A Forming Process Database-Driven Adaptive Optimization Method for Loading Parameters

Abstract

This application provides a forming process database-driven adaptive optimization method for loading parameters, comprising: Step 1, constructing a forming process database for metallic materials; Step 2, searching the forming process database to obtain a loading scheme for the metallic material from an initial macro-micro state to a target macro-micro state; Step 3, continuously loading and forming the metallic material based on the loading scheme and obtaining real-time loading parameter combinations, and estimating the real-time macro-micro state of the metallic material; Step 4, determining whether the deviation between the expected macro-micro state and the real-time macro-micro state exceeds a set threshold. If the determination result is yes, proceed to Step 5; otherwise, return to Step 3; Step 5, re-searching to obtain a loading scheme for the metallic material from the macro-micro state at the moment the deviation exceeds the set threshold to the target macro-micro state, and then returning to Step 3. The method of this application can optimize the forming process under the high-density loading parameter-forming evolution data driving of metallic materials.

Description

Technical Field

[0001] This application belongs to the field of intelligent control technology for metal material forming, and specifically provides a method for adaptive optimization of loading parameters driven by a forming process database. Background Technology

[0002] The determination and optimization of loading process parameters in traditional metal forming manufacturing relies on repeated trial and error: forming test, result measurement, optimization of loading process parameters, and re-forming test. On the one hand, the complex coupling effect of metal forming loading process parameters on the results leads to long trial and error cycles, cumbersome cyclical processes, and high material consumption, significantly increasing time and economic costs. On the other hand, current technologies cannot achieve flexible adjustment during the forming process; once forming begins, processing can only be carried out according to predetermined loading parameters, resulting in the inability to correct dimensional deviations and microstructure variations, making it difficult to improve the scrap rate.

[0003] In recent years, solutions have emerged for real-time measurement of material state during the metal forming and manufacturing process, and dynamic adjustment of loading parameters based on the measurement results. For example, patent CN118824415A discloses a data-driven multi-field, multi-scale dynamic simulation method and process for the entire forging process. This method obtains a material constitutive model through experiments, establishes macro- and micro-scale finite element models of the entire forging process, verifies or corrects the models through relevant experiments and tools, conducts multi-field, multi-scale simulations, establishes a process database, trains a convolutional neural network, monitors and extracts temperature-time data of the region of interest in the forging using an infrared thermometer, converts this data into a time-frequency domain scale map, and inputs it into the trained neural network. Convolutional neural networks (CNNs) are used to predict the real-time mechanical properties of forgings, dynamically adjust process parameters throughout the forging process, and improve the mechanical properties of forgings at the final forging stage. For example, patent CN116663346A provides a method for controlling the temperature field during the forging process of large forgings. This method first establishes a three-dimensional manifold equidistant mapping model of the forging temperature field and a time-series prediction model of the mapped two-dimensional data by performing finite element simulation of the forging process and acquiring prior data of the forging temperature field. Then, during the actual forging process, the surface temperature of the forging is measured in real time, the posterior data and posterior probability of the forging temperature field at the current moment are calculated, the temperature field of the forging at the next moment is predicted, and finally, the forging process parameters are adjusted accordingly.

[0004] Although the above method predicts material properties through a pre-established simulation model and adaptively adjusts the loading parameters based on the prediction results, it directly uses the simulation results or model prediction results as a reliable basis for adjustment. This approach ignores the complex coupling between macro and micro parameters in the metal forming and manufacturing process, which often leads to a large difference between the simulation results and the actual macro and micro states. Using this as the basis for adjusting the loading parameters can easily lead to the inability to achieve the optimization goal when the accuracy of the simulation model is not high. Summary of the Invention

[0005] This application provides a database-driven adaptive optimization method for loading parameters based on a forming process, comprising the following steps:

[0006] Step 1: Construct a forming process database for metallic materials. The forming process database stores structured high-density loading parameter combinations and forming process data obtained through simulation.

[0007] Step 2: Search the forming process database to obtain the loading scheme of the metal material from the initial macro-micro state to the target macro-micro state. The loading scheme includes the expected combination of loading parameters applied to the metal material at each time frame and the expected macro-micro state of the metal material.

[0008] Step 3: Continuously load and shape the metal material based on the loading scheme and obtain real-time loading parameter combinations, and continuously estimate the real-time macro and micro states of the metal material based on the real-time loading parameter combinations and the data in the forming process database;

[0009] Step 4: Determine whether the deviation between the expected macro / micro state and the real-time macro / micro state exceeds a set threshold. If the determination result is yes, proceed to step 5; otherwise, return to step 3.

[0010] Step 5: Re-search to obtain the loading scheme for the metal material from the macro-micro state when the deviation exceeds the set threshold to the target macro-micro state, and then return to step 3.

[0011] Preferably, the high-density loading parameter combination covers the corresponding loading parameter window in each dimension of the loading parameter space in a manner not greater than the preset upper limit of the loading parameter value interval;

[0012] For any set of high-density loading parameters, the corresponding forming process data includes the simulation correction results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metal material at each time frame under the loading of the set of high-density loading parameters.

[0013] Furthermore, the forming process database is constructed through the following steps:

[0014] A1. Determine several sets of basic loading parameter combinations from the loading parameter window of the metallic material;

[0015] A2, Basic forming simulation and basic forming test of metallic materials based on basic loading parameter combinations;

[0016] A3 generates basic residual samples based on the results of basic forming simulation and basic forming test.

[0017] A4, Use the basic residual samples to perform basic training on the residual estimation model;

[0018] A5. The residual level of the basic forming simulation results is evaluated using the residual estimation model after basic training.

[0019] A6. Based on the evaluation results, select several sets of incremental loading parameter combinations from the loading parameter window;

[0020] A7, Incremental forming simulation and incremental forming test of metallic materials based on incremental loading parameter combination;

[0021] A8 generates incremental residual samples based on incremental forming simulation and incremental forming test results;

[0022] A9, use incremental residual samples to incrementally train the residual estimation model;

[0023] A10, Generate high-density load parameter combinations from the load parameter window, wherein the number of high-density load parameter combinations is much greater than the number of basic load parameter combinations;

[0024] A11, based on the high-density loading parameter combination, the high-density forming simulation of metallic materials is carried out, and the simulation results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metallic material under the loading of each set of high-density loading parameter combination are obtained in each time frame.

[0025] A12 uses an incrementally trained residual estimation model to correct the high-density forming simulation results, thereby obtaining forming process data corresponding to each combination of high-density loading parameters.

[0026] Furthermore, the residual estimation model is a three-level Gaussian regression model that integrates macro-micro coupling mechanisms, including:

[0027] At the macroscopic level, based on the input combination of loading parameters, the output simulation results show the preliminary residual estimates of the macroscopic flow parameters at each data point and their uncertainties.

[0028] The micro-layer, based on the combination of input loading parameters and the output results of the macro-layer, outputs the residual estimates and uncertainties of the micro-organism parameters of each data point in the simulation results.

[0029] The macro-micro coupling layer, based on the outputs of the macro and micro layers, outputs the residual estimates and uncertainties of the macroscopic flow parameters at each data point in the simulation results.

[0030] Preferably, the incremental loading parameter combination is determined based on the evaluation function shown below:

[0031] ,

[0032] in, The sequence number of the basic loading parameter combination. The total number of combinations of basic loading parameters. For the first Group basic loading parameter combination, for In the corresponding forming process data, the mean of the normalized estimated values ​​of the macroscopic flow parameter residuals for each data point at each time frame is... for standard deviation for The rate of change of the spatial coordinates of the loading parameters, , These are the weighting coefficients.

[0033] Preferably, the loading scheme for the metallic material from its initial macro-micro state to its final macro-micro state is determined through the following steps:

[0034] B1, to obtain the initial macro- and micro-states and target macro- and micro-states of metallic materials;

[0035] B2, Construct a global evolution quality function, which is used to characterize the evolution quality of the macro-micro state of a metallic material during the forming process from the initial macro-micro state to the final macro-micro state, with the target macro-micro state as the forming target.

[0036] B3 searches the process database to obtain the time series of loading parameter combinations with the optimal global evolution quality function value and generates a loading scheme for the metallic material from the initial macro-micro state to the final macro-micro state.

[0037] Preferably, the evolutionary quality function includes at least the following terms: target macro-micro state deviation term, evolutionary process uniformity evaluation term, macro-state maximum change rate term, and micro-state maximum transformation rate term.

[0038] Preferably, the loading scheme for the metallic material from the initial macro-micro state to the final macro-micro state includes two or more different combinations of high-density loading parameters.

[0039] Preferably, the following steps are used to search and determine the sequence of loading schemes for the metallic material from the macro-micro state at the moment when the deviation exceeds a set threshold to the target macro-micro state:

[0040] C1, to obtain the macro and micro states of the metallic material when the deviation exceeds a set threshold;

[0041] C2, Construct the residual stage evolution quality function, which is used to characterize the evolution quality of the macro-micro state of the metallic material during the forming process from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state, with the target macro-micro state as the forming target.

[0042] C3 searches the process database to obtain the loading parameter combination time series with the optimal remaining stage evolution quality function value and generates a loading scheme for the metallic material from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state.

[0043] Preferably, the remaining stage evolution quality function includes at least the following terms: target macro-micro state deviation term, evolution process uniformity evaluation term, macro-state maximum change rate term, micro-state maximum transformation rate term, and progress evaluation term.

[0044] The forming process database-driven adaptive optimization method for loading parameters provided in this application first searches for the optimal loading scheme under the guidance of high-density stored data on the macro- and micro-state evolution of all data points of the metal material, while simultaneously acquiring the expected macro- and micro-state evolution under the optimal loading scheme throughout the entire process. Then, during the actual loading process, based on the measured results of the macro-loading parameters, the actual macro- and micro-state evolution of all data points of the metal material is estimated again under the guidance of the high-density stored data on the macro- and micro-state evolution of all data points of the metal material and compared with the expected macro- and micro-state evolution. Finally, based on the comparison results, the loading scheme is adjusted a third time under the guidance of the high-density stored data on the macro- and micro-state evolution of all data points of the metal material. In this way, the macro- and micro-states of all data points can be monitored and adjusted throughout the entire forming process of the metal material, and the adjustment direction and effect are controllable, thereby achieving controllable loading for metal forming. Attached Figure Description

[0045] Figure 1 This is a flowchart of a loading parameter adaptive optimization method driven by a forming process database according to an embodiment of this application;

[0046] Figure 2 This is a schematic diagram of a single record stored in the forming process database provided according to an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the architecture of high-density loading combination-forming process data stored in the forming process database provided according to an embodiment of this application;

[0048] Figure 4 This is a flowchart illustrating the construction of a forming process database for metallic materials according to embodiments of this application;

[0049] Figure 5 This is a schematic diagram of a three-dimensional finite element model of a metal billet according to a specific embodiment;

[0050] Figure 6 This is a schematic diagram of a three-dimensional finite element model of a metal billet and a mold according to a specific embodiment;

[0051] Figure 7 This is a schematic diagram of a three-dimensional finite element model of a metal workpiece according to a specific embodiment;

[0052] Figure 8 A physical image of a real forming test piece provided according to a specific embodiment;

[0053] Figure 9 This is a schematic diagram of the architecture of the residual estimation model provided according to the embodiments of this application;

[0054] Figure 10 This is a schematic diagram illustrating the principle of determining the loading scheme according to the embodiments of this application;

[0055] Figure 11 This is a schematic diagram illustrating the principle of the update loading scheme provided according to the embodiments of this application. Detailed Implementation

[0056] The present application will now be further described based on preferred embodiments and with reference to the accompanying drawings.

[0057] Figure 1 The flowchart below shows a database-driven adaptive optimization method for loading parameters in a forming process, according to some embodiments of this application. (Refer to...) Figure 1 The method includes the following steps:

[0058] Step 1: Construct a forming process database for metallic materials. The forming process database stores structured high-density loading parameter combinations and forming process data obtained through simulation.

[0059] Step 2: Search the forming process database to obtain the loading scheme of the metal material from the initial macro-micro state to the target macro-micro state. The loading scheme includes the expected combination of loading parameters applied to the metal material at each time frame and the expected macro-micro state of the metal material.

[0060] Step 3: Continuously load and shape the metal material based on the loading scheme and obtain real-time loading parameter combinations, and continuously estimate the real-time macro and micro states of the metal material based on the real-time loading parameter combinations and the data in the forming process database;

[0061] Step 4: Determine whether the deviation between the expected macro / micro state and the real-time macro / micro state exceeds a set threshold. If the determination result is yes, proceed to step 5; otherwise, return to step 3.

[0062] Step 5: Re-search to obtain the loading scheme for the metal material from the macro-micro state when the deviation exceeds the set threshold to the target macro-micro state, and then return to step 3.

[0063] The specific implementation methods of each of the above steps are explained in detail below with reference to the accompanying drawings.

[0064] <Constructing a database of forming processes for metallic materials>

[0065] In the embodiments of this application, step one constructs a forming process database for metal materials through high-density, high-precision simulation. This database stores structured high-density loading parameter combinations and forming process data. In subsequent steps, this data will be used to search for the optimal loading scheme for forming metal materials and to estimate the macro- and micro-states of the metal materials in real time during the actual loading process. Therefore, the technical solution provided by this application drives the entire forming process of metal materials based on the high-density loading parameter combinations and forming process data in the process database. By utilizing the determinism of the loading parameter combinations and the corresponding macro- and micro-state evolution processes, the entire forming process of metal materials is controlled and monitored. This effectively solves the problem that existing prediction models, when predicting the global state evolution of metal materials based on limited point measured data, cannot distinguish between the mutually coupled loading parameters and macro- and micro-states under the support of a clear physical picture, resulting in low reliability of the prediction results.

[0066] Specifically, in the metal material forming process database established through simulation in step one of this application, each set of high-density loading parameter combinations covers the corresponding loading parameter window in each dimension of the loading parameter space in a manner not greater than the preset upper limit of the loading parameter value interval; at the same time, any set of high-density loading parameter combinations corresponds to a set of forming process data, and the corresponding forming process data includes the simulation correction results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metal material in each time frame under the loading of the set of high-density loading parameter combinations. For the forming process data, it includes the simulation correction results of the macroscopic and microscopic parameters of each data point (the mapping of each real space point of the metal material) of the three-dimensional finite element model of the metal material in each time frame under the loading of each set of loading parameter combinations.

[0067] The following combination Figure 2 and Figure 3 The data structure of the metal forming process database is explained.

[0068] Figure 2 This illustration shows a schematic diagram of a record stored in a forming process database in one specific embodiment, such as... Figure 2 As shown, the high-density loading parameter combination-forming process data is sorted according to the loading parameter combination sequence number, data point sequence number, and time frame sequence number. That is, each set of high-density loading parameters (such as...) is used in sequence... Figure 2 The loading temperature of the billet shown The temperature is 1000℃, and the mold loading rate is 6s.-1 When loading a metal material with a mold corner radius of 8mm, each data point of the metal material in each time frame... Each of them has a set of data describing its macro- and micro-states. These macro- and micro-data include several macro-parameters, including the spatial coordinates of the data point and macro-flow parameters (including stress, strain, temperature, etc.), as well as several micro-structure parameters (including dislocation density, grain size, etc.).

[0069] Loading parameters (also known as process parameters) are control variables applied to metal materials during the forming process that cause changes in their macroscopic and microscopic states. These parameters determine the stress, heating conditions, deformation history, etc. of the metal material during the forming process, and directly affect its physical and mechanical behavior and final properties.

[0070] The number and types of loading parameters vary depending on the metal forming process, for example, Figure 2 The embodiments shown correspond to forming process data for a titanium alloy rib. Each set of high-density loading parameters includes parameters such as the loading temperature of the metal billet, the die loading rate, and the die fillet radius (i.e., the fillet radius at the edge of the formed rib). These parameters can be represented as vectors. ,in, Indicates the number is High-density loading parameter combination, , , These represent the specific loading temperature, mold loading rate, and mold corner radius for this set of loading parameters. This represents the total number of high-density loading parameter combinations stored in the database. It should be understood that the above content is for illustrative purposes only and not to limit the expression of loading parameter combinations; different metal forming processes may vary. The number and format of the included loading parameters can be adaptively adjusted according to specific needs.

[0071] As mentioned above, in order to achieve high-precision loading scheme selection and loading process monitoring, the loading parameter group stored in the process database should be ensured to reach a certain density. Therefore, in some preferred embodiments, the high-density loading parameter combination covers the corresponding loading parameter window in each dimension of the loading parameter space in a manner that is no greater than the preset upper limit of the loading parameter value interval.

[0072] For example, in Figure 2In the illustrated embodiment, the loading temperature of the metal billet, the mold loading rate, and the mold corner radius constitute a three-dimensional loading parameter space. In each dimension, a loading parameter window constrains the values ​​of the loading parameters for that dimension. For example, the loading parameter window for the billet loading temperature is 700°C - 950°C; the loading parameter window for the loading rate is 0.001s. -1 - 10s -1 That is, the loading cycle is between 0.1s and 1000s, and the loading parameter window for the mold fillet radius is 6mm - 8mm; under the above three loading parameter dimensions, the upper limit of the value interval of each loading parameter can be set respectively.

[0073] Taking loading temperature as an example, if the upper limit of its value interval is set to 5°C, that is, the difference between two adjacent loading temperatures does not exceed 5°C, then in some optional embodiments, 5°C can be used as the value interval, and 700°C, 705°C, 710°C, …, 890°C, 895°C, 900°C can be taken sequentially as loading temperatures within the 700°C - 950°C temperature loading parameter window, thereby completing the coverage of the temperature loading parameter window; in other optional embodiments, the value interval can be further reduced, for example, using 1°C as the value interval of the temperature loading parameter, and 700°C, 701°C, 702°C, …, 898°C, 899°C, 900°C can be taken sequentially as loading temperatures within the 700°C - 950°C temperature loading parameter window, thereby performing more intensive value taking within the temperature loading parameter window.

[0074] Taking the corner radius of a mold as an example, if the upper limit of its value interval is set to 0.1mm, then the corner radius of the mold can be selected within the range of 6mm - 8mm with a value interval of no more than 0.1mm.

[0075] It should be noted that, in the embodiments of this application, the specific meaning of "not greater than the upper limit of the loading parameter value interval" should be determined according to the characteristics of the loading parameter. This is because some loading parameters can be expressed in different ways. For example, both loading rate and loading period can reflect the speed at which the material is loaded, but they are reciprocals of each other. Therefore, for the loading rate parameter, if its upper limit of the loading parameter value interval is set to 10s... -1 When the interval is not greater than the upper limit, it actually means that the interval between two adjacent loading frequencies should be equal to or greater than 10 seconds. -1In other words, when using the loading cycle instead of the loading frequency as the loading parameter, the corresponding loading cycle value interval is no greater than 0.1s. That is, the loading cycle value can be selected within a loading cycle window of 0.1s to 1000s at intervals no greater than 0.1s. For example, the loading cycle values ​​can be 0.1s, 0.2s, 0.3s, ..., 999.8s, 999.9s, 1000s, or more densely selected values ​​can be 0.10s, 0.15s, 0.20s, ..., 999.90s, 999.95s, 1000s.

[0076] After selecting the high-density loading parameters for each dimension using the methods described above, orthogonalizing these parameters yields the final high-density loading parameter combination. Further, refer to... Figure 2 For each set of high-density loading parameter combinations Through the simulation process described below, a three-dimensional finite element model of the metallic material can be obtained. During the loading process, the evolution of the macro- and micro-states of all data points (each data point corresponds to a real material point in the metal) can be observed. Clearly, the macro- and micro-states of each data point can also be represented using vectors. ,in, , These represent the data point and time frame numbers, respectively. Represents a group Data points in the metal billet under loading In time frame The spatial location (which can be obtained, for example, through strain accumulation). This indicates the corresponding macroscopic flow parameters (such as stress parameters, strain parameters, and temperature). This represents the corresponding microstructure parameters (such as dislocation density, grain size, etc.), and correspondingly, in Under loading, the three-dimensional finite element model of the metallic material in time frame The combination of macro and micro states of all data points can be represented as .

[0077] Clearly, through structured construction, each group Each of these will correspond to a complete forming process data set for the forming of metal materials. Figure 3 This schematically illustrates the complex evolution of the metallic material during the forming process, as reflected by the various sets of high-density loading parameter combinations-forming process data stored after the data construction is completed.

[0078] Specifically, the diagram includes four sets of high-density loading parameter combinations, respectively... to This indicates that the macro- and micro-states of all data points for the metallic material are represented by two data points respectively. , Representatively, time frames are respectively represented as , The numbering indicates that for each combination of high-density loading parameters, the interval between adjacent time frames is consistent; therefore, the figures with the same numerical number are... This indicates the same moment after loading begins.

[0079] observe Figure 3 It can be seen that at the beginning of each set of loading parameters, the macro- and micro-states of the metallic material are consistent, therefore hour , The macro and micro states may not include the numbers of the loading parameter combinations, i.e., they are represented as follows: and As each set of loading parameters is applied to the metal material, the macro and micro states of each data point begin to evolve along different evolution paths, eventually reaching the desired workpiece shape. Therefore, the forming process data corresponding to each set of high-density loading parameter combinations can be regarded as a "path" (the four paths of different colors in the figure: red, blue, pink, and black). On each path, under the action of the corresponding loading parameters, the macro and micro states of each data point continuously evolve over time, eventually reaching the final shape and macro and micro states of the workpiece.

[0080] Clearly, the macro and micro states of each data point exhibit complex trends of aggregation, dispersion, and intersection during the evolution of each path, and it is possible for the same or similar states to appear in different time frames of different paths. For example, in Under loading, the metal billet is The frame reached the same level as Under the loading The same or similar frames are displayed in the following way: During loading, the metal billet undergoes rapid changes in its macro- and micro-states.

[0081] pass Figure 3 It can also be observed that, due to the use of different combinations of loading parameters, the time required for each forming path is also different. For example, Figure 3 middle The workpiece was formed in just 340 time frames, and This process took 2000 time frames. A faster forming speed reflects a faster change in the macro- and micro-states of each data point (as shown by the red and blue paths in the figure), and may lead to significant fluctuations in the evolution of microstructure parameters during the forming process. Conversely, a longer forming time, as shown by the pink and black forming paths in the figure, reflects a greater likelihood that the macro- and micro-states of each data point are undergoing a more uniform evolution. It is precisely because different combinations of loading parameters lead to different forming paths that, as can be seen from the figure, although the final workpiece's appearance and size generally tend to be consistent (i.e., some macro-states are the same or similar), some states still show significant differences at the end of forming. Therefore, under the loading of four sets of high-density loading parameters, the various data points of the metal billet (including...) , Starting from the same initial macro- and micro-states, the final states reached are respectively represented as:

[0082] ;

[0083] ;

[0084] ;

[0085] .

[0086] The above describes the data structure of the forming process database for metallic materials and the complex interactive relationships of the forming paths of metallic materials under different combinations of loading parameters. Obviously, how to construct such a forming process database with high density and high precision will directly affect the formulation of subsequent loading schemes and the real-time adjustment of loading methods during the forming process.

[0087] Clearly, obtaining such high-density, time-varying data on the macro- and micro-states of various spatial points of a metallic material during the forming process under different combinations of loading parameters can only be achieved through simulation. To improve the accuracy of the simulation results, it is generally necessary to combine the simulation model with a limited number of real experiments for correction. Currently, common simulation model correction methods generally include the following steps:

[0088] (1) First, determine the selectable windows for each dimension of the loading parameter in the high-dimensional loading parameter space, and then select several sets of loading parameter combinations in the loading parameter window.

[0089] Obviously, the combination of loading parameters used for experimental correction should be representative and cover the loading parameter window as much as possible. However, in actual selection, the range of some loading parameters is extremely large, such as... Figure 2In the illustrated embodiment, the possible values ​​of strain rate span four orders of magnitude. As the process window span increases and the dimensions of process parameters increase, if the density of parameter selection is increased in each dimension, the number of loading parameter combinations (i.e., the number of tests for metal material processing) will inevitably explode. Therefore, it is generally necessary to select a set of loading parameter combinations that can cover the multidimensional process window and have an appropriate number of combinations based on existing experience or by using methods such as Latin hypercube design, orthogonal design, or random sampling.

[0090] (2) After determining the combination of loading parameters used to correct the simulation model, the selected combination of loading parameters is used to perform multiple simulations of the metal material processing process and actual metal material processing tests, and the deviation between the actual test results and the simulation results is calculated to obtain the residual sample of the simulation results.

[0091] The data structure of the residual sample consists of the difference between the measured and simulated values ​​of various macroscopic flow parameters and microstructure parameters at various sampling locations under different combinations of loading parameters. Therefore, for any combination of loading parameters, the amount of residual sample data is determined by the number of sampling locations on the workpiece obtained from actual experiments. In particular, if it is necessary to measure the macroscopic and microscopic characteristics at locations inside the workpiece, it may be necessary to perform destructive measurements such as cutting the workpiece. Therefore, the number of sampling locations is generally around tens to hundreds.

[0092] (3) Use residual samples to correct the simulation model using interpolation or fitting methods.

[0093] Because residual samples can only reflect the error between simulation results and actual experimental results at a few sampling and measurement locations on the metal material, and the number of data points in the simulation model is much larger than the number of sampling and measurement points on the formed metal material. For example, the number of nodes in the three-dimensional finite element model of some large metal materials may reach 10. 4 The magnitude is even higher. Therefore, in addition to the sampled measurement location, there are a large number of data points whose simulation results cannot be directly compared with the sampled measurement results. In the existing simulation model correction schemes, for the above data points, the residuals of other data points are generally approximated by difference or polynomial fitting to finally obtain the residuals of all data points for subsequent simulation model correction work.

[0094] However, the above-mentioned correction method for the simulation model is difficult to accurately estimate the residuals of the simulation results of metal material forming with complex macro-micro coupling characteristics. This is because, as analyzed above, during the forming process of metal material, the stress and strain experienced by different parts of the material change from the original blank form to the final workpiece form. The time evolution of its macro-micro properties exhibits a significant nonlinear relationship. Therefore, estimating the residuals of the simulation results for massive data points requires the ability to handle the complex nonlinear relationship between macro-micro parameter residuals and to eliminate the heterogeneity of macro-micro parameters. The above-mentioned scheme only uses the residuals of a limited number of measurement positions and directly uses the results of linear interpolation or fitting as the simulation result residuals of other unmeasured positions, which obviously does not conform to the real evolution of metal forming.

[0095] Therefore, in the embodiments of this application, during the construction of the forming process database in step one, a residual estimation model with macro-micro coupling characteristic extraction capability is proposed. Through two rounds of simulation-experiment process, the preliminary training of the residual estimation model, the optimization selection of representative loading parameter combinations, and the incremental training of the residual estimation model are carried out respectively, thereby effectively improving the accuracy of simulation results. The metal material process database constructed in this way can provide reliable data support for the subsequent selection of loading schemes.

[0096] For details, please refer to Figure 4 In the embodiments of this application, a forming process database for metallic materials is constructed through the following steps:

[0097] A1. Determine several sets of basic loading parameter combinations from the loading parameter window of the metallic material;

[0098] A2, Basic forming simulation and basic forming test of metallic materials based on basic loading parameter combinations;

[0099] A3 generates basic residual samples based on the results of basic forming simulation and basic forming test.

[0100] A4, Use the basic residual samples to perform basic training on the residual estimation model;

[0101] A5. The residual level of the basic forming simulation results is evaluated using the residual estimation model after basic training.

[0102] A6. Based on the evaluation results, select several sets of incremental loading parameter combinations from the loading parameter window;

[0103] A7, Incremental forming simulation and incremental forming test of metallic materials based on incremental loading parameter combination;

[0104] A8 generates incremental residual samples based on incremental forming simulation and incremental forming test results;

[0105] A9, use incremental residual samples to incrementally train the residual estimation model;

[0106] A10, Generate high-density load parameter combinations from the load parameter window, wherein the number of high-density load parameter combinations is much greater than the number of basic load parameter combinations;

[0107] A11, based on the high-density loading parameter combination, the high-density forming simulation of metallic materials is carried out, and the simulation results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metallic material under the loading of each set of high-density loading parameter combination are obtained in each time frame.

[0108] A12 uses an incrementally trained residual estimation model to correct the high-density forming simulation results, thereby obtaining forming process data corresponding to each combination of high-density loading parameters.

[0109] The above steps consist of four stages: the initial training stage of the residual estimation model consisting of steps A1-A4; the incremental loading parameter combination selection stage consisting of steps A5-A6; the incremental training stage of the residual estimation model consisting of steps A7-A9; and the encrypted simulation and correction stage consisting of steps A10 to A12.

[0110] In step A1, appropriate loading parameters (such as loading temperature, loading rate, mold shape parameters, etc.) can be selected and their loading parameter window determined based on the type of metal material, the specific loading method for processing the metal material, and the forming target. In this application, the metal materials include, but are not limited to, steel, copper alloys, aluminum alloys, magnesium alloys, and titanium alloys. The loading methods for metal forming processing include, but are not limited to, forging, extrusion, casting, and drawing. In some specific embodiments, based on existing experience, or by using methods such as Latin hypercube design, orthogonal design, or random sampling, a combination of basic loading parameters that can cover the multidimensional loading parameter window and has an appropriate number can be selected. This is for use in subsequent basic forming simulation and basic forming experiments.

[0111] In step A2, the forming process of the metal material is simulated for each selected combination of basic loading parameters. Specifically, the simulation of the metal material forming process involves using a simulation model to perform time-iterative calculations based on finite element simulation to simulate the macro and micro states of each data point under the drive of a given combination of loading parameters during the metal material forming process. Including location, macroscopic flow parameters (such as equivalent stress) Equivalent change Temperature) and microstructure parameters (such as grain size) dislocation density , Phase volume fraction, The evolution of phase size, grain boundary area density, large-angle grain boundary volume fraction, phase distribution concentration, etc. over time.

[0112] In this application, the simulation model for metal material forming consists of a three-dimensional finite element model of the metal material being processed, a three-dimensional finite element model of the mold used to process the metal material, the constitutive equation of the metal material, and constraint conditions.

[0113] Among them, the three-dimensional finite element model of metallic materials can be established based on the actual geometry and dimensions of the metal billet. Figure 5 A three-dimensional finite element model of a specific cylindrical metal material (bulk) is shown in the figure. The number of data points (nodes) in the three-dimensional finite element model is generally determined by the size of the metal material and the simulation accuracy. Based on factors such as the target shape, the mesh can be refined for the key areas of interest.

[0114] Figure 6 A three-dimensional finite element model of a specific mold is shown, which includes an upper mold and a lower mold for hot pressing cylindrical metal blanks. Figure 6 As shown, the three-dimensional finite element model of the mold can be established based on its actual contour. Typically, the three-dimensional finite element model of the mold is set as an ideal rigid body, that is, it does not undergo strain when applied to the metal material (i.e., when a force is applied).

[0115] Constitutive equations are used to characterize the evolution of the macro- and micro-states of metallic materials over time under a given combination of loading parameters. Generally, constitutive equations contain multiple differential equations that reflect the time-varying characteristics of macro-flow parameters and micro-structure parameters, and the various macro-flow parameters and micro-structure parameters are coupled with each other, thus reflecting the complex macro-micro coupling characteristics in the metal processing and forming process.

[0116] Constraints are used to ensure that the simulation of the deformation process of metallic materials conforms to the physical reality and engineering significance. For example, the temperature of each data point of the blank model is constrained based on the given initial temperature; the displacement limit and loading path of the material can be determined based on the geometric parameters of the mold and the process parameters such as the loading rate of the metallic material; in addition, constraints such as thermo-mechanical boundary conditions and contact friction conditions can be determined based on the material properties of the metallic material and the mold.

[0117] After completing the above simulation model settings, each group can be... The data is loaded onto the billet and mold models, and iterative calculations are performed across time frames based on the constitutive equations. After each time frame calculation is completed, the macro- and micro-states of each data point in the three-dimensional finite element model of the metallic material are updated and used as input conditions for the next time frame. This simulation process continues until the calculations for all time frames are completed, or the forming process terminates. Figure 7 The three-dimensional finite element model of the metal workpiece obtained after a specific basic simulation process is shown.

[0118] While performing basic forming simulations of metallic materials, the above groups Similarly, this loading condition was used to conduct multiple real forming tests on metallic materials, i.e., basic forming tests. Figure 8 It shows the relationship with Figure 7 Workpieces obtained by conducting real forming tests under the same combination of loading parameters.

[0119] After each basic test is completed, sampling and measurement are carried out at several specific parts of the formed metal workpiece. For example, the metal workpiece is cut at specific parts to obtain macroscopic observation data (such as force-displacement curves) and microstructure data (such as grain size, dislocation density, phase composition ratio, etc. obtained from microscopic image analysis).

[0120] In step A3, sampling measurements are performed at several locations on the workpiece obtained from the basic forming test corresponding to each set of basic loading parameter combinations to obtain macro- and micro-state measurement values. These values ​​are then compared with the simulation results at the same locations in the three-dimensional finite element model of the metal material obtained from the basic forming simulation to obtain the basic residual samples corresponding to the basic simulation forming results. It is evident that for any combination of basic loading parameters, the data volume of the basic residual sample is determined by the number of sampling and measurement locations on the workpiece obtained from the actual experiment. Especially if it is necessary to measure the macroscopic and microscopic characteristics inside the workpiece, destructive measurements such as cutting the workpiece may be required. Therefore, for smaller workpieces, the number of sampling and measurement locations is generally around tens, while for larger workpieces it may reach hundreds, with a maximum of no more than 10. 3 Otherwise, it will exceed the limits of acceptable testing time and cost.

[0121] In step A4, the residual estimation model is trained using basic residual samples. As analyzed above, estimating the simulation result residuals at all data points using residual samples from a finite number of measurement locations requires the estimation model to have the ability to handle the complex nonlinear relationship between macro-micro parameter residuals in order to eliminate the heterogeneity of macro-micro parameters. In general deep learning networks, the layer adjustment weights do not have explicit physical constraints. To enable them to accurately estimate residuals, a large training set is necessary, which is limited by real experimental conditions. Therefore, this application proposes a new residual estimation model, a three-layer Gaussian regression model that integrates macro-micro coupling mechanisms. Specifically, as follows... Figure 9 As shown, the model consists of a macroscopic layer, a microscopic layer, and a macro-micro coupling layer, and all of these layers are Gaussian regression models.

[0122] Gaussian regression is a nonparametric regression model based on probability and statistics theory. It assumes that all possible function values ​​follow a multivariate Gaussian distribution. This model measures the correlation between different inputs by constructing a covariance function (kernel function). Unlike traditional regression models, Gaussian regression can not only estimate the expected value of the target variable, but also provide the uncertainty (variance) of the estimation results. This uncertainty reflects the reliability of the estimation results and can provide richer auxiliary reference information for subsequent applications of the estimation results.

[0123] Specifically, refer to Figure 9 The input to the macroscopic layer is the combination of loading parameters used in the simulation of metal forming, and its kernel function includes, for example, parameters used to characterize the effects of strain, stress, and temperature. , and The output is the preliminary residual estimate of the macroscopic flow parameters at each data point in the simulation results of this loading parameter combination, along with their uncertainties, such as the preliminary residual estimates of stress, strain, and temperature. , and And the uncertainty (i.e., variance) of the aforementioned preliminary residual estimates, such as , and This generates a correction basis that covers the entire process window.

[0124] The input to the micro-layer includes the combination of loading parameters and the output of the macro-layer. Its kernel function reflects the constraints of the microstructure mechanism, such as the influence of dislocation density and average grain size. , The output consists of residual estimates of microstructure parameters for each data point in the simulation results of each combination of basic loading parameters, such as residual estimates of dislocation density and grain size. , And the uncertainty of the above estimates, such as the variance of the above residual estimates. , Such as to capture the microscopic nonlinear characteristics during the material forming process.

[0125] The macro-micro coupling layer, based on the outputs of the macro and micro layers, outputs the residual estimates and uncertainties of macroscopic flow parameters for each data point in the simulation results. In the residual coupling model proposed in this application, the macro-micro coupling layer is used to achieve dynamic fusion of macroscopic and microscopic residuals. It takes the preliminary estimates of macroscopic flow parameter residuals output by the macro layer, the estimated microscopic parameter residuals output by the micro layer, and the combination of loading parameters as inputs. Through at least one coupling kernel function included in this layer, it analyzes the interaction between macroscopic and microscopic parameter residuals, applying the influence of microscopic organizational parameters to the macroscopic flow parameter residuals. Finally, it outputs the macro-micro coupling residuals as the residual estimates of macroscopic flow parameters, such as stress, strain, and temperature residuals. , and And the uncertainties of the above estimates, such as the variance of the above residual estimates. , , wait.

[0126] The coupling layer, as its core layer, is used to capture and model the nonlinear coupling effect of micro-parameter residuals on macro-parameter prediction. By designing various heterogeneous coupling kernel functions, the interaction relationship between macro- and micro-parameters can be effectively integrated. For example, in... Figure 9 In the illustrated embodiment, the kernel function of the macro-micro coupling layer includes multiple macro-micro coupling kernel functions that reflect the interaction between macro and micro parameters, wherein... , , These respectively reflect the coupled influence of various microstructure parameters on macroscopic flow parameters such as stress, strain, and temperature. , These parameters respectively reflect the coupled effects of various macroscopic flow parameters on microscopic parameters such as dislocation density and average grain size.

[0127] Furthermore, the kernel function of the macro-micro coupling layer can characterize various coupling mechanisms, such as when stress changes are simultaneously affected by dislocation density. Strain hardening and average grain size caused by changes When dynamic softening caused by change is influenced by two opposing mechanisms, the macro-micro coupling kernel function reflecting the stress residual is considered. It can be constructed using the following formula:

[0128] ,

[0129] in, The macro-micro coupling kernel function corresponding to stress. To account for the effect of dynamic softening on the stress residual, To account for the effect of strain hardening on the stress residual, , These are the combinations of loading parameters to be estimated and the combinations of loading parameters that have already been trained, respectively. , For signal strength hyperparameters, , , These are the residual estimates of stress, average grain size, and dislocation density output by the macroscopic and microscopic layers, respectively. , , This is a length-scale hyperparameter.

[0130] It should be noted that when stress changes are influenced by more complex mechanisms, the corresponding macro-micro coupling kernel function takes a more complex form:

[0131] ,

[0132] in, , These are the influence terms of various microstructure parameters other than dynamic softening and strain hardening on the stress residual. Their functional forms can be found in [reference]. , .

[0133] During the basic training of the residual estimation model, the training set consists of each group as well as Each group Input the residual estimation model, select the data points corresponding to the sampling measurement locations from the residual estimates of the macroscopic flow parameters and microstructure parameters of all the data points output by the model, and use the deviation between the residual estimates of these data points and the basic residual samples to optimize the parameters of the residual estimation model, thereby completing the initial training of the residual estimation model.

[0134] The residual estimation model that has completed basic training has the ability to make preliminary estimates of the residuals of the simulation results. Since the residual estimation model can estimate the residuals of all data points in the simulation results, that is, it can evaluate the deviation between the simulation results and the real experiment from the whole rather than just at a very limited number of measurement positions. Therefore, in step A5 of this application, by analyzing the statistical characteristics of the residual estimates of all data points in the simulation results, it is found that the overall level of the residuals shows significantly abnormal loading parameter intervals. In these intervals, the experimental density is increased in a targeted manner, which can not only significantly improve the representativeness of the loading parameter combinations used for training, but also avoid the problem of increasing the density of all loading parameter windows indiscriminately, which would cause a large number of redundant experiments.

[0135] In some preferred embodiments, in order to select a suitable combination of incremental loading parameters, a basic trained residual estimation model is first used to evaluate each group. Estimate the residuals of the simulation results and obtain the results for each group. In the simulation results, the residual estimates of macroscopic flow parameters for each data point at each time frame are obtained. Then, the residual estimates for each combination of loading parameters are normalized, and the mean of the normalized results is calculated. and its standard deviation Then, the evaluation function corresponding to each combination of basic loading parameters is estimated according to the following formula:

[0136] ,

[0137] in, The sequence number of the basic loading parameter combination. The total number of combinations of basic loading parameters. For the first Group basic loading parameter combination, for In the corresponding forming process data, the mean of the normalized estimated values ​​of the macroscopic flow parameter residuals for each data point at each time frame is... for standard deviation for The rate of change of the spatial coordinates of the loading parameters, , These are the weighting coefficients.

[0138] and This reflects the overall deviation between the simulation results of the macroscopic properties of various parts of the metallic material and the actual macroscopic properties. The first term in the above formula represents the overall deviation evaluation term. The larger this term is, the greater the overall deviation between the simulation results of the macroscopic flow parameters and the actual macroscopic properties under a specific combination of loading parameters. Therefore, it is necessary to increase the number of real experiments around the loading parameter combination. The second term in the above formula is the nonlinear jump evaluation term, which represents the change of the overall residual level of the macroscopic flow parameters with the change of the loading parameter combination. The larger this term is, the more uniform the overall level of the macroscopic property residual in the simulation results has changed at that point. Therefore, it is necessary to increase the density of experiments near that position in the loading combination space to conduct more refined experiments and measurement comparisons in these loading parameter combination intervals where property changes may occur. Based on this, the training sample can be expanded, which can effectively increase the representativeness of the training sample and improve the training effect of the residual estimation model.

[0139] Determine the incremental loading parameter combination using A6. Then, in steps A7-A9, incremental loading parameter combinations can be used to perform incremental forming simulation and incremental forming experiments on metallic materials, thereby obtaining incremental residual samples. And incremental training is performed on the residual estimation model.

[0140] After the residual estimation model completes incremental training, the accuracy of estimating the residuals of the simulation results is further improved. Subsequently, through step A10, the parameter value intervals described above are specified in the loading parameter space with a finer density. By combining loading parameters, a high-density loading parameter combination can be obtained. (Clearly, the number of high-density loading parameter combinations is much greater than the number of basic loading parameter combinations and incremental loading parameter combinations), then proceed through step A11. The forming simulation of a combination of high-density loading parameters is performed, and in step A12, the macro and micro residuals of the simulation results of each combination of high-density loading parameters are estimated using a residual estimation model and the simulation results are corrected. Finally, the high-density loading parameter combination-forming process data is obtained, thereby realizing the construction of the forming process database.

[0141] <Optimal Selection of Molding Scheme>

[0142] Since the forming process database stores the complex macro-micro evolution "paths" of metal materials under different loading parameters in high density, and these "paths" exhibit characteristics such as separation, aggregation, and intersection of macro and micro states, in step two, the optimal solution to achieve the expected forming goal can be searched and obtained in the high-dimensional data space composed of loading parameters-macro-micro states-time frames formed by these structured data.

[0143] Specifically, in some preferred embodiments, the loading scheme for the metallic material from its initial macro-micro state to its target macro-micro state is determined through the following steps:

[0144] B1, to obtain the initial macro- and micro-states and target macro- and micro-states of the metallic material.

[0145] refer to Figure 3 The initial macro- and micro-states of metallic materials generally correspond to the macroscopic and microscopic parameters of metal blanks with the same specifications and dimensions, such as the size of the blank and the overall macroscopic flow parameters and microstructure parameters before processing. These data can be obtained by sampling and measuring samples from a batch of blanks. Correspondingly, the target macro- and micro-states correspond to the macro- and micro-states of the final formed workpiece.

[0146] Based on the preceding text Figure 3 Analysis shows that different combinations of loading parameters can result in "forming paths" that produce workpieces with similar overall shapes and sizes, but still exhibit differences in macroscopic and microscopic states. These differences include variations in stress, strain, lattice size, and dislocation density distribution across different regions within workpieces of the same shape and size, thus reflecting the quality of the final workpiece. Therefore, in the embodiments of this application, the characterization of the target state of the metal material includes not only the overall shape and size but also macroscopic flow parameters and microstructure parameters at various data points.

[0147] B2, construct the global evolutionary quality function.

[0148] The global evolution quality function is used to characterize the evolution quality of a metallic material as it gradually evolves from its initial macro-micro state (including overall shape and size, stress, strain, temperature, dislocation density, and lattice size) to its final macro-micro state when a loading action is applied to the metallic material according to a time series of a certain combination of alternative loading parameters, with the target macro-micro state as the forming target. In other words, whether the metallic material can evolve from its initial macro-micro state in an ideal or relatively ideal form and ultimately achieve or approach the target macro-micro state as closely as possible.

[0149] Obviously, applying different combinations of loading parameters at different time frames will cause different material points of the metal to evolve at different speeds and in different ways. The accumulation of these evolution processes eventually leads to different macro- and micro-state characteristics of workpieces with the same or similar shapes, and ultimately different forming qualities. Therefore, when searching for the optimal "loading path", the global evolution quality function used to evaluate the evolution process should include at least the following terms: target macro- and micro-state deviation term, evolution process uniformity evaluation term, maximum change rate of macro-state term, and maximum transformation rate of micro-state term.

[0150] Among them, the target macro-micro state deviation term is used to evaluate the deviation between the final macro-micro state of the workpiece and the target macro-micro state when processing a metal billet according to a candidate loading parameter combination sequence; the evolution process uniformity evaluation term is used to determine whether the candidate loading parameter combination sequence will cause large fluctuations in the internal stress, strain, microstructure, etc. of the metal material during the entire forming process; the maximum rate of change of macro-state and the maximum rate of change of micro-state are used to constrain the changes of macro-state and micro-state, respectively, to avoid unexpected abrupt changes in macro-micro characteristics at certain times.

[0151] In some specific embodiments, a global evolutionary mass function from the initial macro-microstate to the final macro-microstate can be constructed as shown in the following equation. :

[0152] ,

[0153] in, , , , These are, respectively, the target macro-micro state deviation term, the evolution process uniformity evaluation term, the macro state maximum change rate term, and the micro state maximum transformation rate term. , , , These are the weighting coefficients for each evaluation item. , These are the data point and time frame numbers, respectively. This represents the total number of data points for metallic materials. The total number of time frames for loading the scheme. For time frames The expected macroscopic parameter vector represents the parameters expected in time frames when loading is performed using alternative combinations of loading parameters. At the expected macroeconomic state, correspondingly, when and hour, and Representing the initial time and the final time respectively. The expected macroeconomic situation This is the expected micro-parameter vector, representing the parameters at time frames when loading is performed using alternative combinations of loading parameters. At the expected microstate, correspondingly, when and hour, and Representing the initial time and the final time respectively. The expected microstate, and This constitutes the initial macro-micro state. and This constitutes the final macro-micro state. Let the target be the macroscopic parameter vector. Let be the target microscopic parameter vector. and Constituting the target macro- and micro-states, , The normalization coefficient is... The time frame interval.

[0154] B3 searches the process database to obtain the time series of loading parameter combinations with the optimal global evolution quality function value and generates a loading scheme for the metallic material from the initial macro-micro state to the final macro-micro state.

[0155] Specifically, various path search algorithms known to those skilled in the art, such as genetic algorithms, particle swarm optimization algorithms, and Bayesian optimization algorithms, can be used to search for time series of loading parameter combinations that enable the macro- and micro-states of the metallic material to evolve according to a relatively ideal trend, with the goal of optimizing the global evolutionary quality function, and ultimately achieving the same or as closely as possible the target macro- and micro-states. At the same time, extract from the database During loading, the expected macro- and micro-states of the metallic material are determined, thereby generating a loading scheme for the metallic material from its initial macro- and micro-states to its final macro- and micro-states.

[0156] In this application, the expected combination of loading parameters applied to the metal material at each time frame and the expected macro- and micro-states of the metal material obtained in this way are referred to as the loading scheme for metal material forming.

[0157] Figure 10 A preferred loading scheme is schematically illustrated, which loads from the initial time to the time frame. use Figure 3 Loading parameter combinations As can be seen from the previous analysis, the combination of loading parameters The forming process is relatively long, and its internal microstructure, such as grain arrangement and dislocation density distribution, evolves more uniformly, which is conducive to guiding the strengthening of the microstructure in the early stage of forming; the last 500 time frames of this scheme use a combination of loading parameters. Loading is performed because in Under loading, metallic materials The macro and micro states at the point are basically the same as In time frame The loading scheme is consistent at that point, so it can be changed to [a different scheme] at that time. To accelerate the overall forming process, the entire forming process was completed in 1500 time frames.

[0158] As can be seen, in the embodiments of this application, the loading scheme of the metal material from the initial macro-micro state to the final macro-micro state may include two or more different time series of high-density loading parameter combinations, thereby achieving a good balance between forming quality and forming time.

[0159] <Monitoring the loading process>

[0160] In step three, the metal material is loaded and shaped using the loading scheme obtained from the search. Since the actual loading amount on the metal material often differs from the expected loading amount, it is necessary to monitor the metal forming process in real time. When there is a significant deviation between the actual macro- and micro-states and the expected macro- and micro-states, corrective measures are taken in time through steps four and five.

[0161] Specifically, in this application, the monitoring of the forming process is carried out through the following steps:

[0162] The real-time loading parameter combinations at each sampling time point are continuously acquired at preset time intervals. Then, by looking up tables and searching, based on the high-density loading parameter combination-forming process data stored in the forming process database, the system continuously estimates the real-time loading combination of the metal material according to the time evolution. Under loading, the actual macro and micro states at various moments In step four, the expected macro- and micro-states under the preset loading scheme are determined. (That is, the expected macroscopic parameter vector and the expected microscopic parameter vector at each time frame of the loading scheme are compared. When the deviation exceeds the set threshold, the loading scheme adjustment process in step five is initiated.)

[0163] The determination of the deviation between the actual and expected macro- and micro-states, as well as the selection of the threshold, can be flexibly adjusted according to the actual requirements and indicators of metal material processing. For example, the actual and expected values ​​of the spatial location, macro-flow parameters, and micro-structure parameters of each data point of the metal material can be statistically analyzed to obtain the differences between the statistical results. Then, different weights can be set according to different processing and forming requirements to perform weighted summation (e.g., if the overall forming shape meets the design indicators, the weight of the macro-morphology deviation term can be increased; if the goal is to achieve a good micro-structure, the weight of the micro-state deviation term can be increased). The summation result can then be used as the deviation between the actual and expected macro- and micro-states, and compared with a pre-set threshold. The above methods for determining the deviation and comparing the threshold are known to those skilled in the art and will not be elaborated here.

[0164] Existing adaptive loading optimization techniques typically employ trained predictive models that predict future states by real-time acquisition of state parameters during the forming process, such as temperature distribution, and then adjust the loading based on deviations from the target. However, in actual metal forming processes, some key state parameters are often difficult to obtain. For example, in forging, the critical locations affecting forming quality are where the metal material contacts the die; however, because the material continuously interacts with the die at these locations, the stress and strain at these locations cannot be obtained by deploying sensors. Furthermore, for the same reasons as the residual estimation analysis mentioned earlier, measuring state parameters at limited locations cannot comprehensively estimate the deviation between the overall state of the metal material and the expected state.

[0165] Therefore, this application improves the monitoring method in the existing technology. Its core idea is to monitor the actual loading parameter combination that is easy to obtain (e.g., loading rate, loading temperature, etc. can be measured relatively simply and directly), and then use the high-density loading parameter combination-forming process data in the forming process database to comprehensively compare the real-time macro and micro states of all data points of the metal material with the expected macro and micro states according to the original loading scheme, so as to achieve accurate control of the real state during the forming process.

[0166] <Adjustment of molding scheme>

[0167] Clearly, by using steps three and four, we can not only determine whether the real-time state of the metallic material deviates significantly from the state it should be in according to the loading scheme, but also directly determine the moment when this deviation exceeds a threshold (this moment can be expressed as...). The actual macroscopic and microscopic states of metallic materials can be determined in step five. Real-time macro- and micro-states of metallic materials The loading scheme is re-searched and re-entered as a new initial macro / micro state.

[0168] In some specific embodiments, step five further includes the following steps:

[0169] C1, to obtain the macro and micro states of the metallic material when the deviation exceeds a set threshold;

[0170] C2, Construct the residual stage evolution quality function, which is used to characterize the evolution quality of the macro-micro state of the metallic material during the forming process from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state, with the target macro-micro state as the forming target.

[0171] C3 searches the process database to obtain the loading parameter combination time series with the optimal remaining stage evolution quality function value and generates a loading scheme for the metallic material from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state.

[0172] The implementation of steps C1 to C3 is largely the same as that of steps B1 to B3, except that the estimated macro-micro state of the metallic material at the moment when the deviation exceeds a set threshold is used as the new initial macro-micro state. Therefore, the evolution mass function of the remaining stages... Used for evaluation The evolution quality from time point A to the final time point. In some optional embodiments, It can be adopted with Same form.

[0173] In some other preferred embodiments, it is also possible to... The progress evaluation item is included because, during the metal forming process, deviations from the preset loading scheme in the actual loading parameter combination not only cause discrepancies between the actual and expected macro- and micro-states, but also cause the metal forming progress to deviate from expectations, for example, during scheme adjustments. The actual shape of the formed metal material lags behind the shape expected according to the original plan. Therefore, in the process of searching for new loading schemes, it is necessary to add a schedule evaluation item that constrains the overall forming time schedule, and give priority to those paths that can complete the remaining loading in a faster time to make up for the previous delay.

[0174] Figure 11 The diagram schematically illustrates an adjustment to the loading scheme in a specific embodiment, through... Figure 11 It can be seen that because the loading amount of the real-time loading parameter combination is smaller than that of the original loading scheme (e.g., the actual loading frequency is lower than the loading frequency in the loading scheme), the evolution of the real-time macro and micro states lags behind expectations, and in... This resulted in a significant deviation, clearly indicating that loading could no longer be performed according to the original loading scheme. Therefore, in A new loading scheme was obtained through a re-search. The new loading scheme selected a path with a faster evolution speed, and finally, within 1500 time frames, the metal material was formed basically according to the target macro-micro state.

[0175] The specific embodiments of this application have been described in detail above. For those skilled in the art, several improvements and modifications can be made to this application without departing from the principle of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for adaptive optimization of loading parameters driven by a forming process database, characterized in that, Includes the following steps: Step 1: Construct a forming process database for metallic materials. The forming process database stores structured high-density loading parameter combinations and forming process data obtained through simulation. Step 2: Search the forming process database to obtain the loading scheme of the metal material from the initial macro-micro state to the final macro-micro state. The loading scheme includes the expected combination of loading parameters applied to the metal material at each time frame and the expected macro-micro state of the metal material. Step 3: Continuously load and shape the metal material based on the loading scheme and obtain real-time loading parameter combinations, and continuously estimate the real-time macro and micro states of the metal material based on the real-time loading parameter combinations and the data in the forming process database; Step 4: Determine whether the deviation between the expected macro / micro state and the real-time macro / micro state exceeds a set threshold. If the determination result is yes, proceed to step 5; otherwise, return to step 3. Step 5: Re-search to obtain the loading scheme of the metal material from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state, and then return to step 3; The forming process database is constructed through the following steps: A1. Determine several sets of basic loading parameter combinations from the loading parameter window of the metallic material; A2, Basic forming simulation and basic forming test of metallic materials based on basic loading parameter combinations; A3 generates basic residual samples based on the results of basic forming simulation and basic forming test. A4, Use the basic residual samples to perform basic training on the residual estimation model; A5. The residual level of the basic forming simulation results is evaluated using the residual estimation model after basic training. A6. Based on the evaluation results, select several sets of incremental loading parameter combinations from the loading parameter window; A7, Incremental forming simulation and incremental forming test of metallic materials based on incremental loading parameter combination; A8 generates incremental residual samples based on incremental forming simulation and incremental forming test results; A9, use incremental residual samples to incrementally train the residual estimation model; A10, Generate high-density load parameter combinations from the load parameter window, wherein the number of high-density load parameter combinations is much greater than the number of basic load parameter combinations; A11, based on the high-density loading parameter combination, the high-density forming simulation of metallic materials is carried out, and the simulation results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metallic material under the loading of each set of high-density loading parameter combination are obtained in each time frame. A12 uses an incrementally trained residual estimation model to correct the high-density forming simulation results, thereby obtaining forming process data corresponding to each combination of high-density loading parameters.

2. The adaptive optimization method for loading parameters driven by the forming process database according to claim 1, characterized in that, The high-density loading parameter combination covers the corresponding loading parameter window in each dimension of the loading parameter space in a manner that is no greater than the preset upper limit of the loading parameter value interval; For any set of high-density loading parameters, the corresponding forming process data includes the simulation correction results of the macroscopic and microscopic parameters of each data point of the three-dimensional finite element model of the metal material at each time frame under the loading of the set of high-density loading parameters.

3. The adaptive optimization method for loading parameters driven by the forming process database according to claim 1, characterized in that, The residual estimation model is a three-level Gaussian regression model that integrates macro-micro coupling mechanisms, including: At the macroscopic level, based on the input combination of loading parameters, the output simulation results show the preliminary residual estimates of the macroscopic flow parameters at each data point and their uncertainties. The micro-layer, based on the combination of input loading parameters and the output results of the macro-layer, outputs the residual estimates and uncertainties of the micro-organism parameters of each data point in the simulation results. The macro-micro coupling layer, based on the outputs of the macro and micro layers, outputs the residual estimates and uncertainties of the macroscopic flow parameters at each data point in the simulation results.

4. The adaptive optimization method for loading parameters driven by the forming process database according to claim 3, characterized in that, The combination of incremental loading parameters is determined based on the evaluation function shown below: ， in, The sequence number of the basic loading parameter combination. The total number of combinations of basic loading parameters. For the first Group basic loading parameter combination, for In the corresponding forming process data, the mean of the normalized estimated values ​​of the macroscopic flow parameter residuals for each data point at each time frame is... for standard deviation for The rate of change of the spatial coordinates of the loading parameters, , These are the weighting coefficients.

5. The adaptive optimization method for loading parameters driven by the forming process database according to claim 2, characterized in that, The following steps are used to search for and determine the loading scheme for metallic materials from their initial macro-micro state to their final macro-micro state: B1, to obtain the initial macro- and micro-states and target macro- and micro-states of metallic materials; B2, Construct a global evolution quality function, which is used to characterize the evolution quality of the macro-micro state of a metallic material during the forming process from the initial macro-micro state to the final macro-micro state, with the target macro-micro state as the forming target. B3 searches the forming process database to obtain the loading parameter combination time series with the optimal global evolution quality function value and generates the loading scheme of the metal material from the initial macro-micro state to the final macro-micro state.

6. The adaptive optimization method for loading parameters driven by the forming process database according to claim 5, characterized in that, The global evolution quality function includes at least the following terms: target macro-micro state deviation term, evolution process uniformity evaluation term, macro-state maximum change rate term, and micro-state maximum transformation rate term.

7. The adaptive optimization method for loading parameters driven by the forming process database according to claim 5, characterized in that, The loading scheme for the metallic material from the initial macro-micro state to the final macro-micro state includes two or more different combinations of high-density loading parameters.

8. The adaptive optimization method for loading parameters driven by the forming process database according to claim 5, characterized in that, The following steps are used to search for and determine the loading scheme for the metallic material from the macro-micro state at the moment when the deviation exceeds a set threshold to the final macro-micro state: C1, to obtain the macro and micro states of the metallic material when the deviation exceeds a set threshold; C2, Construct the residual stage evolution quality function, which is used to characterize the evolution quality of the macro-micro state of the metallic material during the forming process from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state, with the target macro-micro state as the forming target. C3 searches the forming process database to obtain the loading parameter combination time series with the optimal remaining stage evolution quality function value and generates a loading scheme for the metal material from the macro-micro state when the deviation exceeds the set threshold to the final macro-micro state.

9. The adaptive optimization method for loading parameters driven by the forming process database according to claim 8, characterized in that, The remaining stage evolution quality function includes at least the following terms: target macro-micro state deviation term, evolution process uniformity evaluation term, macro-state maximum change rate term, micro-state maximum transformation rate term, and progress evaluation term.

Images