Multi-component bonding phase metal-ceramic composite material component design method and system

Abstract

The invention discloses a multi-bonding-phase metal ceramic composite material component design method and a multi-bonding-phase metal ceramic composite material component design system. The method comprises the following steps: constructing a data set; dividing into a training subset and a test subset by adopting a K-fold cross validation mode, training to obtain a plurality of initial models, and evaluating and optimizing by utilizing the test subset to obtain a plurality of improved models; a prediction parameter set of the target multi-element bonding phase metal-ceramic composite material is constructed, and the improved model is traversed for loop execution to obtain final prediction data; and determining design parameters based on the final prediction data. According to the method, loop iteration prediction and a regression model algorithm are combined, and meanwhile, a mapping relation among the components, the process and the alloy performance of the metal-ceramic composite material is more accurately constructed by utilizing a K-fold cross validation algorithm, so that efficient and accurate prediction under the requirement of high performance is realized; the problems that an existing component design method for the multi-component bonding phase metal-ceramic composite material is long in period, low in accuracy and high in cost are solved.

Description

Technical Field

[0001] The invention relates to the technical field of composite material design, and in particular to a method and system for designing components of a multi-component bonding phase metal-ceramic composite material. Background Art

[0002] Metal-ceramic is a composite material with high hardness and high wear resistance. It is known as the "teeth of industry" and has become an indispensable key material for the development of modern industrial society. It is widely used in various fields such as the automotive industry, aerospace, and integrated circuit manufacturing. Traditional metal-ceramic composites usually use tungsten carbide (WC) as the hard phase and cobalt (Co) as the bonding phase. As a key strategic metal element, cobalt plays an irreplaceable role in modern industry and new energy technologies. Although China is the world's largest refined cobalt producer and major consumer, its domestic cobalt resource reserves are very limited, accounting for only 1% of the world's total reserves, and it has long relied on imports. Especially in lithium-ion batteries, cobalt, as an important component of the positive electrode material, greatly improves the energy density, cycle stability and safety of the battery. Therefore, with the rapid growth in demand for electric vehicles and portable electronic devices, the global demand for cobalt has increased significantly, consuming more than 50% of the world's cobalt production. This situation not only leads to drastic fluctuations in cobalt prices, but also poses a risk of resource supply. Therefore, the use of materials such as transition metal elements, rare earth elements and high entropy alloys to replace cobalt as a binder has become an important direction for the development of metal-ceramic composites.

[0003] However, the compositional combination space of multi-phase bonding metal-ceramic composites is complex and diverse. At present, material design mainly relies on traditional trial and error methods, which makes it difficult to achieve efficient and synchronous optimization of comprehensive performance, and even more difficult to achieve precise design of material composition and process, resulting in a long R&D cycle, low efficiency and high cost for new materials. Machine learning-driven material design is a frontier field in modern materials science.

[0004] The introduction of machine learning (ML) has changed this situation. By analyzing large amounts of experimental data, ML algorithms can reveal the complex relationship between material performance and composition and structure. Using this data, researchers can build predictive models to identify potential high-performance materials or optimize the performance of existing materials. In particular, advanced technologies such as deep learning and reinforcement learning can further improve the accuracy of predictions and the efficiency of design. Such technologies not only accelerate the material discovery process, but also reduce experimental costs and promote the rapid development of new functional materials.

[0005] However, for multi-phase bonding metal-ceramic composites with complex interactions, simple machine learning methods cannot meet the design requirements. It is necessary to develop appropriate design methods in a targeted manner to meet the requirements of efficiency and accuracy. However, there is still a blank in the technical solutions in this regard. Summary of the invention

[0006] In view of the deficiencies of the prior art, the object of the present invention is to provide a method and system for designing the composition of a multi-phase bonding metal-ceramic composite material.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention includes:

[0008] In a first aspect, the present invention provides a method for designing a composition of a multi-component bonding phase metal-ceramic composite material, comprising:

[0009] Constructing a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material;

[0010] The data set is divided into a training subset and a test subset by using a K-fold cross-validation method, multiple regression models are trained using the training subset to obtain multiple initial models, and the initial models are evaluated and optimized using the test subset to obtain multiple improved models;

[0011] Constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting the first prediction data into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved models in a loop to obtain final prediction data;

[0012] Based on the final prediction data, design parameters of the target multi-component bonding phase metal-ceramic composite material are determined.

[0013] In a second aspect, the present invention further provides a composition design system for a multi-component bonding phase metal-ceramic composite material, comprising:

[0014] A data construction module, used to construct a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material;

[0015] A model training module is used to divide the data set into a training subset and a test subset by using a K-fold cross-validation method, train multiple regression models using the training subset to obtain multiple initial models, and evaluate and optimize the initial models using the test subset to obtain multiple improved models;

[0016] A cyclic prediction module, used for constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting it into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved model for cyclic execution to obtain final prediction data;

[0017] The parameter determination module is used to determine the design parameters of the target multi-component bonding phase metal-ceramic composite material based on the final prediction data.

[0018] Based on the above technical solution, compared with the prior art, the beneficial effects of the present invention at least include:

[0019] The present invention combines cyclic iterative prediction and regression model algorithms, which can retain the optimal solution predicted by each model in the next iteration process, and add the implicit relationship between different material properties to the material property prediction process, so as to achieve the most accurate solution to the optimal solution; in addition, the K-fold cross-validation algorithm is used to optimize the model hyperparameters, improve the generalization ability of the machine learning algorithm model, and more accurately construct the mapping relationship between the composition, process and alloy performance of the metal-ceramic composite material, so as to achieve efficient and accurate prediction of the composition of the metal-ceramic composite material under high performance requirements, which helps to solve the problems of long cycle, low accuracy and high cost of the existing multi-phase bonding phase metal-ceramic composite material composition design method, thereby promoting the engineering application process of new metal-ceramic composite materials.

[0020] The above description is only an overview of the technical solution of the present invention. In order to enable those skilled in the art to more clearly understand the technical means of the present application and implement them according to the contents of the specification, the following is a description of the preferred embodiments of the present invention in conjunction with detailed drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic flow chart of a composition design method of a multi-component bonding phase metal-ceramic composite material provided by a typical implementation case of the present invention;

[0022] Figure 2a It is a hardness determination coefficient distribution diagram of a composition design method of a multi-component bonding phase metal-ceramic composite material provided by a typical implementation case of the present invention;

[0023] Figure 2b It is a distribution diagram of the fracture toughness determination coefficient of a composition design method of a multi-component bonding phase metal-ceramic composite material provided by a typical implementation case of the present invention;

[0024] Figure 3aIt is a schematic diagram of hardness prediction performance of a composition design method of a multi-component bonding phase metal-ceramic composite material provided by a typical implementation case of the present invention;

[0025] Figure 3b It is a schematic diagram of fracture toughness prediction performance of a composition design method for a multi-component bonding phase metal-ceramic composite material provided by a typical implementation case of the present invention. DETAILED DESCRIPTION

[0026] In view of the deficiencies in the prior art, the inventor of this case has proposed the technical solution of the present invention after long-term research and extensive practice. The technical solution, its implementation process and principle will be further explained as follows.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0028] Moreover, relational terms such as “first” and “second” are merely used to distinguish one component or method step from another with the same name, but do not necessarily require or imply any such actual relationship or order between these components or method steps.

[0029] With the continuous development of big data and artificial intelligence technologies, data-driven research and development of new materials has gradually become a new model to transform the traditional trial and error method. It can greatly improve the efficiency of new material research and development, quickly solve the problem of difficult precise matching of the composition and performance of metal-ceramic composite materials, and promote the engineering application of new materials.

[0030] The present invention develops a new data-driven method for the composition design of complex metal-ceramic composite materials based on a cyclic prediction strategy, so as to unlock the complex implicit relationship between various alloying elements and performance, realize accurate prediction of the performance of metal-ceramic composite materials under high performance requirements and efficient composition design, and solve the problems of long design cycle, low efficiency and high cost of traditional metal-ceramic composite materials.

[0031] See also Figure 1 As shown, an embodiment of the present invention provides a method for designing a composition of a multi-component bonding phase metal-ceramic composite material, which comprises the following steps:

[0032] Constructing a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material;

[0033] The data set is divided into a training subset and a test subset by using a K-fold cross-validation method, multiple regression models are trained using the training subset to obtain multiple initial models, and the initial models are evaluated and optimized using the test subset to obtain multiple improved models;

[0034] Constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting the first prediction data into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved models in a loop to obtain final prediction data;

[0035] Based on the final prediction data, design parameters of the target multi-component bonding phase metal-ceramic composite material are determined.

[0036] By adopting the above-mentioned technical scheme, the main purpose of the present invention is to provide a method for designing the components of multi-phase bonding phase metal-ceramic composite materials based on machine learning. Compared with the traditional trial and error method and simple machine learning method, it can solve the problems of long design cycle, low accuracy and high cost of the components of multi-phase bonding phase metal-ceramic composite materials under high performance requirements.

[0037] The purpose of cyclic prediction is to make predictions again based on existing prediction results. The second prediction does not require experimental measurement but uses the results of the first prediction. The purpose of this is to avoid conducting multiple experiments and predict multiple groups of performance based on preset components.

[0038] The present invention first obtains data on material composition, process, performance, etc. through material databases, public literature and experiments, and establishes a high-quality sample database; on this basis, a machine learning method is used to mine the complex implicit relationship between various alloying elements and mechanical properties through data, and an alloy composition-performance machine learning model is constructed; then the preset multi-element bonding phase metal-ceramic composite material composition is input into the model to obtain its corresponding mechanical properties, thereby quickly and accurately determining the optimal alloy composition.

[0039] In some embodiments, in addition to the above steps, the method for designing the composition of a multi-component bonding phase metal-ceramic composite material may further include the following steps:

[0040] Prepare a sample of the target multi-component bonding phase metal-ceramic composite material based on the design parameters, and perform a test to obtain test data;

[0041] When the degree of difference between the test data and the final prediction data is higher than a preset threshold, the test data and its corresponding design parameters are supplemented as new data in the data set, and the data set is updated to perform a cyclic training and prediction process until the degree of difference between the test data and the final prediction data is below the preset threshold.

[0042] In some embodiments, the process of constructing a data set may specifically include:

[0043] Get the original data;

[0044] Calculating chemical descriptors and thermodynamic descriptors based on compositional parameters and process conditions in the raw data;

[0045] The composition parameters, structural parameters, process parameters, chemical descriptors, thermodynamic descriptors and mechanical property data of each component of the known material are used to establish an original data set;

[0046] Performing data preprocessing on each feature in the original data set respectively;

[0047] The preprocessed data set is subjected to feature engineering screening to obtain an optimal feature set as the data set.

[0048] More specifically in some embodiments, the data preprocessing specifically includes:

[0049] Eliminate abnormal data based on statistical distribution and empirical thresholds;

[0050] and normalize and standardize the data;

[0051] And in some embodiments, the feature engineering screening specifically includes:

[0052] Variance filtering: set a variance threshold. If the feature variable is less than the variance threshold, the feature is removed.

[0053] Correlation coefficient screening: calculate the correlation coefficients between the features and between the features and the mechanical properties data. When the correlation coefficient is lower than the first threshold, the features and mechanical properties data in this column are eliminated. If the correlation coefficient between two features is greater than the second threshold, one of them is eliminated. The first threshold is less than the second threshold.

[0054] The optimal feature set with a preset number is obtained by using a recursive feature elimination method.

[0055] Regarding specific threshold settings, in some implementation schemes, the first threshold value ranges from 0 to 0.10, and the second threshold value ranges from 0.8 to 1.

[0056] In addition, regarding the process of evaluating and optimizing using the test subset, in some embodiments, the process of obtaining the improved model specifically includes:

[0057] Calculating the coefficient of determination and mean square error of the predicted values ​​for the test subset and the corresponding mechanical property data in the test subset;

[0058] When the determination coefficient is above the determination threshold and the mean square error is below the error threshold, the improved model optimization is completed;

[0059] When the determination coefficient is lower than the determination threshold and the mean square error is higher than the error threshold, the optimization of the improved model continues.

[0060] Specifically in some embodiments, the coefficient of determination and root mean square error are expressed as:

[0061]

[0062] Wherein, y is the true value in the test subset, i.e., the mechanical property data; is the predicted value; is the average value of the true value; n is the sample size of the test subset.

[0063] Regarding threshold setting, in some implementation schemes, the decision threshold value ranges from 0.8 to 1.0, and the variance threshold value ranges from 0 to 2000.

[0064] Regarding the types of target prediction data, in some embodiments, the mechanical property data include hardness and fracture toughness; the initial model and the improved model both contain a hardness sub-model and a fracture toughness sub-model, which are used to predict hardness values ​​and fracture toughness values, respectively.

[0065] As some typical examples of the above technical solutions, in order to achieve the above purpose, the composition design method of multi-component bonding phase metal-ceramic composite material based on machine learning provided by the present invention includes the following steps:

[0066] Step 1: Determine the composition selection criteria for metal-ceramic composite materials; select parameters that affect material properties.

[0067] In the specific example, the specific implementation method of step one is as follows:

[0068] Select the components that affect the material properties, including cobalt, nickel, aluminum, silicon, titanium, chromium, molybdenum, vanadium, iron, manganese, copper, ruthenium, rhenium, niobium and other elements;

[0069] Select process parameters that affect material properties, including sintering time, sintering temperature, sintering method, sintering pressure and other parameters;

[0070] Select the microstructural parameters that affect the performance of metal-ceramic composites, including the average grain size.

[0071] Step 2: Acquire the composition data of metal-ceramic composite materials and the corresponding microstructure, process parameters, chemical descriptors, thermodynamic descriptors and target performance data through material databases, public literature and experimental calculations to establish a data set; wherein the data set is divided into a training subset and a test subset;

[0072] In the specific example, the specific implementation method of step 2 is as follows:

[0073] Establish a data set of material composition, structure and process type based on existing research on metal-ceramic composite materials;

[0074] According to the mass percentage of the components of the selected metal-ceramic composites, normalization was performed and the chemical descriptors that may affect the strength and toughness of the alloy were calculated using a python program;

[0075] According to the mass percentage of the components of the selected metal-ceramic composite materials, normalization was performed and the thermodynamic descriptors that may affect the strength and toughness of the alloy were calculated using pandat software;

[0076] A data set is established by using the mass fraction of each component of the known material, microstructural parameters, process parameters, chemical descriptors, thermodynamic descriptors and target performance to provide data support for the next step of data preprocessing and machine learning.

[0077] Step 3: Using the training subset to perform model training, respectively establish different performance machine learning models of metal-ceramic composite material composition-performance, and verify their accuracy;

[0078] The specific implementation method of step three is as follows:

[0079] The mass fraction of each component of the material, microstructure parameters, process parameters, chemical descriptors, thermodynamic descriptors and target performance data obtained in step 2 are respectively preprocessed;

[0080] Perform feature engineering screening on the preprocessed data set to obtain the optimal feature set;

[0081] Furthermore, the data preprocessing methods include: eliminating data with large deviations according to data distribution and experience; normalization processing; standardization processing. The methods for obtaining the optimal feature set include: variance filtering, setting a threshold, if the feature variable is less than the threshold, then eliminate the feature; correlation coefficient screening, calculate the correlation coefficient between each feature and between the feature and the target variable, including the Pearson correlation coefficient and the Spearman correlation coefficient, if the correlation coefficient between the feature and the variable is less than 0.05, then eliminate the feature variable in this column, if the correlation coefficient between two features is greater than 0.9, then eliminate one feature from the two; use recursive feature elimination to obtain a feature set with a certain number.

[0082] Divide the data set into training subset and test subset. Use the training subset for model training, and the test subset for verifying and optimizing the final model.

[0083] The material target performance parameters are used as model outputs, and other parameters are used as model inputs. A variety of regression model algorithms are selected for training to obtain an initial model.

[0084] Furthermore, the regression model algorithm adopts any one of random forest tree regression, gradient boosting regression tree, support vector regression, and back propagation neural network;

[0085] The initial model is optimized by using a K-Fold Cross Validation algorithm to obtain an improved model;

[0086] Using the test subset to evaluate the prediction effect and optimize the design of the improved model, calculate the performance evaluation index of the machine learning model, and use the best machine learning model as the candidate model;

[0087] Further, the improved model for predictive effect evaluation and optimization design includes:

[0088] Using the experimental and predicted values ​​of the alloy's mechanical properties, the coefficient of determination R is calculated. 2 and mean square error MSE;

[0089] Determination coefficient R 2 Is it greater than or equal to the preset threshold, and is the mean square error MSE less than or equal to the preset threshold:

[0090] When the coefficient of determination R 2 >=preset threshold and mean square error MSE<=preset threshold, the model optimization design is completed;

[0091] When the coefficient of determination R 2 <=preset threshold or mean square error MSE>=preset threshold, repeat the optimization design steps until the determination coefficient R2>=preset threshold and the mean square error MSE<=preset threshold.

[0092] Furthermore, the best machine learning prediction model for multiple target performances can be obtained through step three.

[0093] Step 4: Input the target system metal-ceramic composite material composition, process and other characteristic parameters into the machine learning model to obtain the corresponding performance. Through the cyclic prediction strategy, obtain multiple sets of corresponding performance prediction values, and determine the optimal alloy composition based on the material with the best comprehensive performance.

[0094] As a specific example, the specific implementation method of step 4 is as follows:

[0095] Determine the target system metal-ceramic composite material, set the limit range of each component, and construct the system material composition space with a certain step size;

[0096] Taking the best feature combination obtained above as the standard, calculate each feature variable in the component space to determine the prediction data set;

[0097] Inputting the obtained prediction data set into a determined machine learning model to obtain first performance data of different component combinations;

[0098] Further, the obtained first performance data is added to the prediction data set, and the updated prediction data set is input into the prediction model of the second performance to obtain the prediction data of the second performance, and this cycle is repeated to achieve the purpose of cyclic iterative prediction.

[0099] Step 5: Prepare samples using the data on the optimal composition and content of the metal-ceramic composite material in step 4, and conduct experimental verification of relevant mechanical properties.

[0100] As a specific example, the specific implementation method of step five is as follows:

[0101] Select the material with the best comprehensive performance in the predicted results for preparation and testing to obtain the actual performance parameters;

[0102] Determine whether the error between the predicted performance and the actual performance is greater than 15%; if so, the design is successful; if not, the data of this material is added to the original data set, and the above steps are repeated until a metal-ceramic composite material that meets the design requirements is obtained.

[0103] The design process of a composite material with WC ceramic particles as a hard phase and multiple metal elements as a bonding phase is demonstrated as a typical representative case below. However, it should be noted that although the present invention uses a carbide system as a representative case, it cannot be considered that the design method provided by the present invention is only applicable to this material system. In fact, other hard ceramic-transition metal element hard phase-bonding phase systems with similar properties are also similarly applicable to the design method and ideas provided by the present invention.

[0104] First, the composition and performance data of metal-ceramic composites were obtained through material databases, public literature and experiments to establish a data set. The composition of the metal-ceramic composite material is WC as the hard phase, and the multi-element alloying element is used as the bonding phase. The multi-element alloying element is at least one of cobalt, nickel, aluminum, silicon, titanium, chromium, molybdenum, vanadium, iron, manganese, copper, ruthenium, rhenium and niobium.

[0105] It is worth mentioning that the metal-ceramic composite material in the present invention uses WC as the hard phase, and the WC content has a great influence on the mechanical properties of the alloy. Considering the relationship between material structure and performance, the WC mass fraction percentage range in the data set is 70-96.1wt.%.

[0106] In addition, since it is difficult to sinter and densify metal-ceramic composite powders, the publicly reported high-performance sample data is mainly obtained through rapid sintering methods such as spark plasma sintering, microwave sintering, and high-frequency induction heating sintering. However, due to its complex process and high unit cost, and the high level of participating porosity and anisotropy of the sintered samples, the current data volume is relatively limited. In order to improve the accuracy of the model, the alloy samples in the data set of the present invention are mainly obtained by traditional vacuum sintering methods to eliminate the influence of factors such as porosity.

[0107] According to the mass percentage of the elements and components of the bonding phase of the selected metal-ceramic composite material, normalization is performed, and the chemical descriptors that may affect the strength and toughness of the alloy are calculated using the python program; the thermodynamic descriptors that may affect the strength and toughness of the alloy are calculated using the pandat software;

[0108] Specifically, the initial characteristic variables include: mass fraction of WC, mass fraction of Co, mass fraction of Ni, mass fraction of Al, mass fraction of Si, mass fraction of Ti, mass fraction of Cr, mass fraction of Mo, mass fraction of V, mass fraction of Fe, mass fraction of Mn, mass fraction of Cu, mass fraction of Ru, mass fraction of Re, mass fraction of Nb, mass fraction of grain refiner, average grain size of WC, sintering temperature, sintering time, sintering pressure, average atomic number, atomic number difference, average atomic radius, atomic radius difference, average Pauli electronegativity, Pauli electronegativity difference, average valence electron concentration, valence electron concentration difference, average electron affinity, electron affinity difference, average ionization energy, ionization energy difference, average melting point, melting point difference, average density, density difference, Gibbs free energy, reaction enthalpy, mixing entropy, hot melt, molar volume. The mechanical properties of metal-ceramic composites include hardness and fracture toughness.

[0109] The methods for preprocessing data are: removing data with large deviations according to data distribution and experience; normalization; standardization. The methods for obtaining the optimal feature set are: variance filtering, setting a threshold, if the feature variable is less than the threshold, then remove the feature; correlation coefficient screening, calculate the correlation coefficient between each feature and between the feature and the target variable, including the Pearson correlation coefficient and the Spearman correlation coefficient, if the correlation coefficient between the feature and the variable is less than 0.05, then remove the feature variable in this column, if the correlation coefficient between two features is greater than 0.9, then remove one feature from the two; use recursive feature elimination to obtain a feature set with a certain number.

[0110] In the embodiment of the present invention, the data set is divided into a training subset and a test subset, the data volume in the training subset accounts for 70% to 80% of the total data volume of the data set, the data volume in the test subset accounts for 20% to 30% of the total data volume of the data set, and the sum of the proportions of the training subset and the test subset is always 1. Using the obtained composition, process, and performance data of the metal-ceramic composite material, in the process of establishing the data set, the system will randomly and automatically generate a training subset and a test subset, wherein both the training subset and the test subset contain the composition process of the metal-ceramic composite material and the corresponding alloy mechanical property data. In order to facilitate distinguishing and managing, the data in the training subset is called training data, which is used for model establishment; the data in the test subset is called test data, which is used for effect evaluation and optimization of the established model.

[0111] In order to improve the accuracy and applicability of the model, machine learning models with different mechanical properties are trained using the same data set. The machine learning algorithm in the present invention is mainly one or more of support vector regression, random forest regression, gradient boosting decision tree regression, and back propagation neural network suitable for small-scale data sets. Among them, the grid optimization of hyperparameters is performed corresponding to the machine learning model algorithm, and the hyperparameter setting can be adjusted as needed without specific limitation.

[0112] The K-Fold Cross Validation algorithm is used to optimize the hyperparameters of the initial model to obtain an improved model; in the example of the present invention, the K value in the K-Fold Cross Validation algorithm is 5 or 10. It should be noted that the K value can be 5 or 10, and can also be designed according to actual requirements, without specific requirements.

[0113] The test subset is used to evaluate the prediction effect and optimize the design of the improved model, improve the model accuracy and generalization ability, and obtain the best model. In the embodiment of the present invention, the evaluation indicators used are the determination coefficient R2 and the mean square percentage error MSE, and the expression is as follows:

[0114]

[0115] Among them, y is the true value; is the predicted value; is the average value of the true value; n is the sample size of the test subset.

[0116] Model evaluation and optimization design specifically include the following steps:

[0117] Input the characteristic parameters in the test subset into the improved model to obtain the corresponding predicted values;

[0118] Using the experimental and predicted values ​​of the alloy properties, the coefficient of determination R is calculated. 2 and mean square error MSE;

[0119] Determination coefficient R 2 Is it greater than or equal to the preset threshold, and is the mean square error MSE less than or equal to the preset threshold:

[0120] When the coefficient of determination R 2 ≥ preset threshold and mean square error MSE ≤ preset threshold, the model optimization design is completed;

[0121] When the coefficient of determination R 2 <Preset threshold or mean square error MSE> Preset threshold, repeat the optimization design steps until the determination coefficient R 2 ≥ preset threshold and mean square error MSE ≤ preset threshold.

[0122] Where R 2 The preset threshold can be 0.9 and the MSE can be 1000. Stricter error limits are conducive to improving the generalization ability of the model.

[0123] Input multiple sets of preset composite material components and processes into the optimal model to obtain the corresponding multiple sets of alloy performance prediction data. Multiple sets of alloy input parameters can be pre-set and input into the obtained optimal model to obtain the corresponding multiple sets of alloy different performance prediction data.

[0124] Correspondingly, the second aspect of the embodiment of the present invention further provides a multi-component bonding phase metal-ceramic composite material composition design system, which includes:

[0125] A data construction module, used to construct a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material;

[0126] A model training module is used to divide the data set into a training subset and a test subset by using a K-fold cross-validation method, train multiple regression models using the training subset to obtain multiple initial models, and evaluate and optimize the initial models using the test subset to obtain multiple improved models;

[0127] A cyclic prediction module, used for constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting it into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved model for cyclic execution to obtain final prediction data;

[0128] The parameter determination module is used to determine the design parameters of the target multi-component bonding phase metal-ceramic composite material based on the final prediction data.

[0129] The technical solution of the present invention is further described in detail below through several embodiments and in conjunction with the accompanying drawings. However, the selected embodiments are only used to illustrate the present invention, and do not limit the scope of the present invention.

[0130] Example 1

[0131] This embodiment illustrates a composition design process of a multi-component bonding phase metal-ceramic composite material of a WC-transition metal system, as shown below:

[0132] Establishing the initial data set: Obtaining the data on the composition, process, and performance of WC-based multi-component bonding phase metal-ceramic composites from the material database, public literature, and experimental test data to establish the initial data set; the hard phase is WC, and the bonding phase element is

[0133] Data processing: preprocess the data in the initial data set, remove data with large deviations according to the data distribution; normalize the data; use the correlation coefficient screening method to calculate the correlation coefficients between each feature and between feature variables, and remove features with correlations less than 0.05 and greater than 0.9; obtain the processed data set.

[0134] Establish the initial model of mechanical properties of WC-based metal-ceramic composite materials: divide the data set into training subset and test subset, the data volume in the training subset accounts for 80% of the total data volume of the data set, and the data volume in the test subset accounts for 20% of the total data volume of the data set; use the BPNN back-propagation neural network algorithm to train the model, and use the K-Fold Cross Validation algorithm to optimize the hyperparameters of the BPNN model. Among them, K=5, the optimization range of the BPNN algorithm hyperparameters hidden layers and hidden size is 1~10 and 10N100, the learning rate is 0.01, the batch size is 12, and the numepochs is 1000. The determination coefficients R2 of the final optimized model are 0.913 and 0.906, respectively, as shown in Figure 2.

[0135] An orthogonal design was performed on the Fe and Ni components in the Co, Fe, Ni ternary bonding phases of the WC-based metal-ceramic composite material to obtain multiple sets of preset component combinations. The multiple sets of preset material compositions were input into the optimal hardness model to obtain the hardness performance values ​​corresponding to the preset alloy compositions. The obtained hardness data were input into the optimal fracture toughness model to obtain the fracture toughness performance value of the material. The performance diagram predicted by the model is shown in Figure 3.

[0136] The optimal composition of the composite material is determined based on multiple sets of preset component combinations and their corresponding hardness and fracture toughness prediction data.

[0137] The WC-CoFeNi composite material samples were prepared by vacuum sintering technology using the above optimal composition. The specific process parameters were: sintering temperature 1450°C, sintering holding time 60 min, and vacuum sintering. The hardness of the obtained samples was tested, and the fracture toughness value was calculated based on the indentation crack.

[0138] The error percentage between the predicted value and the true value is compared, and the experimental data is added to the established data set. The prediction accuracy and generalization ability of the model are continuously improved through iterative feedback of experimental data.

[0139] Table 1 Composition design of WC-based metal-ceramic composites and corresponding predicted data and experimental data of hardness and fracture toughness

[0140]

[0141] It can be seen from Table 1 that for the WC-based multi-component bonding phase metal-ceramic composite material, the hardness and fracture toughness of the metal-ceramic composite material predicted by the machine learning model proposed in the present invention are 6.07% and 3.74% respectively, which are both less than 10%. Therefore, the accuracy of the obtained performance prediction model meets the expected use requirements, and can realize the rapid and accurate prediction of the mechanical properties of the multi-component bonding phase metal-ceramic composite material, thereby accurately and efficiently determining the optimal material composition.

[0142] Based on the above implementation cases, it can be clearly understood that the design method provided in the embodiment of the present invention relates to the field of metal-ceramic composite materials made by powder metallurgy process using metal carbide as hard phase, transition metal element or rare earth element as bonding phase. At present, by mining a large amount of data on the composition, process and performance of metal-ceramic composite materials in existing literature, atomic descriptors and thermodynamic descriptors are calculated and established. Based on these data, machine learning methods are used to explore the complex relationship between "composition-performance" of metal-ceramic composite materials, build an initial model, and optimize the initial model through K-fold Cross Validation technology, and use the test subset to evaluate the prediction effect of the improved model to obtain the best model; multiple groups of preset material components are input into the best model to obtain corresponding prediction performance; a strategy of multiple cyclic iterations is adopted to add the existing prediction results to the data set, and different properties of the material are predicted as the target variable of machine learning, so as to optimize the design of material components. Compared with the traditional trial and error method, the design method in the present invention can significantly shorten the composition design cycle of multi-component bonding phase metal-ceramic composite materials, improve design accuracy, and reduce R&D costs.

[0143] It should be understood that the above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for designing the composition of a multi-component bonding phase metal-ceramic composite material, characterized in that: include: Constructing a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material; The data set is divided into a training subset and a test subset by using a K-fold cross-validation method, multiple regression models are trained using the training subset to obtain multiple initial models, and the initial models are evaluated and optimized using the test subset to obtain multiple improved models; Constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting the first prediction data into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved models in a loop to obtain final prediction data; Based on the final prediction data, design parameters of the target multi-component bonding phase metal-ceramic composite material are determined.

2. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 1, characterized in that: Also includes: Prepare a sample of the target multi-component bonding phase metal-ceramic composite material based on the design parameters, and perform a test to obtain test data; When the degree of difference between the test data and the final prediction data is higher than a preset threshold, the test data and its corresponding design parameters are supplemented as new data in the data set, and the data set is updated to perform a cyclic training-prediction process until the degree of difference between the test data and the final prediction data is below the preset threshold.

3. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 1, characterized in that: The process of building a dataset specifically includes: Get the original data; Calculating chemical descriptors and thermodynamic descriptors based on compositional parameters and process conditions in the raw data; The composition parameters, structural parameters, process parameters, chemical descriptors, thermodynamic descriptors and mechanical property data of each component of the known material are used to establish an original data set; Performing data preprocessing on each feature in the original data set respectively; The preprocessed data set is subjected to feature engineering screening to obtain an optimal feature set as the data set.

4. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 3, characterized in that: The data preprocessing specifically includes: Eliminate abnormal data based on statistical distribution and empirical thresholds; and normalize and standardize the data; And / or, the feature engineering screening specifically includes: Variance filtering: set a variance threshold. If the feature variable is less than the variance threshold, the feature is removed. Correlation coefficient screening: calculate the correlation coefficients between the features and between the features and the mechanical properties data. When the correlation coefficient is lower than the first threshold, the features and mechanical properties data in this column are eliminated. If the correlation coefficient between two features is greater than the second threshold, one of them is eliminated. The first threshold is less than the second threshold. The optimal feature set with a preset number is obtained by using a recursive feature elimination method.

5. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 4, characterized in that: The first threshold value ranges from 0 to 0.10, and the second threshold value ranges from 0.8 to 1.

6. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 1, characterized in that: The process of obtaining the improved model specifically includes: Calculating the coefficient of determination and mean square error of the predicted values ​​for the test subset and the corresponding mechanical property data in the test subset; When the determination coefficient is above the determination threshold and the mean square error is below the error threshold, the improved model optimization is completed; When the determination coefficient is lower than the determination threshold and the mean square error is higher than the error threshold, the optimization of the improved model continues.

7. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 6, characterized in that: The coefficient of determination and the root mean square error are expressed as: Wherein, y is the true value in the test subset, i.e., the mechanical property data; is the predicted value; is the average value of the true value; n is the sample size of the test subset.

8. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 6, characterized in that: The decision threshold has a value range of 0.8-1.0, and the variance threshold has a value range of 0-2000.

9. The composition design method of multi-component bonding phase metal-ceramic composite material according to claim 1, characterized in that: The mechanical properties data include hardness and fracture toughness; The initial model and the improved model both include a hardness sub-model and a fracture toughness sub-model, which are used to predict hardness values ​​and fracture toughness values ​​respectively.

10. A multi-component bonding phase metal-ceramic composite material composition design system, characterized in that: include: A data construction module, used to construct a data set, wherein the data set includes composition parameters, structural parameters, process conditions and mechanical property data of a known multi-component bonding phase metal-ceramic composite material; A model training module is used to divide the data set into a training subset and a test subset by using a K-fold cross-validation method, train multiple regression models using the training subset to obtain multiple initial models, and evaluate and optimize the initial models using the test subset to obtain multiple improved models; A cyclic prediction module, used for constructing a prediction parameter set of a target multi-element bonding phase metal-ceramic composite material, inputting the prediction parameter set into one of the improved models to obtain first prediction data, merging the first prediction data into the prediction parameter set and inputting it into another improved model as a new prediction parameter set to obtain second prediction data, and traversing the improved model for cyclic execution to obtain final prediction data; The parameter determination module is used to determine the design parameters of the target multi-component bonding phase metal-ceramic composite material based on the final prediction data.

Images