is placed into an extrusion cylinder, which can be... Figure 2 The space where the moving mold is located is shown. The moving mold pushes the billet into the mold cavity through the coordinated movement of the extrusion shaft and the extrusion mandrel.

[0097] The extrusion shaft applies pressure, pushing the billet through the mold cavity. The punch end, the portion where the extrusion shaft applies pressure, directly pushes the billet into the mold cavity; the free end, the portion of the mandrel furthest from the extrusion shaft, can move freely or separate after extrusion. The relative movement between the punch end and the free end directly affects the metal flow direction and forming quality.

[0098] Correspondingly, see Figure 3 The displacement example diagram of the tubular extrusion process provided in this application is shown.

[0099] like Figure 3 The figure shows the displacement changes of the tubular part before and after extrusion, mainly showing the displacement of the punch end of the billet and the free end of the tubular part.

[0100] The DNN proxy model used in this application adopts a fully connected structure, which may include only one hidden layer in the middle.

[0101] For example, see Figure 4 The structural example diagram of the deep neural network model provided in this application is shown.

[0102] like Figure 4 As shown, the deep neural network model used in this application can be a deep neural network surrogate model including an input layer, a hidden layer and an output layer.

[0103] Optionally, the set of process parameters to be optimized can be input into a deep neural network model to obtain the displacement of the punch end of the billet and the displacement of the free end of the tubular part.

[0104] Among them, the displacement of the blank punch end and the displacement of the free end of the tubular part are essentially coordinate values ​​of spatial points. Specifically, they can be replaced by the grid under steady-state results in the entire extrusion deformation region of the tubular part, as well as the maximum and minimum y-coordinate values ​​of the data.

[0105] It is understandable that the shape prediction results output by the deep neural network model are essentially the x and y coordinates of spatial points, specifically represented as: (x1, y1), (x2, y2), (x3, y3), (x4, y4), (x5, y5), etc.

[0106] It should also be noted that when using a deep neural network model for shape prediction, the line boundaries can be extracted from the 2D mesh data. The line boundaries are discretized using NURBS (Non-Uniform Rational B-Spline) or B-spline curves to describe the control point parameters. The control point parameters are then incorporated into the output parameters of the punch end displacement and free end displacement for prediction by the DNN model.

[0107] Step 102: Input the simulation results and shape prediction results into the interpolation model to obtain the uniform regular grid after interpolation.

[0108] The simulation results are those for the extrusion process of tubular parts mentioned above, and can be specifically represented as CAE result files. The shape of the tubular part changes during extrusion. In this application, a deep neural network model is primarily used to predict the shape of the tubular part after extrusion. While the shape of the tubular part changes during extrusion, the field data in the entire space containing the tubular part also change, such as displacement and temperature values.

[0109] Specifically, for finite element simulation in the structural domain, i.e., CAE simulation mentioned above, unstructured meshes are generally used to discretize the entire physical domain, and the physical space where the tubular part is located is discretized into small unstructured meshes. The mesh distortion of the tubular part under extrusion simulation is very large. Due to the large deformation of the original unstructured mesh during the forging process, the mesh needs to be re-divided. This results in the base mesh of the final simulation results under different input parameters being inconsistent. The mesh points and the number of meshes are inconsistent. Specifically, the original mesh before distortion and the mesh after distortion are not the same set.

[0110] It is understandable that in the original simulation data, the unstructured field results under different process parameters correspond to different unstructured meshes, which leads to inconsistent dimensionality of the field results under different process parameters.

[0111] The solution proposed in this application is to use a uniform and regular structured grid to unify the dimension of the field quantity results under different process parameters. Specifically, the field quantity data of different unstructured grids with different process parameters are interpolated into the uniform and regular grid to be interpolated to obtain the interpolated uniform and regular grid.

[0112] Specifically, the unstructured grid is interpolated onto the structured grid. In particular, the field data at the unstructured grid points are interpolated to other locations in space. These other locations in space are predetermined, relatively regular and uniform structured grids, i.e., the uniform and regular grids to be interpolated.

[0113] The interpolation process can be implemented based on an interpolation model, and the specific operation of the interpolation model can be as follows:

[0114] First, based on the displacement of the blank punch end and the displacement of the free end of the tubular part predicted by the deep neural network model above, the grid coordinates of the uniform regular grid to be interpolated are updated.

[0115] Then, the field data of each grid point of different unstructured grids with different process parameters in the simulation results of the tubular part are interpolated to the corresponding grids in the uniform regular grid to be interpolated. The initial unstructured temperature field and displacement structure are interpolated to the corresponding grids. The initial unstructured temperature field and displacement structure mentioned here are obtained by simulation using a 2D axisymmetric equivalent model, which is the simulation result of the tubular part extrusion process mentioned above.

[0116] It should also be noted that interpolation can be performed in sections to improve the overall interpolation accuracy.

[0117] For example, see Figure 5 Example diagrams of interpolation effects provided in this application.

[0118] like Figure 5 As shown, the left side shows an unstructured mesh, and the right side shows the interpolation effect of a structured mesh. The mesh topology and number of points of the structured mesh are fixed, and the coordinates of its network nodes change with the shape.

[0119] The interpolated structured mesh is obtained by interpolating the variable values ​​on the original mesh onto a new, uniform mesh.

[0120] It should be noted that the uniform grid after interpolation is the basis for the intrinsic field quantity prediction model to predict field quantities. Specifically, the intrinsic field quantity prediction model uses the nodal values ​​on the uniform grid after interpolation as the basis for field quantity prediction.

[0121] Step 103: Input the interpolated uniform grid into the intrinsic field quantity prediction model to obtain the field quantity prediction results.

[0122] The intrinsic field quantity prediction model is mainly used to predict the field quantity data after the extrusion molding of tubular parts, which can specifically include the displacement value, temperature value, etc. at each grid point.

[0123] The intrinsic field quantity prediction model provided in this application is a prediction model that integrates the intrinsic orthogonal decomposition model and the modal participation coefficient solution model. The intrinsic orthogonal decomposition model is a POD (Print Orthogonal Decomposition) model, a physical field proxy model established using the POD method. POD is an intrinsic decomposition calculated using SVD (Singular Value Decomposition) to obtain eigenvalues ​​and eigenvectors. Then, the field quantity results in the physical space are reduced to the modal space, which is a linearization process. Essentially, it extracts the main features from high-dimensional physical field data to establish a low-dimensional model to approximate the original system. By retaining the main modes and ignoring secondary modes, order reduction is achieved, and linear modal order reduction reduces the degrees of freedom.

[0124] The calculation process of the intrinsic orthogonal decomposition model can be as follows:

[0125] The snapshot matrix S is obtained based on the sampling points in the interpolated uniform grid:

[0126] ;

[0127] Where N is the number of sampling points, p is the dimension of the field quantity, and u(x) i ) represents the field result at the i-th sampling point, which can be a temperature field or a strain field, etc.

[0128] Then, the snapshot matrix is ​​decomposed using SVD, and the calculation formula is as follows:

[0129] ;

[0130] Where U is the left singular vector matrix, It is a singular value matrix. It is the transpose of the right singular vector matrix.

[0131] Take the first few columns of U to form the POD base:

[0132] ;

[0133] in, It is POD-based. Let M be the i-th POD mode, and M be the order of the mode space. M can be determined according to the minimum value of p with a tolerance of 0.99. The specific value needs to be determined according to the user's required efficiency.

[0134] The formula for the intrinsic orthogonal decomposition model provided in this application can be specifically as follows:

[0135] Where x represents the spatial location of the field quantity data. For field quantity data, M is the order of the modal space. These are predicted values ​​from the field volume data. This is the average value of the field quantity data. These are the scalar coefficients for each mode, also known as modal participation coefficients. These are spatial modal basis functions.

[0136] As can be seen from the formula of the intrinsic orthogonal decomposition model, the predicted value of the field quantity data can also be calculated by this formula. The unknown quantity in this formula is mainly the modal participation coefficient. Therefore, it is necessary to use the modal participation coefficient solution model to calculate the modal participation coefficient.

[0137] The modal participation coefficient (MRE) solution model provided in this application can be a BPNN (Back Propagation Neural Network) model. This model has strong nonlinear mapping capabilities and high-intensity self-learning and adaptive capabilities. The input layer is the process parameters to be optimized, the output layer is the modal participation coefficient, and the loss function is the MRE deviation value of the temperature field and displacement field results. The calculation formula is as follows:

[0138]

[0139] Where N is the sample size. For the predicted value of the participation coefficient of the i-th mode, This represents the actual value of the participation coefficient for the i-th mode. It is a very small positive number. To obtain the maximum absolute value between the predicted value and the actual value of the participation coefficient of the i-th mode.

[0140] In addition, the BPNN model can also be used to predict scalars, such as grain size and grain uniformity index.

[0141] The training steps for a BPNN model can be as follows:

[0142] Step 1: Initialize the weights. The weights are usually randomly assigned by setting a weight range.

[0143] Step 2: Calculate the output values ​​of the hidden layer and the output layer.

[0144] Step 3: Calculate the reverse error of each layer.

[0145] Step 4: Determine if the training requirements are met. If they are met, the model training is complete, and the model is output. If not, update the weights and repeat steps 2 and 3.

[0146] In summary, the modal participation coefficients in the POD model are updated based on the output of the BPNN, thereby calculating the field quantity prediction results.

[0147] The formula for predicting the field volume can be expressed as:

[0148]

[0149] in, These are predicted values ​​from the field volume data. Let M be the average value of the field quantity data, and M be the order of the modal space. Scalar coefficient predictions for each mode. These are spatial modal basis functions.

[0150] Optionally, an intrinsic orthogonal decomposition model is used to process the field quantity data in the interpolated uniform regular grid into low-order linear modal field quantity data. The modal participation coefficient solving model is used to calculate the modal parameter coefficients required for the intrinsic orthogonal decomposition model to predict the field quantity data. The intrinsic orthogonal decomposition model calls the modal parameter coefficients calculated by the modal participation coefficient solving model to calculate the field quantity prediction results, thus obtaining the field quantity prediction results.

[0151] Step 104: Input the set of process parameters to be optimized and the field quantity prediction results into the fast simulation proxy model to obtain the optimized process parameter set of the tubular part.

[0152] It should be noted that the rapid simulation surrogate model is a nonlinear mapping surrogate model between process parameters and objective function. This model is based on the Kriging model to construct the nonlinear relationship between process parameters and objective function, and constructs a freeform problem through the expected improvement (EI) point addition criterion.

[0153] The specific calculation process can be summarized as follows:

[0154] When the minimum value of the objective function in the current sample is known to be... For samples to be updated The estimated value of its fast simulation surrogate model is ,and The expected improvement of the objective function can be expressed by the following formula:

[0155] ;

[0156] in, This is an estimated value. The predicted mean of the objective function value. For variance, To minimize the objective function value, For samples to be updated For the samples to be updated in the Kriging model The prediction variance The cumulative distribution function of the standard normal distribution. Let be the probability density function of the standard normal distribution. These are estimated values ​​for process parameters.

[0157] The rapid simulation surrogate model employs cross-validation, and the accuracy evaluation will use the coefficient of determination. The formula for the coefficient of determination can be:

[0158] ;

[0159] in, As the coefficient of determination, Let i be the actual value of the i-th observation point. Let i be the predicted value for the i-th observation point. is the average of all actual values, i is the sample number, and p is the total number of samples.

[0160] It should also be noted that the optimization model and algorithm used in the rapid simulation surrogate model can also be used to solve the Pareto front using other intelligent optimization algorithms or discrete optimization algorithms such as NSGA-II (Non-dominated Sorting Genetic Algorithm II).

[0161] Optionally, the field quantity prediction results are input into the Kriging model to obtain the nonlinear mapping relationship between the process parameters to be optimized and the objective function in the set of process parameters to be optimized. Combined with the expected improvement point addition criterion, the objective function value of the process parameters to be optimized in the set of process parameters to be optimized is calculated. The objective function value is the optimized process parameter.

[0162] In the extrusion process parameter optimization method provided in this application, both linear order reduction and nonlinear mapping are considered when establishing a rapid simulation proxy model. The spatial degrees of freedom are reduced by linear mode order reduction, and the neglect of nonlinearity in the intrinsic orthogonal decomposition process is compensated by BPNN.

[0163] In summary, the extrusion process parameter optimization method provided in this application first employs a deep neural network to process the process parameters to be optimized, obtaining the predicted shape of the tubular part. The simulation results and shape prediction results of the tubular part are then input into an interpolation model to obtain an interpolated uniform grid. Next, the interpolated uniform grid is input into an intrinsic field prediction model to obtain field prediction results. Finally, the process parameters to be optimized and the field prediction results are input into a rapid simulation proxy model to obtain the optimized process parameter set for the tubular part. This application, through a rapid simulation proxy model, can quickly calculate the optimized process parameters for the tubular part, significantly improving the efficiency of optimizing the extrusion process.

[0164] It should be noted that this application uses an intrinsic field quantity prediction model to predict field quantity data. For scalar results, in addition to using the BPNN model mentioned above for prediction, time series-based models such as RNN (Recurrent Neural Network) or LSTM (Long Short-Term Memory) can also be used for prediction.

[0165] In addition, an SBO optimization framework can be established, under which the fast simulation surrogate model and the optimization algorithm model can be two independent modules, which can be extended and combined with various algorithms for different problems in the future.

[0166] The above describes a method for optimizing extrusion process parameters provided by embodiments of this application. The following describes the apparatus for performing the above-described method for optimizing extrusion process parameters.

[0167] Please see Figure 6 , Figure 6 This is a schematic diagram of an extrusion process parameter optimization device provided in this application. Figure 6 As shown, the device includes:

[0168] The system comprises: shape prediction unit 10, interpolation unit 20, field quantity prediction unit 30, and process parameter optimization unit 40; wherein:

[0169] The shape prediction unit 10 is used to input the set of process parameters to be optimized for the tubular part into the deep neural network model to obtain the shape prediction result of the tubular part; the set of process parameters to be optimized is: the set of various key sensitive process parameters obtained after sampling the simulation results of the tubular part extrusion process and performing parameter sensitivity analysis.

[0170] Interpolation unit 20 is used to input simulation results and shape prediction results into the interpolation model to obtain the interpolated uniform regular grid; the interpolation model is used to: interpolate the field quantity data of different unstructured grids with different process parameters to the uniform regular grid to be interpolated, to obtain the interpolated uniform regular grid; the different unstructured grids with different process parameters are obtained by discretizing the simulation results;

[0171] The field quantity prediction unit 30 is used to input the interpolated uniform regular grid into the intrinsic field quantity prediction model to obtain the field quantity prediction result.

[0172] The process parameter optimization unit 40 is used to input the set of process parameters to be optimized and the field quantity prediction results into the fast simulation proxy model to obtain the optimized process parameter set of the tubular part. The fast simulation proxy model is used to process the process parameters to be optimized in the set of process parameters to be optimized through the nonlinear mapping relationship between process parameters and objective function to obtain the optimized process parameter set.

[0173] In one embodiment, the shape prediction result in the interpolation unit 20 includes the punch end displacement result of the blank and the free end displacement result of the tubular part; the interpolation unit 20 is specifically used for:

[0174] The grid coordinates of the uniform regular grid to be interpolated are updated using the displacement results of the punch end and the displacement results of the free end.

[0175] The field quantity data of each grid of different unstructured grids with different process parameters are interpolated to each grid of the uniform regular grid to be interpolated.

[0176] In one embodiment, the intrinsic field quantity prediction model in the field quantity prediction unit 30 is a prediction model that integrates the intrinsic orthogonal decomposition model and the modal participation coefficient solution model. The field quantity prediction unit 30 is specifically used for:

[0177] The intrinsic orthogonal decomposition model is used to process the field quantity data in the interpolated uniform regular grid into low-order linear modal field quantity data;

[0178] The modal participation coefficients required for predicting field quantity data are calculated using the intrinsic orthogonal decomposition model by solving the model using modal participation coefficients.

[0179] The intrinsic orthogonal decomposition model calls the modal participation coefficient calculation model to calculate the modal participation coefficients, and then solves for the field quantity prediction results.

[0180] In one embodiment, the modal participation coefficient solution model in the field prediction unit 30 includes a backpropagation neural network model, which is used to update the modal participation coefficients in the intrinsic orthogonal decomposition model and also to predict the scalar data of the tubular component.

[0181] In one embodiment, the rapid simulation surrogate model in the process parameter optimization unit 40 is a nonlinear surrogate model between process parameters and the objective function constructed based on the Kriging model; the process parameter optimization unit 40 is specifically used for:

[0182] The field quantity prediction results are input into the Kriging model to obtain the nonlinear mapping relationship between the process parameters to be optimized and the objective function in the set of process parameters to be optimized.

[0183] The objective function values ​​of the process parameters to be optimized are calculated by combining the expected improvement criteria. The objective function values ​​are the optimized process parameters.

[0184] In one embodiment, the extrusion process parameter optimization device further includes a process parameter set determination unit;

[0185] The process parameter set determination unit is used to establish a 2D axisymmetric equivalent model of tubular parts using numerical simulation tools;

[0186] The process parameters input during the establishment of the 2D axisymmetric equivalent model are sampled to obtain a set of process parameters;

[0187] Parameter sensitivity analysis is performed on each process parameter in the process parameter set to obtain the set of process parameters to be optimized.

[0188] This application also provides an extrusion process parameter optimization device in its embodiments. (Reference) Figure 7 The diagram illustrates a structural schematic suitable for implementing the extrusion process parameter optimization device provided in this application. The extrusion process parameter optimization device in this embodiment may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 7 The extrusion process parameter optimization equipment shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0189] like Figure 7 As shown, the extrusion process parameter optimization device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the extrusion process parameter optimization device is powered on, the RAM 603 also stores various programs and data required for the operation of the extrusion process parameter optimization device. The processing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0190] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, memory cards, hard drives, etc.; and communication devices 609. Communication device 609 allows the extrusion process parameter optimization equipment to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 An extrusion process parameter optimization apparatus with various devices is shown; however, it should be understood that implementation or having all of the devices shown is not required. More or fewer devices may be implemented alternatively.

[0191] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the extrusion process parameter optimization methods provided in this application.

[0192] This application also provides a computer storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the extrusion process parameter optimization methods provided in this application.

[0193] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0194] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0195] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0196] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for optimizing extrusion process parameters, characterized in that, include: The set of process parameters to be optimized for the tubular component is input into a deep neural network model to obtain the shape prediction result of the tubular component. The set of process parameters to be optimized is: a set of key sensitive process parameters obtained after sampling and parameter sensitivity analysis of the simulation results of the tubular extrusion process; the shape prediction results include the displacement results of the punch end of the billet and the displacement results of the free end of the tubular part; the simulation results are determined by the 2D axisymmetric equivalent model of the tubular part. The simulation results and the shape prediction results are input into the interpolation model to obtain the interpolated uniform regular grid. The interpolation model is used to: update the grid coordinates of the uniform regular grid to be interpolated using the displacement results of the punch end and the displacement results of the free end; and interpolate the field data of different unstructured grids with different process parameters to the uniform regular grid to be interpolated to obtain the interpolated uniform regular grid. The different unstructured grids with different process parameters are obtained by discretizing the simulation results. The interpolated uniform grid is input into the intrinsic field quantity prediction model to obtain the field quantity prediction results; The set of process parameters to be optimized and the field quantity prediction results are input into the fast simulation proxy model to obtain the optimized process parameter set of the tubular component. The rapid simulation proxy model is a nonlinear proxy model between the process parameters and the objective function constructed based on the Kriging model; the field quantity prediction results are input into the Kriging model to obtain the nonlinear mapping relationship between the process parameters to be optimized and the objective function in the set of process parameters to be optimized; The objective function value of the process parameters to be optimized in the set of process parameters to be optimized is calculated by combining the expected improvement point criteria. The objective function value is the optimized process parameter.

2. The method for optimizing extrusion process parameters according to claim 1, characterized in that, The intrinsic field quantity prediction model is a prediction model that integrates the intrinsic orthogonal decomposition model and the modal participation coefficient solution model. The step of inputting the interpolated uniform grid into the intrinsic field quantity prediction model to obtain the field quantity prediction results includes: The intrinsic orthogonal decomposition model is used to process the field quantity data in the interpolated uniform regular grid into low-order linear modal field quantity data; The modal participation coefficients required for predicting field quantity data are calculated using the modal participation coefficient solution model based on the intrinsic orthogonal decomposition model. The intrinsic orthogonal decomposition model calls the modal participation coefficients calculated by the modal participation coefficient solving model to solve for the field quantity prediction results.

3. The method for optimizing extrusion process parameters according to claim 2, characterized in that, The modal participation coefficient solution model includes a backpropagation neural network model; The backpropagation neural network model is used to update the modal participation coefficients in the intrinsic orthogonal decomposition model and also to predict the scalar data of the tubular component.

4. The method for optimizing extrusion process parameters according to claim 1, characterized in that, Before inputting the set of process parameters to be optimized for the tubular component into the deep neural network model, the following steps are also included: A 2D axisymmetric equivalent model of the tubular component was established using numerical simulation tools; The process parameters input during the establishment of the 2D axisymmetric equivalent model are sampled to obtain a set of process parameters; Parameter sensitivity analysis is performed on each process parameter in the set of process parameters to obtain the set of process parameters to be optimized.

5. An extrusion process parameter optimization device, characterized in that, include: The shape prediction unit is used to input the set of process parameters to be optimized for the tubular part into the deep neural network model to obtain the shape prediction result of the tubular part. The set of process parameters to be optimized is: a set of key sensitive process parameters obtained after sampling and parameter sensitivity analysis of the simulation results of the tubular extrusion process; the shape prediction results include the displacement results of the punch end of the billet and the displacement results of the free end of the tubular part; the simulation results are determined by the 2D axisymmetric equivalent model of the tubular part. An interpolation unit is used to input the simulation results and the shape prediction results into an interpolation model to obtain an interpolated uniform regular grid. The interpolation model is used to: update the grid coordinates of the uniform regular grid to be interpolated using the displacement results of the punch end and the displacement results of the free end; and interpolate the field data of different unstructured grids with different process parameters to the uniform regular grid to be interpolated to obtain the interpolated uniform regular grid. The different unstructured grids with different process parameters are obtained by discretizing the simulation results. The field quantity prediction unit is used to input the interpolated uniform regular grid into the intrinsic field quantity prediction model to obtain the field quantity prediction result. The process parameter optimization unit is used to input the set of process parameters to be optimized and the field quantity prediction results into the fast simulation proxy model to obtain the optimized process parameter set of the tubular component. The rapid simulation proxy model is used to process the process parameters to be optimized in the set of process parameters to be optimized through the nonlinear mapping relationship between process parameters and objective function, so as to obtain the optimized process parameter set; wherein, the rapid simulation proxy model is a nonlinear proxy model between the process parameters and objective function constructed based on the Kriging model; the field quantity prediction results are input into the Kriging model to obtain the nonlinear mapping relationship between the process parameters to be optimized and objective function in the set of process parameters to be optimized; The objective function value of the process parameters to be optimized in the set of process parameters to be optimized is calculated by combining the expected improvement point criteria. The objective function value is the optimized process parameter.

6. An extrusion process parameter optimization device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the extrusion process parameter optimization device can implement the extrusion process parameter optimization method as described in any one of claims 1 to 4.

7. A computer program product, characterized in that, Includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the extrusion process parameter optimization method as described in any one of claims 1 to 4.

8. A computer storage medium, characterized in that, The storage medium carries one or more computer programs that, when executed by an electronic device, enable the electronic device to implement the extrusion process parameter optimization method as described in any one of claims 1 to 4.

Images