Precipitated alloy performance prediction method and system fused with physical law

Abstract

The invention discloses an alloy performance prediction method and system fused with a physical law. The method comprises the following steps: obtaining a characteristic factor set of all elements in a periodic table of elements; performing feature screening to obtain key features; inputting the key features into a constructed and trained element performance prediction model, and predicting to obtain a performance value of each element; and then the alloy elements are compared with the corresponding preset threshold values, and the alloy elements meeting the requirements are obtained. The method comprises the following steps: setting component data of an alloy in combination with existing alloy elements, and calculating and obtaining process parameters and aging precipitation parameters during alloy manufacturing according to thermodynamic data of the alloy; and the component data, the process parameters and the aging precipitation parameters of the alloy serve as input and are input into the performance prediction model of the precipitation type alloy fused with the physical law, and the corresponding aging performance is output. According to the method, the problem of multi-element alloy component and process multi-factor strong coupling efficient integrated design is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of material design, and in particular to a method and system for predicting the performance of precipitation-type alloys integrating physical laws. Background Art

[0002] With the rapid development of the current industry, especially in high-tech industries such as computers, semiconductors, instrumentation, information and communications, high load, high reliability and long service life have become the main development direction of conductor functional devices, which makes the service conditions of alloys used in signal transmission, structural support and thermal management systems more demanding, and therefore puts forward higher requirements on the mechanical properties of materials. In view of the excellent performance of copper alloys, aluminum alloys, titanium alloys, magnesium alloys and other metal materials in the manufacture of integrated circuit lead frames, precision electronic components, connectors and other aspects, they have become key structural functional materials in many high-tech fields. Considering the national development strategy, the preparation of metal materials with ultra-high strength and good conductivity has extremely high application value and industry potential.

[0003] Microalloying is one of the important means to improve the comprehensive properties of alloys. Adding Fe, Co, Nb, V, Ti, Si, P, Zr, Zn, and B to Cu-Ni-Sn alloys can effectively inhibit the nucleation and growth of discontinuous precipitation phases. For example, the V element can change the growth mode of the precipitation phase, promote the precipitation of the second phase and improve the softening resistance of the alloy, thereby improving the electrical conductivity of the alloy. As a typical precipitation-strengthened alloy, the amplitude modulation structure (poor Sn, rich Sn) of Cu-Ni-Sn can be gradually transformed into γ-DO by ordered phase transformation. 22 and γ-L1 2 Ordered phase, significantly improving the mechanical properties and electrical conductivity of the alloy; but discontinuous precipitation of γ-DO 3 The phase nucleates and grows at the grain boundary, which seriously affects the mechanical properties of the alloy. Therefore, the high-level matching of the alloy composition and process becomes the key to balancing the mechanical properties and electrical conductivity of the alloy.

[0004] Faced with the huge exploration space of alloy composition and process, efficient alloy design methods have become the key to rapid material development. The first principles calculate the phase mechanical properties of alloys from the perspective of atomic and electron migration, thereby revealing the influencing laws of their mechanical properties and electrical properties. The phase diagram is used to calculate the thermodynamic properties of alloys at different temperatures, and the type and volume fraction of precipitated phases can be predicted. The above methods provide effective guidance for material development to a certain extent, but quantitative prediction of alloy properties is still challenging. Machine learning, as a multi-field interdisciplinary subject involving probability theory and statistics, is essentially a method that uses algorithms to allow computers to efficiently and autonomously construct complex nonlinear relationships between input data and output targets from existing data, and use this relationship to build a model to predict unknown data. If this method can be applied to material research and development, it is expected to achieve performance-oriented efficient material design and make up for the shortcomings of the above calculation methods.

[0005] Therefore, in the alloy design process, there are still contradictions in alloy performance and problems that need to be solved urgently by existing machine learning methods. Summary of the invention

[0006] The present invention provides a prediction method and system for the performance of precipitation-type alloys that integrate physical laws. The method solves the problem of efficient design of multi-factor strong coupling of multi-element alloy components and processes, and realizes efficient and precise integrated design of high-performance alloy components, processes, structure and performance.

[0007] In a first aspect, the present invention provides a method for predicting the properties of precipitation-type alloys incorporating physical laws, comprising:

[0008] S1: Obtain the physical and chemical properties of all elements in the periodic table and obtain a set of element characteristic factors;

[0009] S2: Feature screening of feature factor set is performed through correlation analysis and recursive feature elimination to obtain key features;

[0010] S3: The key features are input into a plurality of different element performance prediction models that have been constructed and trained to predict the performance values ​​of each element; the element performance prediction model is mainly a performance model for the inverse relationship, and the performance includes hardness, conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.;

[0011] S4: comparing the predicted performance value of each element with the corresponding preset threshold value, and screening out one or more alloy elements that meet the performance requirements;

[0012] S5: Based on the obtained multiple alloying elements and in combination with the existing alloying elements, the composition data of the alloy is set, and the process parameters and aging precipitation parameters when manufacturing the alloy are calculated according to the thermodynamic data of the alloy;

[0013] S6: The alloy composition data, process parameters, and aging precipitation parameters are input into an alloy performance prediction model based on the constructed and trained fusion physical laws, and the corresponding aging performance is output.

[0014] Furthermore, the physical and chemical property characteristics of the elements in S1 include atomic size information, electronic information, energy information, and physical information.

[0015] Furthermore, the specific process of S2 is:

[0016] S21: Performing a preliminary screening of the element characteristic factor set using the Pearson correlation coefficient to obtain a preliminary screening result; wherein the preliminary screening result includes characteristic factors whose correlation coefficient values ​​between any characteristic factors and the correlation coefficient values ​​between the characteristic factors and the target are both greater than the corresponding preset preliminary screening threshold. In specific implementation, the preliminary screening threshold can be set according to the actual application scenario, is not limited to it, and is set to 0.9;

[0017] S32: Based on the initial screening results, recursive feature elimination is performed to perform fine feature screening, and the machine learning model is used for judgment, and the lowest RMSE value is used as the key feature for final screening;

[0018] The calculation formula of RMSE is as follows:

[0019]

[0020] Among them, MAPE is the mean absolute error of the model; V Predictive value. and V Measured value. are the model prediction value and the true value of the sample respectively, and N is the number of samples.

[0021] Furthermore, the construction process of the element performance prediction model in S3 is specifically as follows:

[0022] S31: Acquire a first historical alloy data set, wherein each alloy data in the data set includes alloy composition data, structure data, and alloy performance data; the alloy includes multiple elements;

[0023] S32: performing data conversion on the first historical alloy data set, converting each alloy component data in the historical alloy data set into a physical and chemical property characteristic factor of a corresponding element;

[0024] S33: Pearson correlation coefficient is used to calculate the correlation coefficient between any characteristic factors and the correlation coefficient between the characteristic factors and the target, and the initial feature screening is performed, and the screening threshold is set to 0.9;

[0025] S34: Based on the results after the initial screening, feature fine screening is performed in combination with recursive feature elimination, and the judgment is made through the machine learning model. The feature factor with the lowest RMSE value is used as the key feature affecting the element performance, and the key features and the corresponding element performance are combined into a key feature factor combination; wherein, the machine learning module can adopt support vector regression SVR, random forest RF, linear regression LR, artificial neural network ANN, decision tree DT, extreme gradient boosting XGBOOST, etc.

[0026] S35: Divide the key feature factor combination into a training set, a test set, and a validation set according to a preset ratio; take the key features as input and the element performance value as output, pre-train the element performance prediction model built based on the machine learning model, and obtain the element performance prediction model.

[0027] Furthermore, the construction process of the alloy performance prediction model integrating physical laws is as follows:

[0028] S101: Acquire a second historical alloy data set; wherein each alloy data in the data set includes alloy composition data, process parameters, performance data of the solid solution-cold working-aging process stages under different conditions, and aging precipitation parameters; wherein the solid solution state performance includes hardness, electrical conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.; the cold working performance includes hardness, electrical conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.; the aging state performance includes hardness, electrical conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.;

[0029] S102: using an optimization algorithm to construct a large number of composite models with different model hyperparameter combinations according to data conditions, taking alloy composition data, process parameters, performance data of solution-cold working-aging process stages under different conditions, and aging precipitation parameters as inputs, and aging state performance as outputs, and performing hyperparameter self-optimization and model self-screening on the composite model; wherein the composite model combines a time series model and a deep learning model; the time series model can adopt a long-term and short-term neural network model LSTM, a recursive neural network RNN, a recursive neural network GRU, a Transformer, etc.;

[0030] S103: Obtain an alloy performance prediction model that integrates physical laws based on the model generalization ability and test accuracy.

[0031] Furthermore, in the prediction process of the alloy performance prediction model integrating physical laws:

[0032] The alloy composition data is used as the input of the first time unit of the time series model to predict the corresponding solid solution properties;

[0033] The solid solution properties, process parameters and hidden cell unit Ct are used as the input of the second time unit of the time series model to obtain the alloy properties after aging; wherein the aging precipitation parameters of the alloy are integrated into the second time unit;

[0034] The alloy properties after aging and the hidden cell unit Ct1 are used as the input of the deep learning model to obtain the corresponding aging state performance.

[0035] Furthermore, the different conditions of the solutionizing-cold working-aging process stages are as follows:

[0036] The cold working process deformation includes 50%, 70%, and 90%; the aging temperatures include 300℃, 325℃, 350℃, 375℃, 400℃, 425℃, 450℃, 470℃, and 500℃; the aging time includes 15min, 30min, 60min, 120min, 180min, and 240min.

[0037] Furthermore, the hyperparameters include the number of neural network layers, the number of neuron nodes, the optimizer, the learning rate, and the dropout rate. The number of neural network layers ranges from 2 to 10, the number of neuron nodes ranges from 8 to 1024, the optimizer is Adam, RMSprop, or SGD, and the learning rate is 10 -6 Up to 10 -1 , the dropout rate is 0.01 to 0.6

[0038] In a second aspect, the present invention provides a performance prediction system for precipitation alloys integrating physical laws, comprising:

[0039] Data acquisition and preprocessing module: used to obtain the physical and chemical properties of all elements in the periodic table and obtain the characteristic factor set; the characteristic factor set is screened by correlation analysis and recursive feature elimination to obtain key features;

[0040] Element screening module: used to input key features into multiple different element performance prediction models that have been built and trained, and predict the performance value of each element; wherein the element performance includes hardness, conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.; compare the predicted performance value of each element with the corresponding preset threshold value, and screen out one or more alloy elements that meet the performance requirements;

[0041] Alloy performance prediction module: Based on the obtained alloying elements and combined with the existing alloying elements, the alloy composition data is set, and the process parameters and aging precipitation parameters for manufacturing the alloy are calculated based on the thermodynamic data of the alloy; the obtained alloy composition data, processing technology, and aging precipitation parameters are used as input to the constructed and trained alloy performance prediction model that integrates physical laws, and the corresponding aging performance is output.

[0042] Beneficial Effects

[0043] The present invention proposes a prediction method and system for the performance of precipitation-type alloys that integrate physical laws. The method targets the performance contradictions of alloys and the dilemma faced by existing machine learning methods in the process of new material research and development, integrates phase change precipitation laws and physical neural networks, drives an efficient and accurate design method for alloy composition-process integration, and provides a new perspective and approach for exploring the design and mechanism research of high-performance alloys with complex components. The method uses the spatial exploration capability of machine learning models based on small samples to efficiently screen unknown elements, and combines thermodynamics to predict the alloy aging precipitation sequence temperature; combines the time series model and the deep learning model DNN network structure, integrates the phase change precipitation law for modeling, captures the intrinsic relationship between "alloy composition-process-organization transformation law-performance", and solves the problem of efficient integrated design with strong coupling of multi-factor alloy composition and process. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a flow chart of a prediction method of a performance prediction model of an alloy integrating physical laws provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0047] Alloy design is divided into two stages: Stage 1 is the selection of microalloying elements; Stage 2 is composition optimization and preparation (melting → homogenization → hot working / cold working → solution → cold working → aging). For stage 1, all candidate elements are independent of each other, and model construction belongs to "parallel" development; for stage 2, all processes have a sequence, and model construction belongs to "serial" development. It is difficult to solve the dilemma faced by both stages at the same time if a single machine learning method is used. Deep neural network is a multi-layer feedforward parallel machine learning method that uses topological structure to deal with nonlinear problems. It has a fast processing speed and strong generalization ability. Long short-term memory network is a special recursive neural network that stores and forgets important information and noise information in each process by constructing cell units and gate units (forget gate, input gate, output gate). Its complex internal structure may lead to insufficient training efficiency and generalization ability.

[0048] Example 1

[0049] like Figure 1 As shown, this embodiment provides a method for predicting the properties of precipitation-type alloys that integrates physical laws. This embodiment takes the prediction of Cu-Ni-Sn-X (X is an unknown element or multiple elements) alloy from element selection to performance prediction as an example to be described in detail, including:

[0050] S1: Obtain the physical and chemical property characteristics of all elements in the periodic table and obtain a set of characteristic factors.

[0051] All elements in the elemental table are known. The physical and chemical properties of elements not only affect the performance of the elements, but also the performance of the alloys. Among them, the physical and chemical properties of elements include the atomic size information, electronic information, energy information, and physical information of each element. Among them, atomic size information includes: lattice constant (a, b, c), atomic radius (experimental value and calculated value), Cordero covalent radius and Pyykko covalent radius; electronic information includes: valence electron concentration, electronegativity (Pauling, Sanderson, Allredrochow, Ghosh-Gupta, Nagle, Smith), electronic energy band, valence electron, electron affinity, conductivity, resistivity, contact potential and photoelecrtic work function; energy information includes: melting point, boiling point, enthalpy (fusion, vaporization), cohesive energy, heat of fusion, heat of vaporization, specific heat, thermal conductivity and thermal expansion; physical information includes: atomic mass, nuclear charge number, density, mole fraction, atomic hardness (Brimell, Mohs, Bulk), shear modulus, Young's modulus, Poisson's ratio, family number, period number, space group number. The above information is combined to obtain the characteristic factor set.

[0052] S2: Feature screening of feature factor set is performed through correlation analysis and recursive feature elimination to obtain key features.

[0053] S21: Use the Pearson correlation coefficient to perform a preliminary screening of the element characteristic factor set to obtain a preliminary screening result; wherein, the preliminary screening result includes the characteristic factors whose correlation coefficient values ​​between any characteristic factors and the correlation coefficient values ​​between the characteristic factors and the target are both greater than the corresponding preset preliminary screening threshold. The specific process is: take the characteristic factor set as input, use the Pearson correlation analysis in the machine learning feature engineering packaging method to screen the characteristic factors in the input respectively, and calculate the Pearson correlation coefficient between any alloy factor (X, Y) and the correlation coefficient between the characteristic factor and the target according to the following formula. By comparing the correlation coefficients between the characteristic factors and the characteristic factors, and between the characteristic factors and the target (organization, performance) in turn, a characteristic factor combination that can better describe the target is screened. When the absolute value of the correlation between the characteristic factors is greater than 0.90, it means that there is a strong correlation between the two characteristic factors X and Y, and the information they contain and the degree of contribution to the model construction are similar. In order to prevent the model from being overfitted due to excessive data dimensions, the correlation between the above two alloy characteristic factors and the target (organization, performance) is determined, and the one with a higher degree of correlation with the target is retained.

[0054]

[0055] Among them, r is the correlation coefficient; x i is the i-th characteristic factor; x m is the average value of the characteristic factor; x j are j characteristic factors.

[0056] S32: Based on the results of the initial screening, recursive feature elimination is performed to perform feature fine screening, and the SVR model is used for judgment, and the lowest RMSE value is used as the key alloy element feature for the final screening;

[0057] Specifically, the REF gradient descent method in machine learning feature engineering is used to establish a support vector model with the remaining feature factors and the target. To prevent the model from overfitting or underfitting, a 10-fold cross validation is used to calculate the model accuracy corresponding to each gradient descent. The calculation method of the model error is shown in the following formula. By obtaining a feature number-model error curve and a feature importance curve, the key feature factor combination that has the greatest impact on the target is screened out according to the change in the obtained data model error.

[0058] The calculation formula of RMSE is as follows:

[0059]

[0060] Among them, MAPE is the mean absolute error of the model; V Predictive value. and V Measured value. are the model prediction value and the true value of the sample respectively, and N is the number of samples.

[0061] S3: Take the key features as inputs and input them into multiple different element performance prediction models that have been built and trained to predict the performance values ​​of each element; among them, the element performance prediction model is mainly for performance models with inverted relationships. Performance includes hardness, conductivity, strength, elongation, softening resistance, wear resistance, corrosion resistance, etc.; in specific implementation, it can be selected according to actual needs. In this embodiment, alloy hardness and conductivity are taken as examples of inverted relationships. Therefore, the element performance prediction model includes a hardness prediction model and a conductivity prediction model.

[0062] Specifically, S31: acquiring a first historical alloy data set, wherein each alloy data in the data set includes composition data, structure data, and performance data of the alloy; the alloy includes multiple elements;

[0063] S32: performing data conversion on the first historical alloy data set, converting each alloy component data in the historical alloy data set into physical and chemical property characteristics of corresponding elements;

[0064] S33: using the Pearson correlation coefficient to calculate the correlation coefficient values ​​between any characteristic factors and the correlation coefficient values ​​between the characteristic factors and the target, and performing a preliminary feature screening. The setting of the screening threshold can be modified according to the actual application scenario, and is set to 0.9 in this embodiment;

[0065] S34: Based on the results after the initial screening, recursive feature elimination is combined to perform feature fine screening, and the machine learning model is used for judgment. The physical and chemical property characteristics with the lowest RMSE value are used as the key features affecting the performance of the element. The key element characteristics and the corresponding element performance are combined into key factor characteristics. The type of machine learning model used can be adjusted according to the actual situation. The SVR model is used in this implementation.

[0066] S35: Divide the key factor feature combination into a training set, a test set, and a validation set according to a preset ratio; take the key features as input and the element performance value as output, pre-train the alloy performance prediction model constructed based on the SVR model, and obtain the element performance prediction model.

[0067] S4: Compare the predicted performance value of each element with the corresponding preset threshold value, and screen out one or more alloy elements that meet the performance requirements. It should be noted that the preset threshold value can be set according to actual conditions and is not limited. In this embodiment, the preset threshold value is 0.9; when the predicted performance value of each element is greater than the preset threshold value, it indicates that the element meets the performance requirements.

[0068] S5: Based on the obtained alloying elements and in combination with the existing alloying elements, the composition data of the alloy is set, and the process parameters and aging precipitation parameters when manufacturing the alloy are calculated according to the thermodynamic data of the alloy. The composition data of the alloy is set so that the sum of the mass fraction and molar fraction of all alloying elements of each alloy is equal to 100%.

[0069] Thermodynamic data of each alloy calculated by Thermo-calc, including solvus temperature, phase transition temperature, phase type, and phase volume fraction.

[0070] The process of each alloy processing includes: smelting → homogenization → hot working / cold working → solution treatment → cold working → aging;

[0071] Process parameters include hot working / cold working deformation, hot working / cold working temperature, and aging temperature / time;

[0072] The aging precipitation parameters include an ordered phase transformation factor PT and a transformation efficiency factor DT;

[0073] PT=AGT-AT

[0074]

[0075] Among them, AGT is the aging temperature in the sample; AT is the highest spinodal temperature of the TTT curve corresponding to the sample composition; BT is the fastest spinodal precipitation temperature of the TTT curve corresponding to the sample composition.

[0076] Take Cu-Ni-Sn alloy as an example: when PT>0, the alloy directly forms discontinuous DO during the aging process. 3 phase; when PT<0, the alloy undergoes an ordered phase transformation during the aging process. When DT>0, the precipitation sequence of the alloy during the aging process is "Spinodal→γ-DO 22 ", the larger the value, the faster the atomic diffusion speed, the more difficult it is to form the amplitude modulation structure, and the γ-DO precipitates in the alloy matrix. 22 When DT<0, the precipitation sequence of the alloy during the aging process is "Spinodal→γ-DO 22 →γ-L1 2 ", the smaller the value, the slower the atomic diffusion rate, the more stripe organization, γ-DO 22 and γ-L1 2 The slower the formation speed, but with the extension of aging time, γ-DO precipitates in the alloy matrix 22 and γ-L1 2 As the content of MgO increases, a complete ordered phase transition occurs.

[0077] S6: The obtained alloy composition data, process parameters, and aging precipitation parameters are used as inputs to the performance prediction model of the precipitation alloy that integrates physical laws and is constructed and trained, and the corresponding aging performance is output. In this embodiment, the processing parameters include cold working deformation CR, artificial aging temperature AGT, and artificial aging time AGt. The performance prediction model of the precipitation alloy that integrates physical laws is a composite model that combines a timing model and a deep learning model. The timing model can use LSTM, RNN, GRU, Transformer, etc. The timing model in this embodiment is constructed and trained using a long-term and short-term neural network model combined with a deep learning model (i.e., LSTM-DNN) as an example to obtain a performance prediction model of the precipitation alloy that integrates physical laws.

[0078] Specifically, the alloy performance prediction model integrating physical laws uses the LSTM model to distinguish the alloy composition design and processing technology design by time units, and then combines it with DNN to build it. The construction process is:

[0079] S101: Acquire a second historical alloy data set; wherein each alloy data in the data set includes alloy composition data, process parameters, performance data of the solid solution-cold working-aging process stages under different conditions, and aging precipitation parameters; wherein the solid solution state properties include hardness and electrical conductivity; the cold working state properties include hardness and electrical conductivity; and the aging state properties include hardness and electrical conductivity;

[0080] S102: Use the optimization algorithm to build a large number of LSTM-DNN models with different model hyperparameter combinations according to the data situation, take the alloy composition data, process parameters, performance data of the solution-cold working-aging process stage under different conditions, aging precipitation parameters as input, and aging state performance as input, and perform hyperparameter self-optimization and model self-screening on the LSTM-DNN model;

[0081] S103: Obtaining a prediction model of the alloy performance prediction model integrating physical laws according to the model generalization ability and the test accuracy.

[0082] More specifically, in the prediction process of the alloy performance prediction model integrating physical laws:

[0083] The alloy composition is used as the input of the first time unit of the LSTM model to predict the corresponding solid solution properties;

[0084] The solid solution properties, process parameters and hidden cell unit Ct are used as the input of the second time unit of the LSTM model to obtain the alloy properties after aging; among them, the aging precipitation parameters of the alloy are integrated into the second time unit;

[0085] The alloy properties after aging and the hidden cell unit Ct1 are used as the input of the DNN model to obtain the corresponding aging state properties. Taking Cu-Ni-Sn alloy as an example, the specific process is as follows: in the first time unit, the alloy composition (Cu, Ni, Sn, X) is used as input to predict the corresponding solid solution state properties ht (MH0, EC0) and use it as the information input of the next unit; in the second time unit, the process parameters (CR, AGT, AGt), the previous solid solution state properties ht (MH0, EC0) and the hidden cell unit (Ct) are used as input to output the alloy properties after aging ht1 (MH1, EC1); at the same time, a third time unit (composed of DNN) is added after the second time unit. This stage uses the alloy properties (MH1, EC1) and hidden cell unit (Ct1) predicted in the previous stage as input, and predicts the corresponding aging properties (MH1_, EC1_) as output. In addition, the aging precipitation behavior of the alloy is incorporated into the second time unit as the second important parameter to improve the model's physical characterization ability of the phase change process during the training process. The Bayesian optimization algorithm is used to optimize the model hyperparameters during the model training process.

[0086] Furthermore, the different conditions of the solutionizing-cold working-aging process stages are as follows:

[0087] The cold working process deformation includes 50%, 70%, and 90%; the aging temperatures include 300℃, 325℃, 350℃, 375℃, 400℃, 425℃, 450℃, 470℃, and 500℃; the aging time includes 15min, 30min, 60min, 120min, 180min, and 240min.

[0088] Furthermore, the hyperparameters include the number of neural network layers, the number of neuron nodes, the optimizer, the learning rate, and the dropout rate. The number of neural network layers ranges from 2 to 10, the number of neuron nodes ranges from 8 to 1024, the optimizer is Adam, RMSprop, or SGD, and the learning rate is 10 -6 Up to 10 -1 , the dropout rate is 0.01 to 0.6

[0089] Example 2

[0090] This embodiment provides a prediction system for precipitation alloy properties integrating physical laws, including:

[0091] Data acquisition and preprocessing module: used to obtain the physical and chemical properties of all elements in the periodic table and obtain the element characteristic factor set; perform feature screening on the alloy element characteristic factor set through correlation analysis and recursive feature elimination to obtain key element characteristics;

[0092] Element screening module: used to input the key alloy element characteristics as input to multiple different element performance prediction models that have been built and trained, and predict the performance value of each element; the element performance prediction model is mainly aimed at performance models with inverted relationships, and the performance includes hardness, conductivity, strength, elongation, softening resistance, wear resistance, corrosion, etc.; compare the predicted performance value of each element with the corresponding preset threshold value, and screen out one or more alloy elements that meet the performance requirements;

[0093] Alloy performance prediction module: Based on the obtained alloy elements, the alloy composition data is set, and the process parameters and aging precipitation parameters when manufacturing the alloy are calculated according to the thermodynamic data of each alloy element; the alloy composition data, process parameters, and aging precipitation parameters are used as input to the alloy performance prediction model that integrates physical laws and has been built and trained, and the corresponding aging performance is output;

[0094] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0095] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

[0096] It should be understood that in the embodiments of the present invention, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the type of device.

[0097] The readable storage medium is a computer-readable storage medium, which may be an internal storage unit of the controller described in any of the foregoing embodiments, such as a hard disk or memory of the controller. The readable storage medium may also be an external storage device of the controller, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the controller. Furthermore, the readable storage medium may also include both an internal storage unit of the controller and an external storage device. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0098] Based on this understanding, the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned readable storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., various media that can store program codes.

Claims

1. A method for predicting the properties of precipitation-type alloys incorporating physical laws, characterized in that: include: S1: Obtain the physical and chemical property characteristics of all elements in the periodic table and obtain the characteristic factor set; S2: Feature screening of feature factor set is performed through correlation analysis and recursive feature elimination to obtain key features; S3: The key features are input into the constructed and trained element performance prediction model to predict the performance value of each element; S4: comparing the predicted performance value of each element with the corresponding preset threshold value, and screening out one or more alloy elements that meet the performance requirements; S5: Based on the obtained alloying elements and in combination with the existing alloying elements, the composition data of the alloy is set, and the process parameters and aging precipitation parameters when manufacturing the alloy are calculated according to the thermodynamic data of the alloy; S6: The alloy composition data, process parameters, and aging precipitation parameters are input into the alloy performance prediction model that integrates physical laws and has been constructed and trained, and the corresponding aging performance is output.

2. The method for predicting the properties of precipitation-type alloys incorporating physical laws according to claim 1, characterized in that: The physical and chemical property characteristics of the elements in S1 include atomic size information, electronic information, energy information, and physical information.

3. The prediction method of the alloy performance prediction model integrating physical laws according to claim 1, characterized in that: The specific process of S2 is: S21: Performing a preliminary screening of the characteristic factor set using the Pearson correlation coefficient to obtain a preliminary screening result; wherein the preliminary screening result includes characteristic factors whose correlation coefficient values ​​between any characteristic factors and the correlation coefficient values ​​between the characteristic factors and the target are both greater than the corresponding preset preliminary screening threshold value; S32: Based on the initial screening results, recursive feature elimination is performed to perform fine feature screening, and the machine learning model is used for judgment, and the lowest RMSE value is used as the key feature for final screening; The calculation formula of RMSE is as follows: Among them, MAPE is the mean absolute error of the model; V Predictivevalue. and V Measuredvalue. are the model prediction value and the true value of the sample respectively, and N is the number of samples.

4. The method for predicting the properties of precipitation-type alloys incorporating physical laws according to claim 1, characterized in that: The construction process of the element performance prediction model in S3 is specifically as follows: S31: Acquire a first historical alloy data set, wherein each alloy data in the data set includes composition data, structure data, and performance data of the alloy; the alloy includes multiple elements; S32: performing data conversion on the first historical alloy data set, converting each alloy component data in the historical alloy data set into a physical and chemical property characteristic factor of a corresponding element; S33: using the Pearson correlation coefficient to calculate the correlation coefficient values ​​between any characteristic factors and the correlation coefficient values ​​between the characteristic factors and the target, to perform initial feature screening; S34: Based on the results of the initial screening, recursive feature elimination is combined to perform feature fine screening, and the machine learning model is used to make a judgment. The feature factor with the lowest RMSE value is used as the key feature affecting the performance of the element, and the key features and the corresponding element performance are combined into a key feature factor combination; S35: Divide the key feature factor combination into a training set, a test set, and a validation set according to a preset ratio; take the key features as input and the element performance value as output, pre-train the element performance prediction model built based on the machine learning model, and obtain the element performance prediction model.

5. The prediction method of the alloy performance prediction model integrating physical laws according to claim 1, characterized in that: The process of constructing the performance prediction model of the alloy integrating physical laws is as follows: S101: Acquire a second historical alloy data set; wherein each alloy data in the data set includes alloy composition data, process parameters, performance data of the solid solution-cold working-aging process stage under different conditions, and aging precipitation parameters; S102: using an optimization algorithm to construct a large number of composite models with different model hyperparameter combinations according to data conditions, taking alloy composition data, process parameters, performance data of solution-cold working-aging process stages under different conditions, and aging precipitation parameters as inputs, and aging state performance as outputs, and performing hyperparameter self-optimization and model self-screening on the composite model; the composite model combines a time series model and a deep learning model; S103: Obtain an alloy performance prediction model that integrates physical laws based on the model generalization ability and test accuracy.

6. The method for predicting the properties of precipitation-type alloys incorporating physical laws according to claim 5, characterized in that: The prediction process of the alloy performance prediction model that integrates physical laws: The alloy composition data is used as the input of the first time unit of the time series model to predict the corresponding solid solution properties; The solid solution properties, process parameters and hidden cell unit Ct are used as the input of the second time unit of the time series model to obtain the alloy properties after aging; wherein the aging precipitation parameters of the alloy are integrated into the second time unit; The alloy properties after aging and the hidden cell unit Ct1 are used as the input of the deep learning model to obtain the corresponding aging state performance.

7. The method for predicting the properties of precipitation-type alloys incorporating physical laws according to claim 5, characterized in that: The different conditions in the solution-cold working-aging process are as follows: The cold working process deformation includes 50%, 70%, and 90%; the aging temperatures include 300℃, 325℃, 350℃, 375℃, 400℃, 425℃, 450℃, 470℃, and 500℃; the aging time includes 15min, 30min, 60min, 120min, 180min, and 240min.

8. The method for predicting the properties of precipitation-type alloys incorporating physical laws according to claim 5, characterized in that: Hyperparameters include the number of neural network layers, the number of neuron nodes, the optimizer, the learning rate, and the loss rate. The number of neural network layers ranges from 2 to 10, the number of neuron nodes ranges from 8 to 1024, the optimizer is Adam, RMSprop, or SGD, and the learning rate is 10 -6 Up to 10 -1 , dropout rate is 0.01 to 0.6 9. A prediction system for precipitation alloy properties integrating physical laws, characterized in that: include: Data acquisition and preprocessing module: used to obtain the physical and chemical property characteristics of all elements in the periodic table and obtain the characteristic factor set; The feature factor set is screened by correlation analysis and recursive feature elimination to obtain key features; Element screening module: used to input key element characteristics into multiple different element performance prediction models that have been built and trained, and predict the performance value of each element; compare the predicted performance value of each element with the corresponding preset threshold value, and screen one or more alloy elements that meet the performance requirements; Alloy performance prediction module: Based on the obtained alloying elements and combined with the existing alloying elements, the alloy composition data is set, and the process parameters and aging precipitation parameters when manufacturing the alloy are calculated based on the thermodynamic data of the alloy; the alloy composition data, process parameters, and aging precipitation parameters are used as input to the performance prediction model of the alloy that integrates physical laws and is constructed and trained, and the corresponding aging performance is output.

Images