Forging process simulation parameter recommendation method

Abstract

The embodiment of the invention relates to the technical field of metal plastic processing, and provides a forging process simulation parameter recommendation method which comprises the following steps: constructing a finite element system model based on a geometric model corresponding to a forge piece, and performing simulation processing on the forge piece through the finite element system model to obtain multiple pieces of simulation result data corresponding to the forge piece; acquiring a plurality of experimental data corresponding to the forge piece through acquisition equipment; optimizing the multiple pieces of simulation result data and the multiple pieces of experiment data based on target complexity to obtain optimized data; wherein the target complexity comprises single target optimization or multi-target optimization; and based on the optimization data, determining an optimal simulation parameter corresponding to the finite element system model. Thus, the optimal simulation parameters can be obtained according to the simulation result data and the experiment data, simulation is conducted through the optimal simulation parameters, the geometric prediction error of the forge piece can be effectively reduced, the load prediction precision is improved, and the simulation accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal plastic working, in particular to a forging process simulation parameter recommendation method. BACKGROUND

[0002] Forging process is one of the important processes of metal material processing, and its process involves high temperature, high pressure and complex material deformation. Traditional forging process optimization usually relies on experimental trial and error method, which not only consumes time and materials, but also is difficult to cover all combinations of process parameters comprehensively. Therefore, with the development of computer technology, finite element simulation technology is introduced into the research of forging process, so that the key data of the process can be quickly obtained through numerical simulation, thereby reducing the number of experiments and improving the research efficiency.

[0003] However, when performing finite element simulation, the material constitutive model and process parameters need to be set, and the selection of process parameters is usually based on past experience, and the parameters of the material constitutive model are usually fitted from experimental data or empirical formula, which has certain errors. Therefore, the accuracy of the finite element simulation result data is poor. SUMMARY

[0004] The present application provides a forging process simulation parameter recommendation method, which can improve the simulation accuracy.

[0005] In a first aspect, the present application provides a forging process simulation parameter recommendation method, comprising: Based on the corresponding geometric model of the forging, a finite element system model is constructed, and the forging is simulated by the finite element system model to obtain a plurality of simulation result data corresponding to the forging; A plurality of experimental data corresponding to the forging are collected by a collection device; The plurality of simulation result data and the plurality of experimental data are optimized based on a target complexity to obtain optimization data; wherein the target complexity includes single-objective optimization or multi-objective optimization; Based on the optimization data, the optimal simulation parameters corresponding to the finite element system model are determined.

[0006] In some embodiments, the plurality of simulation result data and the plurality of experimental data are optimized based on the target complexity to obtain optimization data, comprising: The plurality of simulation result data and the plurality of experimental data are aligned to obtain a plurality of aligned data; The plurality of aligned data are optimized based on the target complexity to obtain the optimization data.

[0007] In some embodiments, each of the aligned data includes target simulation result data and target experimental data; In a case where the target complexity is the single-target optimization, the optimization processing of the multiple alignment data based on the target complexity to obtain the optimization data comprises: determining multiple comprehensive errors corresponding to the multiple alignment data based on the target simulation result data and the target experimental data included in each of the multiple alignment data, wherein the comprehensive error is a weighted sum of a geometric error, a load error, and a temperature error; constructing a proxy model and iteratively updating the proxy model based on the multiple comprehensive errors until the proxy model converges to obtain multiple prediction data; determining the most potential prediction data in the multiple prediction data as the optimization data.

[0008] In some embodiments, in a case where the target complexity is the multi-target optimization, the optimization processing of the multiple alignment data based on the target complexity to obtain the optimization data comprises: randomly generating an initial population based on the multiple alignment data, and generating multiple reference points based on the initial population; non-dominantly sorting the initial population to obtain a non-dominant level corresponding to each of the initial population, and determining a distance between each of the initial population and a target reference point in the multiple reference points; wherein the target reference point is the reference point in the multiple reference points closest to the corresponding initial population; determining a parent population from the initial population based on the non-dominant level corresponding to each of the initial population and the distance between each of the initial population and the target reference point in the multiple reference points; generating a child population based on the parent population by a simulated binary crossover algorithm, merging the parent population and the child population to obtain a new initial population, and repeating the above steps until a preset condition is met, and outputting the optimization data.

[0009] In some embodiments, the determination of the optimal simulation parameter corresponding to the finite element system model based on the optimization data comprises: determination of the optimal simulation parameter corresponding to the finite element system model based on the optimization data by a target recommendation model; or, determination of the optimal simulation parameter corresponding to the optimization data from the material database based on the optimization data and outputting.

[0010] In some embodiments, the optimization data comprises material composition, material parameter, process parameter, and reference simulation parameter; wherein the material parameter comprises strain rate and strain amount; the process parameter comprises temperature data; and the reference simulation parameter comprises mold heat conduction data and friction coefficient. The target recommendation model determines the optimal simulation parameter corresponding to the finite element system model based on the optimization data, which includes: determining a thermal force coupling parameter based on the temperature data, the strain amount, and the strain rate; determining a mold material cross feature based on the mold heat conduction data and the friction coefficient; inputting the optimization data, the thermal force coupling parameter, and the mold material cross feature into the target recommendation model, and determining a candidate simulation parameter through the target recommendation model; determining an uncertainty corresponding to the candidate simulation parameter through the target recommendation model, and determining the candidate simulation parameter as the optimal simulation parameter when the uncertainty corresponding to the candidate simulation parameter is less than a first preset value.

[0011] In some embodiments, the method further includes: when the uncertainty corresponding to the candidate simulation parameter is greater than the first preset value and less than a second preset value, generating a plurality of sets of sample data, training the target recommendation model based on the sample data to obtain a new target recommendation model, inputting the optimization data, the thermal force coupling parameter, and the mold material cross feature into the new target recommendation model, and determining a candidate simulation parameter through the new target recommendation model; when the uncertainty corresponding to the candidate simulation parameter is greater than the second preset value, outputting risk information.

[0012] In some embodiments, after the target recommendation model determines the optimal simulation parameter corresponding to the finite element system model based on the optimization data, the method further includes: updating the material database based on the optimization data and the optimal simulation parameter.

[0013] In some embodiments, the method further includes: obtaining training data, validation data, and test data; training the initial recommendation model based on a loss function and the training data to obtain a first recommendation model; optimizing the first recommendation model through the validation data to obtain a second recommendation model; evaluating the second recommendation model through test data, and determining the second recommendation model with the best evaluation as the target recommendation model.

[0014] In some embodiments, the loss function is as follows:

[0015] wherein, λ is a friction coefficient, m is a heat conduction coefficient, h is a heat convection coefficient, α, β, γ are sensitive coefficients, MSE represents a mean square error loss function, Huber represents a Huber loss function, and CosLoss represents a cosine loss function.

[0016] The embodiment of the present application provides a forging process simulation parameter recommendation method, including constructing a finite element system model based on a corresponding geometric model of a forging, and performing simulation processing on the forging through the finite element system model to obtain a plurality of simulation result data corresponding to the forging; collecting a plurality of experimental data corresponding to the forging through a collection device; optimizing the plurality of simulation result data and the plurality of experimental data based on a target complexity to obtain optimized data; wherein the target complexity includes single-target optimization or multi-target optimization; determining optimal simulation parameters corresponding to the finite element system model based on the optimized data. In this way, the optimal simulation parameters can be obtained according to the simulation result data and the experimental data, and the simulation is performed through the optimal simulation parameters, which can effectively reduce the geometric prediction error of the forging and improve the load prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A flowchart of a forging process simulation parameter recommendation method provided by the embodiment of the present application is shown in the figure. Figure 2 A forging simulation diagram provided by the embodiment of the present application is shown in the figure. Figure 3 Another forging simulation diagram provided by the embodiment of the present application is shown in the figure. Figure 4 A flowchart of another forging process simulation parameter recommendation method provided by the embodiment of the present application is shown in the figure. Figure 5 A structural diagram of a forging process simulation parameter recommendation device provided by the embodiment of the present application is shown in the figure. Figure 6 A structural diagram of an electronic device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present application more clear, the technical solutions in the present application will be described clearly and completely in the following with reference to the drawings in the present application.

[0020] Forging process is one of the important processes of metal material processing, and its process involves high temperature, high pressure and complex material deformation. The traditional forging process optimization usually relies on experimental trial and error method, which not only consumes time and materials, but also is difficult to comprehensively cover all combinations of process parameters. Therefore, with the development of computer technology, finite element simulation technology is introduced into the forging process research, so that the key data of the process can be quickly obtained through numerical simulation, thereby reducing the number of experiments and improving the research efficiency.

[0021] However, when performing finite element simulation, it is usually necessary to set the material constitutive model and process parameters, and the selection of process parameters is usually based on past experience, and the parameters of the material constitutive model usually depend on experimental data fitting or empirical formula, which has certain errors. Therefore, the accuracy of the finite element simulation result data is poor.

[0022] To solve the above technical problems, the forging process simulation parameter recommendation method provided by the embodiments of the present application can obtain the optimal simulation parameters according to the simulation result data and the experimental data, and the simulation can be performed through the optimal simulation parameters, which can effectively reduce the geometric prediction error of the forgings, improve the load prediction accuracy, and improve the simulation accuracy.

[0023] The forging process simulation parameter recommendation method provided by the embodiments of the present application will be described below with reference to the accompanying drawings. Figure 1 It should be noted that the forging process simulation parameter recommendation method provided by the embodiments of the present application can be implemented by a control device, and the control device can be a server or a terminal device. The terminal device can include at least one of a personal computer, a notebook computer, a smart phone, a tablet computer and a portable wearable device; the server can include a stand-alone server or a server cluster composed of multiple servers, and the embodiments of the present application do not limit this.

[0024] As shown in the figure, the forging process simulation parameter recommendation method provided by the embodiments of the present application includes S101-S104. Figure 1

[0025] S101, based on the geometric model corresponding to the forgings, a finite element system model is constructed, and the forgings are simulated by the finite element system model to obtain a plurality of simulation result data corresponding to the forgings.

[0026] In some embodiments, the finite element system model is used to simulate the forgings, and the simulation result data is the data output by the finite element system.

[0027] Exemplarily, the simulation result data can include at least one of the displacement data corresponding to the forgings, the temperature field data and the grain size.

[0028] ​In some embodiments, a geometric model is first constructed according to the upper and lower molds and the blank, and then a finite element system model is constructed based on the geometric model, material parameters corresponding to the blank, load conditions (such as forging pressure), and boundary constraints (such as mold fixed positions). The material parameters include density, elasticity, plasticity, thermal conductivity, specific heat capacity, and the like.

[0029] Exemplarily, after obtaining the geometric model, the geometric model can be input into a simulation system, and then a hexahedral element type is selected, a corresponding material constitutive relation is guided, a mesh is automatically divided, a full connection relationship is set, a load condition and a boundary constraint are configured, and a finite element system model is constructed. After the finite element system model is constructed, the forging can be simulated by setting different simulation parameters and solving parameters, and a plurality of simulation result data is obtained. The simulation parameters include initial temperature, heating / cooling rate, heating / cooling rate, travel speed, friction coefficient, thermal conductivity coefficient, and heat convection coefficient.

[0030] S102, a plurality of experimental data corresponding to the forging is collected by a collection device.

[0031] In some embodiments, the collection device can include at least one of a three-dimensional scanner, an infrared thermal imager, and an electron backscatter diffraction instrument.

[0032] Exemplarily, the geometric data corresponding to the forging can be collected by the three-dimensional scanner, the surface temperature field of the forging can be collected by the infrared thermal imager, and the EBSD grain size after slicing the forging can be collected by the electron backscatter diffraction instrument.

[0033] S103, the plurality of simulation result data and the plurality of experimental data are optimized based on a target complexity, and optimized data is obtained.

[0034] In some embodiments, the target complexity includes single-target optimization or multi-target optimization. The single-target optimization refers to geometric size optimization. The multi-target optimization includes geometric size optimization, temperature optimization, and the like. The number and types of optimization targets included in the multi-target optimization are not limited in the embodiments of the present application.

[0035] Exemplarily, the target complexity can be determined according to a trigger instruction input by a user. For example, if the user only clicks the geometric size optimization button, only the trigger instruction corresponding to the geometric size optimization is triggered, and the target complexity is single-target optimization. If the user not only clicks the geometric size optimization button but also clicks the temperature optimization button, two trigger instructions of the geometric size optimization and the temperature optimization are triggered, and the target complexity is multi-target optimization.

[0036] In some embodiments, the optimizing the plurality of simulation result data and the plurality of experimental data based on the target complexity to obtain the optimized data can include: performing alignment processing on the plurality of simulation result data and the plurality of experimental data to obtain a plurality of alignment data; and performing optimization processing on the plurality of alignment data based on the target complexity to obtain the optimized data.

[0037] It can be understood that the alignment processing on the plurality of simulation result data and the plurality of experimental data can unify the format and standard between the simulation result data and the experimental data, and improve the recommendation accuracy.

[0038] Two optimization methods for optimizing the plurality of simulation result data and the plurality of experimental data are provided in the embodiments of the present application, and the control device can select a corresponding optimization method according to the target complexity to optimize the plurality of simulation result data and the plurality of experimental data.

[0039] In some embodiments, each alignment data includes target simulation result data and target experimental data. The target simulation result data is simulation result data aligned with the experimental data, and the target experimental data is experimental data aligned with the simulation result data.

[0040] When the target complexity is single-target optimization, the optimization processing on the plurality of alignment data based on the target complexity to obtain the optimized data includes: constructing a surrogate model, iteratively updating the surrogate model based on a sampling function and the plurality of alignment data until the surrogate model converges, determining the most potential alignment data, and determining the most potential alignment data as the optimized data.

[0041] In some embodiments, the surrogate model is used to evaluate the comprehensive error of each alignment data, and the comprehensive error is a weighted sum of the geometric error, the load error and the temperature error; wherein the respective weights of the geometric error, the temperature error and the load error are preset values, which can be set according to actual needs. The respective weights of the geometric error, the temperature error and the load error are not limited in the embodiments of the present application.

[0042] In some embodiments, when the target complexity is single-target optimization, an adaptive optimization Bayesian optimization algorithm can be used to perform optimization processing on the plurality of alignment data to obtain the optimized data.

[0043] Exemplarily, a surrogate model corresponding to a target function can be constructed by a Gaussian process, and the target function can be predicted by the surrogate model, wherein the target function is a black-box function for minimizing the comprehensive error. That is, the surrogate model can evaluate the comprehensive error corresponding to each alignment data.

[0044] In some embodiments, the sampling function can be an expected improvement function.

[0045] Exemplarily, 20 alignment data can be acquired, one of the 20 alignment data is input into the surrogate model, and the comprehensive error of the alignment data is evaluated by the surrogate object to complete one update of the surrogate model and obtain the comprehensive error corresponding to the alignment data. Then, a new alignment data is selected from the 20 alignment data by a sampling function (such as an expected improvement function) and input into the surrogate model, and the comprehensive error of the new alignment data is evaluated by the surrogate object to complete a second update of the surrogate model and obtain the comprehensive error corresponding to the new alignment data. The above steps are repeated in this way, when the optimal error in the comprehensive error is improved by less than 1% for 3 iterations, that is, it is determined that the surrogate model converges, at this time, the surrogate model will output the most potential alignment data, and the most potential alignment data is determined as the optimization data.

[0046] In some embodiments, the surrogate model can be represented as: prediction function (surrogate model) ≈ mean function + exploration factor × standard deviation function. Wherein, the exploration factor is initially set to 2.5 and is dynamically reduced with iteration.

[0047] In some embodiments, when the target complexity is multi-objective optimization, the optimization processing of the plurality of alignment data based on the target complexity to obtain the optimization data can include: generating an initial population based on the plurality of alignment data, and generating a plurality of reference points based on the initial population; performing non-dominated sorting on the initial population to obtain non-dominated levels corresponding to each initial population, and determining distances between each initial population and target reference points in the plurality of reference points; determining a parent population from the initial population based on the non-dominated levels corresponding to each initial population and the distances between each initial population and the target reference points in the plurality of reference points; generating a child population based on the parent population by a simulated binary crossover algorithm, and merging the parent population and the child population to obtain a new initial population and repeating the above steps until a preset condition is met, and outputting the optimization data.

[0048] In some embodiments, the target reference point is the reference point closest to the corresponding initial population in the plurality of reference points In some embodiments, when the target complexity is multi-objective optimization, the plurality of alignment data can be optimized by an NSGA-III algorithm to obtain the optimization data.

[0049] Exemplarily, 100 initial populations are randomly generated based on the plurality of alignment data, and a plurality of reference points are generated based on each initial population to maintain diversity of the solution set by using the reference points. Then, the 100 initial populations are non-dominantly sorted to obtain a non-dominant level corresponding to each initial population, and a distance between each initial population and a target reference point is determined; then, a plurality of initial populations with high non-dominant level and close distance to the target reference point are selected from the 100 initial populations as parent populations, and enter a cross-variation stage. After entering the cross-variation stage, the parent populations are recombined by a simulated binary crossover algorithm to generate offspring populations. Then, the parent populations and the offspring populations are combined to obtain new initial populations and repeat the above steps until a preset condition is met, and output the optimization data. The preset condition can be 250 iterations, or the solution set (optimization data) is unchanged for 20 consecutive generations.

[0050] In some embodiments, based on the optimization data, the optimal simulation parameter corresponding to the finite element system model is determined, comprising: The optimal simulation parameter corresponding to the finite element system model is determined based on the optimization data by the target recommendation model; or, based on the optimization data, the optimal simulation parameter corresponding to the optimization data is determined from the material database and output.

[0051] Exemplarily, when the parameter corresponding to the optimization data exists in the material database, the optimal simulation parameter corresponding to the optimization data can be determined from the material database based on the optimization data. When the parameter corresponding to the optimization data does not exist in the material database, the optimal simulation parameter corresponding to the finite element system model can be determined based on the optimization data by the target recommendation model.

[0052] Exemplarily, when the parameter corresponding to the optimization data exists in the material database, the optimal simulation parameter corresponding to the optimization data can be determined from the material database based on the optimization data. When the parameter corresponding to the optimization data does not exist in the material database, the optimal simulation parameter corresponding to the finite element system model can be determined based on the optimization data by the target recommendation model.

[0053] In some embodiments, the material database includes a mapping relationship between the material process parameters and the optimal simulation parameters. The material database can include material number, material composition, material parameter, applicable process, optimal simulation parameter, confidence index, etc. Since the optimization data is obtained based on the experimental data and the simulation result data corresponding to the forging, the optimization data can include the material composition of the forging, the material parameter, the process parameter, and the reference simulation parameter. The material composition is used to indicate the material composition of the blank corresponding to the forging; the material parameter is used to indicate the attribute of the blank corresponding to the forging; the process parameter is used to indicate the basic data or control variable required when the forging is performed; and the reference simulation parameter is used to indicate the simulation parameter configured when S101 is performed, i.e., the simulation result data is obtained based on the reference simulation parameter when S101 is performed. After the control device obtains the optimization data, it can determine whether there is an optimal simulation parameter corresponding to the optimization data in the material database, and if so, output the optimal simulation parameter.

[0054] In some embodiments, the target recommendation model can be a Transformer model. The Transformer model adopts a double-channel attention mechanism, which is component attribute attention and process gate attention respectively. The component attribute attention can be represented as . Wherein Q is the main element vector, K is the impurity vector, V is the attribute vector, is the weight matrix, is the scaling factor. The process gate attention introduces the LSTM gate mechanism. After the optimization data is input into the target recommendation model, the optimization data is subjected to feature fusion after normalization processing through the double channel, and the optimal simulation parameter will be obtained.

[0055] In some embodiments, after determining the optimal simulation parameter corresponding to the finite element system model based on the optimization data by the target recommendation model, the forging process simulation parameter recommendation method provided by the embodiments of the present application further comprises: updating the material database based on the optimization data and the optimal simulation parameter.

[0056] It can be understood that in this way, the data in the material database can be supplemented, so that the data in the material database is more comprehensive.

[0057] In some embodiments, the optimization data includes material composition, material parameters, process parameters and reference simulation parameters. The material composition includes main component content and impurity content; the material parameters include strain, strain rate, density, elasticity, plasticity, thermal conductivity, specific heat capacity, expansion coefficient; the process parameters include temperature data and forging data, the temperature data includes initial temperature, heating temperature, heating rate, cooling temperature and cooling rate; the forging data includes travel rate (forging rate) and die displacement amount; the reference simulation parameters include friction coefficient, die heat conduction data, heat convection coefficient.

[0058] Determining the optimal simulation parameter corresponding to the finite element system model based on the optimization data by the target recommendation model comprises: determining the thermal force coupling parameter based on the temperature data, the strain and the strain rate; determining the die material cross characteristic based on the die heat conduction data and the friction coefficient; inputting the optimization data, the thermal force coupling parameter and the die material cross characteristic into the target recommendation model to determine the candidate simulation parameter by the target recommendation model; determining the uncertainty corresponding to the candidate simulation parameter by the target recommendation model, and determining the candidate simulation parameter as the optimal simulation parameter when the uncertainty corresponding to the candidate simulation parameter is less than a first preset value.

[0059] In some embodiments, the temperature data and the strain amount in the process parameters can be multiplied to obtain a thermal coupling parameter; the mold heat conduction and the friction coefficient in the reference simulation parameters can be multiplied to obtain a mold material cross feature. Then, the optimization data, the thermal coupling parameter and the mold material cross feature are input into the target recommendation model to determine the candidate simulation parameters through the target recommendation model; finally, the dropout of the target recommendation model is enabled to determine the uncertainty corresponding to the candidate simulation parameters, and when the uncertainty corresponding to the candidate simulation parameters is less than a first preset value, the candidate simulation parameters are determined as the optimal simulation parameters.

[0060] Exemplarily, the first preset value can be 0.15. When the uncertainty is lower than 0.15, the candidate simulation parameters are determined as the optimal simulation parameters and output.

[0061] In some embodiments, when the uncertainty corresponding to the candidate simulation parameters is greater than the first preset value and less than a second preset value, a plurality of groups of sample data are generated, the target recommendation model is trained based on the sample data to obtain a new target recommendation model, and the optimization data, the thermal coupling parameter and the mold material cross feature are input into the new target recommendation model to determine the candidate simulation parameters through the new target recommendation model; when the uncertainty corresponding to the candidate simulation parameters is greater than the second preset value, risk information is output.

[0062] Exemplarily, the second preset value can be 0.25. When the uncertainty is between 0.15 and 0.25, virtual exploration is started, 50 groups of sample data are generated, and the target recommendation model is trained based on the sample data. When the uncertainty is higher than 0.25, risk information is generated. The risk information is used to recommend a staff to perform a physical experiment.

[0063] It can be understood that, in the embodiments of the present application, the simulation result data of the finite element simulation is combined with the experimental data, the simulation result data and the experimental data are optimized by using an optimization algorithm, and finally the optimal simulation parameters are obtained. Through simulation by using the optimal simulation parameters, the geometric prediction error of the forgings can be effectively reduced, and the load prediction accuracy can be improved.

[0064] Exemplarily, assuming that the material of the forgings is titanium-aluminum alloy, a stress diagram obtained by simulation through the prior art is as shown in Figure 2 A stress diagram obtained by simulation through the optimal simulation parameters obtained by executing S101-S104 is as shown in Figure 3 .

[0065] As shown in Figure 4 , in some embodiments, the forging process simulation parameter recommendation method provided by the embodiments of the present application further includes S401-S404.

[0066] S401, obtain training data, verification data and test data.

[0067] In some embodiments, the data set can be obtained from historical data and / or a database, and the data set is divided into a training set (training data), a validation set (validation data) and a test set (test data) in proportions of 70%, 20% and 10%.

[0068] S402, training the initial recommendation model based on the loss function and the training data to obtain a first recommendation model.

[0069] In some embodiments, the loss function is as follows:

[0070] wherein λ is a friction coefficient, m is a thermal conductivity coefficient, h is a heat convection coefficient, α, β, γ are sensitive coefficients, MSE represents a mean square error loss function, Huber represents a Huber loss function, and CosLoss represents a cosine loss function.

[0071] In some embodiments, the sensitive coefficients can be set according to actual needs, for example, α, β, γ can be 0.52, 0.42 and 0.16 respectively.

[0072] S403, optimizing the first recommendation model through the validation data to obtain a second recommendation model.

[0073] S404, evaluating the second recommendation model through the test data, and determining the second recommendation model with the best evaluation as the target recommendation model.

[0074] In some embodiments, the initial recommendation model can be trained through the training data (i.e., S402), the trained initial recommendation model (i.e., the first recommendation model) can be optimized through the validation data (i.e., S403), the optimized initial recommendation model (i.e., the second recommendation model) can be evaluated through the test data, and finally the model with the best evaluation is determined as the target recommendation model (i.e., step S404).

[0075] Corresponding to the foregoing embodiments of the forging process simulation parameter recommendation method, the present application also provides embodiments of a forging process simulation parameter recommendation device.

[0076] Referring to Figure 5 The present embodiment provides a forging process simulation parameter recommendation device, which comprises: A simulation module 501 is configured to construct a finite element system model based on a geometric model corresponding to a forging, and perform simulation processing on the forging through the finite element system model to obtain a plurality of simulation result data corresponding to the forging. A collection module 502 is configured to collect a plurality of experimental data corresponding to the forging through a collection device. The optimization module 503 is configured to optimize the plurality of simulation result data and the plurality of experimental data based on a target complexity to obtain optimization data, wherein the target complexity includes single-target optimization or multi-target optimization. The processing module 504 is configured to determine the optimal simulation parameter corresponding to the finite element system model based on the optimization data.

[0077] In some embodiments, the optimization module 503 is configured to perform alignment processing on the plurality of simulation result data and the plurality of experimental data to obtain a plurality of alignment data, and perform optimization processing on the plurality of alignment data based on a target complexity to obtain the optimization data.

[0078] In some embodiments, each of the alignment data includes target simulation result data and target experimental data. The optimization module 503 is configured to determine a plurality of comprehensive errors corresponding to the plurality of alignment data based on the target simulation result data and the target experimental data included in each of the alignment data, wherein the comprehensive error is a weighted sum of a geometric error, a load error, and a temperature error; construct a surrogate model, and iteratively update the surrogate model based on a plurality of the comprehensive errors until the surrogate model converges to obtain a plurality of prediction data; and determine the most promising prediction data in the plurality of prediction data as the optimization data.

[0079] In some embodiments, the optimization module 503 is configured to randomly generate an initial population based on the plurality of alignment data, and generate a plurality of reference points based on the initial population; perform non-dominated sorting on the initial population to obtain a non-dominated level corresponding to each of the initial population, and determine a distance between each of the initial population and a target reference point in the plurality of reference points; wherein the target reference point is the reference point in the plurality of reference points closest to the corresponding initial population; determine a parent population from the initial population based on the non-dominated level corresponding to each of the initial population and the distance between each of the initial population and the target reference point in the plurality of reference points; generate a child population based on the parent population by using a simulated binary crossover algorithm, and combine the parent population and the child population to obtain a new initial population and repeat the above steps until a preset condition is met, and output the optimization data.

[0080] In some embodiments, the processing module 504 is configured to determine the optimal simulation parameter corresponding to the finite element system model based on the optimization data, including: determining the optimal simulation parameter corresponding to the finite element system model based on the optimization data by using a target recommendation model; or determining the optimal simulation parameter corresponding to the optimization data from the material database based on the optimization data and outputting.

[0081] In some embodiments, the optimization data comprises material composition, material parameters, process parameters and reference simulation parameters; wherein the material parameters comprise strain rate and strain amount; the process parameters comprise temperature data; the reference simulation parameters comprise mold heat conduction data and friction coefficient; The processing module 504 is configured to determine thermal force coupling parameters based on the temperature data, the strain amount and the strain rate; determine mold material cross characteristics based on the mold heat conduction data and the friction coefficient; input the optimization data, the thermal force coupling parameters and the mold material cross characteristics into the target recommendation model, determine a candidate simulation parameter through the target recommendation model; determine an uncertainty corresponding to the candidate simulation parameter through the target recommendation model, and determine the candidate simulation parameter as the optimal simulation parameter when the uncertainty corresponding to the candidate simulation parameter is less than a first preset value.

[0082] In some embodiments, the processing module 504 is configured to generate a plurality of groups of sample data when the uncertainty corresponding to the candidate simulation parameter is greater than the first preset value and less than a second preset value, train the target recommendation model based on the sample data to obtain a new target recommendation model, input the optimization data, the thermal force coupling parameters and the mold material cross characteristics into the new target recommendation model, and determine a candidate simulation parameter through the new target recommendation model; and output risk information when the uncertainty corresponding to the candidate simulation parameter is greater than the second preset value.

[0083] In some embodiments, the processing module 504 is configured to update the material database based on the optimization data and the optimal simulation parameter.

[0084] In some embodiments, the apparatus further comprises an acquisition module configured to acquire training data, verification data and test data. The training module is configured to train the initial recommendation model based on a loss function and the training data to obtain a first recommendation model; optimize the first recommendation model based on the verification data to obtain a second recommendation model; evaluate the second recommendation model based on test data, and determine the second recommendation model with the best evaluation as the target recommendation model.

[0085] In some embodiments, the loss function is as follows:

[0086] Wherein, λ is the friction coefficient, m is the thermal conductivity coefficient, h is the heat convection coefficient, and α, β and γ are the sensitivity coefficients.

[0087] As Figure 6As shown, the electronic device provided by the embodiment of the present application can include a processor 610, a communications interface 620, a memory 630 and a communications bus 640, wherein the processor 610, the communications interface 620 and the memory 630 complete mutual communication through the communications bus 640. The processor 610 can invoke the logic instructions in the memory 630 to execute the above-mentioned methods.

[0088] In addition, the logic instructions in the memory 630 described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the switch device mechanical state monitoring method described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0089] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the above-mentioned methods.

[0090] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0091] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0092] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for recommending forging process simulation parameters, characterized in that: The method comprises the following steps: Based on the corresponding geometric model of the forging, a finite element system model is constructed, and the forging is simulated through the finite element system model to obtain a plurality of simulation result data corresponding to the forging; A plurality of experimental data corresponding to the forging are collected by a collection device; Based on a target complexity, the plurality of simulation result data and the plurality of experimental data are optimized to obtain optimization data; wherein the target complexity includes single-objective optimization or multi-objective optimization; Based on the optimization data, the optimal simulation parameters corresponding to the finite element system model are determined.

2. The method of claim 1, wherein, The method comprises the following steps: The plurality of simulation result data and the plurality of experimental data are aligned to obtain a plurality of alignment data; Based on the target complexity, the plurality of alignment data are optimized to obtain the optimization data.

3. The method of claim 2, wherein, Each of the alignment data includes target simulation result data and target experimental data; When the target complexity is the single-objective optimization, the plurality of alignment data are optimized based on the target complexity to obtain the optimization data, which comprises the following steps: An agent model is constructed; the agent model is used to evaluate the comprehensive error of each of the alignment data, and the comprehensive error is the weighted sum of geometric error, load error and temperature error; Based on a sampling function and the plurality of alignment data, the agent model is iteratively updated until the agent model converges, the most potential alignment data are determined, and the most potential alignment data are determined as the optimization data.

4. The method of claim 2, wherein, When the target complexity is the multi-objective optimization, the plurality of alignment data are optimized based on the target complexity to obtain the optimization data, which comprises the following steps: An initial population is randomly generated based on the plurality of alignment data, and a plurality of reference points are generated based on the initial population; The initial population is non-dominantly sorted to obtain the non-dominant level corresponding to each of the initial population, and the distance between each of the initial population and a target reference point in the plurality of reference points is determined; wherein the target reference point is the reference point in the plurality of reference points closest to the corresponding initial population; Based on the non-dominant level corresponding to each of the initial population and the distance between each of the initial population and the target reference point in the plurality of reference points, a parent population is determined from the initial population; A child population is generated based on the parent population by using a simulated binary crossover algorithm, the parent population and the child population are combined to obtain a new initial population, and the above steps are repeated until a preset condition is met, and the optimization data is output.

5. The method according to claim 3 or 4, characterized in that, Based on the optimization data, the optimal simulation parameters corresponding to the finite element system model are determined, which comprises the following steps: The optimal simulation parameters corresponding to the finite element system model are determined based on the optimization data by using a target recommendation model; or Based on the optimization data, the optimal simulation parameters corresponding to the optimization data are determined from a material database and output.

6. The method of claim 5, wherein, The optimization data includes material composition, material parameters, process parameters and reference simulation parameters; wherein, the material parameters include strain rate and strain amount; the process parameters include temperature data; the reference simulation parameters include mold heat conduction data and friction coefficient; The method further comprises: determining thermal force coupling parameters based on the temperature data, the strain amount and the strain rate; determining mold material cross characteristics based on the mold heat conduction data and the friction coefficient; inputting the optimization data, the thermal force coupling parameters and the mold material cross characteristics into the target recommendation model to determine a candidate simulation parameter through the target recommendation model; determining uncertainty corresponding to the candidate simulation parameter through the target recommendation model, and determining the candidate simulation parameter as the optimal simulation parameter when the uncertainty corresponding to the candidate simulation parameter is less than a first preset value.

7. The method of claim 6, wherein, The method further comprises: when the uncertainty corresponding to the candidate simulation parameter is greater than the first preset value and less than a second preset value, generating a plurality of sets of sample data, training the target recommendation model based on the sample data to obtain a new target recommendation model, inputting the optimization data, the thermal force coupling parameters and the mold material cross characteristics into the new target recommendation model, and determining a candidate simulation parameter through the new target recommendation model; when the uncertainty corresponding to the candidate simulation parameter is greater than the second preset value, outputting risk information.

8. The method of claim 5, wherein, After determining the optimal simulation parameter corresponding to the finite element system model based on the optimization data through the target recommendation model, the method further comprises: updating the material database based on the optimization data and the optimal simulation parameter.

9. The method of claim 5, wherein, The method further comprises: obtaining training data, verification data and test data; training an initial recommendation model based on a loss function and the training data to obtain a first recommendation model; optimizing the first recommendation model through the verification data to obtain a second recommendation model; evaluating the second recommendation model through test data, and determining the second recommendation model with the best evaluation as the target recommendation model.

10. The method of claim 9, wherein, The loss function is as follows: ； wherein, λ is the friction coefficient, m is the thermal conductivity coefficient, h is the heat convection coefficient, α, β, γ are the sensitivity coefficients, MSE represents the mean square error loss function, Huber represents the Huber loss function, and CosLoss represents the cosine loss function.

Images