Directional design method and device for high-entropy ceramics

Abstract

The invention discloses a directional design method and device for high-entropy ceramics in the technical field of new material design and development, and the method comprises the steps: obtaining initial sample data of a high-entropy ceramic material, and constructing an initial sample data set; calculating input feature data based on the initial sample data set; performing feature screening on the input feature data, and constructing a training set and a test set based on the screened features and the corresponding mechanical property parameters of the high-entropy ceramic material; constructing a hardness prediction model based on the training set and the test set; and performing component optimization on the hardness prediction model to obtain a target composition element solution set, and determining target composition element components of the high-hardness and high-entropy ceramic according to the target composition element solution set as a design result. According to the directional design method and device for the high-entropy ceramic, the influence of the component inherent property class characteristics and the component element class characteristics on the performance of the high-entropy ceramic is considered, and the problem that a model is difficult to reverse only based on the component inherent property class characteristics is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of new material design and development, and in particular to a method and device for directional design of high entropy ceramics. Background Art

[0002] As an emerging class of multi-component ceramic materials, high entropy ceramics break through the limitations of traditional material design by utilizing the high entropy effect to stabilize complex structures. Among the many high entropy ceramic systems, high entropy carbides, high entropy nitrides and high entropy carbonitride ceramics have become a hot topic of research due to their excellent hardness, high thermal stability and good chemical inertness, and have shown great potential in application fields such as cutting tools and protective coatings. However, traditional material design methods are difficult to meet the design requirements of high entropy ceramics, which greatly limits their development speed.

[0003] In 2020, Kaufmann from the Vecchio team at the University of California, USA, applied machine learning methods to the field of high-entropy ceramics for the first time, successfully predicted the single-phase formation ability of high-entropy carbide ceramics, and opened a new chapter in the interdisciplinary research of machine learning and materials science. On this basis, many researchers have further expanded machine learning methods to the prediction of high-entropy ceramic properties, and successfully achieved the prediction of mechanical properties such as hardness and elastic modulus, which has accelerated the development of high-entropy ceramics to a certain extent. However, current research is still in the prediction stage, and no scholar has yet proposed an efficient directional design method for high-entropy ceramics.

[0004] Existing technologies mainly guide the design of new high-entropy ceramics by constructing and predicting virtual space and screening virtual samples that meet target performance. However, this method has poor adaptability and is difficult to efficiently guide the directional design of new high-entropy ceramics.

[0005] Therefore, it is urgent to design a method and device that can quickly and accurately guide the directional design of high-hardness and high-entropy ceramics. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and device for the directional design of high entropy ceramics. The method comprehensively considers the influence of the inherent property characteristics of the components and the characteristics of the constituent elements on the performance of high entropy ceramics, avoids the large amount of data sets required to construct a model using only the characteristics of the constituent elements, and overcomes the difficulty of inverse engineering when constructing a model based only on the inherent property characteristics of the components.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] In one aspect, the present invention provides a method for directional design of high entropy ceramics, comprising:

[0009] Obtaining initial sample data of high entropy ceramic materials and constructing an initial sample data set;

[0010] Calculate input feature data based on the initial sample data set;

[0011] Performing feature screening on the input feature data to obtain screened features, and constructing a training set and a test set based on the screened features and corresponding mechanical property parameters of high entropy ceramic materials;

[0012] Building a hardness prediction model based on the training set and the test set;

[0013] The hardness prediction model is subjected to composition optimization to obtain a target component element solution set, and the target component element composition of the high-hardness high-entropy ceramic is determined according to the target component element solution set as a directional design result of the high-entropy ceramic.

[0014] Furthermore, the initial sample data includes mechanical property parameters, component intrinsic property parameters, component element parameters, preparation method parameters and preparation process parameters of the high entropy ceramic material;

[0015] Wherein, the mechanical property parameters of the high entropy ceramic material include one or more of the Vickers hardness, Rockwell hardness, Brinell hardness and nanoindentation hardness of the high entropy ceramic material;

[0016] And / or, the obtaining of initial sample data of high-entropy ceramic materials and the construction of an initial sample data set include: obtaining material data of bulk high-entropy ceramic materials from an SCI literature library, wherein the bulk high-entropy ceramic materials include high-entropy ceramic systems of three different anions: high-entropy carbides, high-entropy nitrides and high-entropy carbonitrides.

[0017] Furthermore, the input characteristic data includes the inherent property characteristics of the components, the component composition characteristics and the preparation process characteristics;

[0018] The step of calculating input feature data based on the initial sample data set includes:

[0019] Obtaining the intrinsic property characteristics of the component based on the intrinsic property parameters of the component;

[0020] Obtain component composition characteristics Com and carbon-nitrogen ratio C / N based on component element parameters;

[0021] The preparation process characteristics Syn are obtained based on the preparation method parameters and the preparation process parameters.

[0022] Furthermore, the intrinsic property characteristics of the components are obtained based on the intrinsic property parameters of the components, wherein the intrinsic property characteristics of the components include the average melting temperature, melting temperature deviation, average atomic radius, atomic radius deviation, average density, density deviation, average mass, mass deviation, average electron affinity, electron affinity deviation, average Pauling electronegativity, Pauling electronegativity deviation, average ionization energy and fractional weighted average of the ionization energy deviation of all constituent elements. and variance , the calculation method is:

[0023]

[0024] in, It is the first of all the components The fractional weighted average of the intrinsic properties, For all the components The variance of the intrinsic property, Indicates The first component An inherent property, Indicates The mole fraction of an element, is the number of constituent elements;

[0025] The intrinsic property characteristics of the components also include the average valence electron concentration of the corresponding precursor of the synthesized high entropy ceramic material. and valence electron concentration difference , the calculation method is:

[0026]

[0027]

[0028] in , , is the mole fraction of carbon, is the mole fraction of nitrogen, Indicates The mole fraction of an element, is the number of cations in the high entropy ceramic material, For the The valence electron concentration of each constituent element, and They represent the valence electron concentrations of carbon and nitrogen in the corresponding precursors for synthesizing high entropy ceramic materials;

[0029] And / or, the carbon-nitrogen ratio C / N has a value range of [0, 1], when the atomic percentage of nitrogen is 0, the carbon-nitrogen ratio is assigned a value of 1; when the atomic percentage of carbon is 0, the carbon-nitrogen ratio is assigned a value of 0;

[0030] And / or, the obtaining of component composition characteristics Com based on component element parameters includes:

[0031] Perform one-hot encoding on the constituent element parameters to obtain the binary component features;

[0032] The binary component features are reduced to 1, 3, 5, 7 and 9 dimensions respectively to obtain Com of different dimensions;

[0033] And / or, obtaining the preparation process characteristics Syn based on the preparation method parameters and the preparation process parameters includes:

[0034] Performing target encoding on the preparation method parameters to obtain encoded preparation method parameters;

[0035] Performing mode filling on the missing values ​​of the preparation process parameters to obtain the filled preparation process parameters;

[0036] The encoded preparation method parameters and the filled preparation process parameters are reduced to 1 dimension to obtain the preparation process characteristics, which are denoted as Syn.

[0037] Furthermore, the input feature data is subjected to feature screening to obtain screened features, including:

[0038] Calculate the Pearson correlation coefficient between the input feature data pairs, and select the input feature data pairs with the Pearson correlation coefficient greater than 0.7;

[0039] Calculate the Pearson correlation coefficient between two input feature data and the target feature in all input feature data pairs, delete the input feature data with the Pearson correlation coefficient between each pair of input feature data and the target feature, and obtain the input feature data for preliminary screening;

[0040] The genetic algorithm is used to train the model and the random forest maximization model evaluation criteria are As the objective function, the iterative process is performed to obtain the selected features, wherein the genetic algorithm is iterated 100-500 times, the population value is set to 50-100, and the crossover factor and mutation factor range is 0.1-0.5.

[0041] Furthermore, the hardness prediction model is constructed based on the training set and the test set, including:

[0042] Calling Support Vector Regression, RandomForest and eXtreme Gradient Boosting algorithms from the scikit-learn library in Python, and building Vickers hardness prediction models under each algorithm based on the training set and the test set;

[0043] The Scikit-learn library in Python is used to call GridSearchCV to perform hyperparameter tuning on the Vickers hardness prediction model under each algorithm, and the Vickers hardness prediction model under each algorithm after hyperparameter tuning is obtained;

[0044] Calculate the determination coefficient of the Vickers hardness prediction model under each algorithm after the hyperparameter tuning With the root mean square error RMSE, choose The algorithm with the largest value and the smallest RMSE is used as the final algorithm of the Vickers hardness prediction model under each algorithm after hyperparameter tuning, and the optimal machine learning algorithm is obtained;

[0045] Based on the preferred machine learning algorithm, Vickers hardness prediction models containing Com of different dimensions are constructed respectively;

[0046] Calculation of the coefficient of determination for the Vickers hardness prediction model involving different dimensions of Com ,choose The dimension Com greater than 0.8 is taken as the target dimension Com, and the Vickers hardness prediction model including the target dimension Com is obtained, which is the hardness prediction model finally constructed.

[0047] Furthermore, the coefficient of determination The calculation formula is:

[0048] ;

[0049] The root mean square error RMSE calculation formula is:

[0050] ;

[0051] in, is the total number of samples, It is The true hardness value of the samples, Representative The hardness prediction value of samples, Indicates the true hardness average value.

[0052] Furthermore, the hardness prediction model is optimized to obtain a target element solution set, and the target element composition of the high-hardness high-entropy ceramic is determined according to the target element solution set, including:

[0053] A genetic algorithm is used to perform target optimization on the Vickers hardness prediction model including the target dimension Com, so as to obtain a target solution set returned by the Vickers hardness prediction model including the target dimension Com;

[0054] According to the size of the target solution set returned by different models, the principal component analysis and dimensionality reduction reconstruction of the component composition characteristics Com are performed to restore the constituent elements of the target high entropy ceramics. The system with the largest number of restored target high entropy ceramics is selected, and the model Models with values ​​greater than or equal to 0.8 are used as hardness prediction models for directional design;

[0055] The Com in the target solution set of the hardness prediction model used for directional design is reconstructed to obtain the target element composition of high-hardness and high-entropy ceramics.

[0056] Furthermore, the genetic algorithm is used to perform target optimization on the Vickers hardness prediction model containing the target dimension Com, including calling the genetic algorithm in the DEAP library in python to perform target optimization on the Vickers hardness prediction model containing the target dimension Com, wherein the objective function is the predicted hardness value of the Vickers hardness prediction model, the constraint condition is that the predicted hardness value is greater than a preset target hardness value, the population size is set to 10-100, the number of iterations is set to 100-1000, and the crossover factor and the variation factor are set to 0.1-0.5.

[0057] On the other hand, the present invention provides a device for directional design of high entropy ceramics, which is used to implement the directional design method of high entropy ceramics, and the device comprises:

[0058] A data acquisition module is used to obtain initial sample data of high entropy ceramic materials and construct an initial sample data set;

[0059] A feature calculation module, used to calculate input feature data based on an initial sample data set;

[0060] A feature screening module performs feature screening on the input feature data to obtain screened features, and constructs a training set and a test set based on the screened features and corresponding mechanical property parameters of high entropy ceramic materials;

[0061] A model training module, used for building a hardness prediction model based on the training set and the test set;

[0062] The output optimization module optimizes the composition of the hardness prediction model to obtain a target component element solution set, and determines the target component element composition of the high-hardness high-entropy ceramic according to the target component element solution set as a directional design result of the high-entropy ceramic.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] The directional design method of high entropy ceramics provided by the present invention realizes accurate prediction and directional design of the Vickers hardness of high entropy ceramics by introducing machine learning algorithms and heuristic optimization algorithms, breaking the current situation of "can be predicted but difficult to reverse" in related research, and is expected to meet the requirements of on-demand design of high entropy ceramics and accelerate the research progress in the field of high entropy ceramics.

[0065] The present invention comprehensively considers the influence of the inherent property characteristics of the components and the characteristics of the constituent elements on the performance of high entropy ceramics, avoiding the large amount of data sets required to construct the model using only the characteristics of the constituent elements, and also overcomes the difficulty of inverse engineering when constructing the model based only on the inherent property characteristics of the components, and improves the model accuracy to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a schematic flow chart of a method for directional design of high entropy ceramics provided in an embodiment of the present invention;

[0067] Figure 2 It is a schematic diagram comparing the actual value and predicted value of the hardness of the one-dimensional Com corresponding model provided by an embodiment of the present invention;

[0068] Figure 3 It is a schematic diagram comparing the actual value and predicted value of the hardness of the five-dimensional Com corresponding model provided by an embodiment of the present invention;

[0069] Figure 4 is a SHAP feature importance ranking diagram of the hardness prediction model provided by an embodiment of the present invention;

[0070] Figure 5 is a SHAP summary diagram of the hardness prediction model provided by an embodiment of the present invention;

[0071] Figure 6 The embodiment of the present invention provides Corresponding ICE and PDP curves;

[0072] Figure 7 The embodiment of the present invention provides Corresponding ICE and PDP curves;

[0073] Figure 8 The embodiment of the present invention provides (Ta 0.25 Zr 0.25 Nb 0.25 Hf 0.25 )C and (Ti 0.25 Zr 0.25 Nb 0.25 Hf 0.25 ) The corresponding XRD pattern of C. DETAILED DESCRIPTION

[0074] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0075] Example 1

[0076] This embodiment provides a method for directional design of high entropy ceramics, the flow chart is as follows Figure 1 As shown, the method includes:

[0077] S1. Obtain the initial sample data set of high entropy carbide, nitride and carbonitride ceramic materials: each sample data includes the Vickers hardness performance parameters, component intrinsic property parameters, component element parameters, preparation method parameters and preparation process parameters of the high entropy ceramic materials. In addition, the missing values ​​in the above parameters are filled with the mode, the preparation method parameters and preparation process parameters are further reduced in dimension, and the entire data set is standardized;

[0078] S2. Feature screening to obtain training and test sets: Feature screening is performed based on the Pearson correlation coefficient and genetic algorithm, and the training set and test set are constructed in a ratio of 7:3;

[0079] S3. Based on the input feature data and the corresponding Vickers hardness performance parameters, a machine learning model is constructed using a variety of algorithms, and the model is validated through five-fold cross validation. The optimal model was selected by comparing with RMSE. On this basis, the Shapley additive feature interpretation method (SHAP), individual conditional expectation diagram (ICE) and partial dependence diagram (PDP) methods were used to analyze the interpretability of the model.

[0080] S4. Optimize the constructed Vickers hardness prediction model based on genetic algorithm to obtain the target solution set; on this basis, use principal component analysis and dimensionality reduction reconstruction to restore the component-related features in the target solution set and use it as the directional design result of high-hardness and high-entropy ceramics;

[0081] S5. Based on the above directional design results, the target high entropy ceramics were prepared by spark plasma sintering method in argon atmosphere with a pressure of 30 MPa maintained for 15 minutes, and the phase formation was analyzed by XRD method; the Vickers hardness was measured by DHN-1000Z microhardness tester under a load of 100-1000 gf for 5 seconds, repeated 3 times.

[0082] In this embodiment, obtaining an initial sample data set of high entropy ceramic materials includes:

[0083] The material data of high entropy carbide, nitride and carbonitride ceramic samples were obtained from the SCI literature library to form the initial sample data set.

[0084] The initial sample data set directly affects the performance and generalization of the constructed model. To ensure the reliability of the data, the present invention uses "high entropy ceramics" and "mechanical properties" as keywords to obtain the data from published high-level literature.

[0085] It should be noted that the sample data of high-entropy ceramic materials obtained include at least 10 different systems of high-entropy ceramics.

[0086] The indicators closely related to the mechanical properties of high entropy ceramics mainly include Vickers hardness, Rockwell hardness, Brinell hardness and nanoindentation hardness. In this embodiment, Vickers hardness is selected as a parameter indicator for measuring hardness performance.

[0087] The intrinsic properties of high entropy ceramic materials include the average valence electron concentration ( )、valence electron concentration difference( )、Average melting temperature( )、Melting temperature difference( ), average atomic radius ( ), atomic radius difference ( ), average electron affinity ( )、Electron affinity difference( )、Average ionization energy( )、Ionization energy difference( ), average Pauling electronegativity ( ), Pauling electronegativity is poor ( )、Average density( ), density difference ( ), average quality ( ) and poor quality ( ), the component composition characteristics include the component element parameters after one-hot encoding, the input feature data also includes the carbon-nitrogen ratio (C / N), the value range of the carbon-nitrogen ratio C / N is [0, 1], when the atomic percentage of nitrogen element is 0, the carbon-nitrogen ratio is assigned a value of 1; when the atomic percentage of carbon element is 0, the carbon-nitrogen ratio is assigned a value of 0, the preparation method parameters include the sintering method, and the preparation process parameters include the sintering time, sintering pressure and sintering temperature.

[0088] In one embodiment of the present invention, input feature data is obtained based on the intrinsic property parameters of the components, the constituent element parameters, the preparation method parameters and the preparation process parameters in the initial sample data set. The specific implementation process is as follows:

[0089] The calculation method for obtaining the intrinsic property characteristics of the component based on the intrinsic property parameters of the component is:

[0090]

[0091]

[0092] in, It is the first of all the components The fractional weighted average of the intrinsic properties, For all the components The variance of the intrinsic property, Indicates The first component of An inherent property, Indicates The mole fraction of an element, is the number of constituent elements;

[0093] The intrinsic property characteristics of the components also include the average valence electron concentration of the corresponding precursor of the synthesized high entropy ceramic material. and valence electron concentration difference , the calculation method is:

[0094]

[0095]

[0096] in , , is the mole fraction of carbon, is the mole fraction of nitrogen, Indicates The mole fraction of an element, is the number of cations in the high entropy ceramic material, For the The valence electron concentration of each constituent element, and They represent the valence electron concentrations of carbon and nitrogen in the corresponding precursors for synthesizing high entropy ceramic materials;

[0097] Get the characteristic Com method based on the component element parameters:

[0098] Use one-hot encoding to binarize the constituent element parameters, and perform principal component analysis to reduce them to 1, 3, 5, 7, and 9 dimensions. After the model is established, select The Com dimension greater than 0.8 is taken as the target dimension Com, and the Com dimension is determined.

[0099] Acquisition of characteristic Syn methods based on preparation method parameters and preparation process parameters:

[0100] The principal component analysis method was used to reduce the sintering method, sintering time, sintering pressure and sintering temperature to one dimension and named Syn to prevent the "dimensional disaster".

[0101] In this embodiment, machine learning is performed based on the above input feature data and the corresponding hardness performance parameters to construct a hardness prediction model. The specific implementation process is as follows:

[0102] Call XGBoost, RF and SVR algorithms from the scikit-learn library in Python, use the called algorithms for machine learning based on input feature data and hardness performance parameters, and build hardness prediction models under each algorithm;

[0103] Calculate the determination coefficient of the Vickers hardness prediction model under each algorithm after the hyperparameter tuning With the root mean square error RMSE, choose The algorithm with the largest value and the smallest RMSE is used as the final algorithm of the Vickers hardness prediction model under each algorithm after hyperparameter tuning, and the optimal machine learning algorithm is obtained.

[0104] In this embodiment, it is finally determined to build a hardness prediction model based on the XGBoost algorithm.

[0105] 70% of the data set is randomly selected as the training set, and the other 20% is used as the test set and does not participate in the model training process. After the model training is completed, the test set is used to verify whether the model is accurate.

[0106] Coefficient of determination The calculation formula is: .

[0107] The root mean square error RMSE calculation formula is: .

[0108] in, is the total number of samples, It is The true hardness value of the samples, Representative The hardness prediction value of samples, Indicates the true hardness average value.

[0109] In this embodiment, hyperparameter tuning of the constructed hardness prediction model algorithm is also included, as follows:

[0110] The scikit-learn library in Python is used to call GridSearchCV to tune the hyperparameters of the algorithm; the hyperparameter search space is selected with the default parameters as the center.

[0111] In this embodiment, the XGBoost algorithm hyperparameters used to construct the hardness prediction model include n_estimators, min_child_weight, max_depth, Subsample, colsample_bytree, and learning_rate; the RF algorithm hyperparameters include n_estimators, max_leaf_nodes, max_depth, min_samples_leaf, and min_samples_split; the SVR algorithm hyperparameters include kernel, C, gamma, and epsilon. The search space is shown in Table 1:

[0112] Table 1 Hyperparameter search space

[0113]

[0114] Furthermore, the SHAP method was used to analyze the importance of features, and the influence of important features on hardness was determined based on the SHAP summary diagram combined with the ICE and PDP methods, which provided preliminary guidance for the subsequent design of high-hardness and high-entropy ceramics.

[0115] In this embodiment, the constructed hardness prediction model is optimized based on a genetic algorithm to obtain a target solution set, including:

[0116] Use the DEAP library in Python to call the genetic algorithm, set the genetic algorithm parameters, determine the constraints and objective function; determine the population size, number of iterations, crossover factor and mutation factor.

[0117] Specifically, the objective function is the constructed hardness prediction model, and the constraint condition is that the objective function is greater than the specified hardness.

[0118] The population size is set to 50, the number of iterations is set to 100, and the crossover factor and mutation factor are set to 0.5.

[0119] Furthermore, the Com eigenvalue in the result returned by the genetic algorithm is the result of PCA dimensionality reduction, so it is reconstructed by PCA dimensionality reduction in the hope of obtaining the constituent elements of the specific target high-entropy ceramic. However, PCA dimensionality reduction will lead to information loss, and the reconstructed information cannot be accurately reversed to integer values ​​of 0 and 1, so 0.7 is defined as the threshold. If it is greater than the threshold, it is considered to be 1, indicating that this element is included, otherwise it is considered to be 0, that is, it is not included.

[0120] In the practical engineering application of structural materials, quantitative mechanical performance index requirements are often required. In this embodiment, the above method is used to output a directional design method for high entropy ceramics with a Vickers hardness greater than 25GPa. The Vickers hardness of high entropy ceramics is predicted by the XGBoost algorithm using the above-screened features, and then according to the genetic algorithm, the predicted hardness value of the above-mentioned hardness prediction model is used as the objective function, and the predicted result greater than 25GPa is used as a constraint condition for target optimization, and the target solution set is returned. The Com eigenvalue is reconstructed by dimensionality reduction, and 0.7 is used as the threshold to select elements with a result greater than 0.7 as the specific component elements of the target high entropy ceramics.

[0121] In this embodiment, relevant literature was screened by using the SCI literature library with high entropy ceramics and mechanical properties as keywords, and a total of 110 sets of Vickers hardness data were collected. The initial feature space of the prediction model was formed according to the above formulas (1)-(4) and the feature engineering method. Feature screening was then performed based on Pearson correlation coefficient analysis and genetic algorithm, and the final results are shown in Table 2:

[0122] Table 2 Feature screening results

[0123]

[0124] 70% of the data are randomly selected to train the XGBoost model, and the remaining 30% of the data are used as a test set to verify the model performance. The prediction model verification results corresponding to different dimensions of Com are shown in Table 3:

[0125] Table 3 Corresponding performance parameters of Com models of different dimensions

[0126]

[0127] It can be seen from Table 3 that the high entropy ceramic performance prediction model has the most accurate prediction results when Com is 1 dimension. The comparison between the predicted value and the true value is as follows. Figure 2 Therefore, SHAP, ICE and PDP interpretability analysis were performed on the model, and the results are as follows Figures 3 to 6 shown.

[0128] Using genetic algorithm, the population size is determined to be 50, the number of iterations is 200, the crossover factor is 0.5, and the mutation factor is 0.1. In order to solve the problem of component component feature Com information verification caused by feature dimensionality reduction, the 5-dimensional Com model is selected as the objective function to optimize the hardness prediction model. The predicted value is compared with the true value. Figure 3 A total of 79 groups of high entropy ceramics of different systems were returned, among which the results verified by previous studies and the experiments of this embodiment are shown in Table 4:

[0129] Table 4 Verified genetic algorithm optimization results

[0130]

[0131] Among them (Ta 0.25 Zr 0.25 Nb 0.25 Hf 0.25 )C and (Ti 0.25 Zr 0.25 Nb 0.25 Hf 0.25 )C is experimentally verified by this embodiment, and the XRD spectrum is as follows Figure 8 shown.

[0132] Example 2

[0133] This embodiment provides a device for directional design of high entropy ceramics, which is used to implement the directional design method of high entropy ceramics described in Example 1, and the device includes:

[0134] A data acquisition module is used to obtain initial sample data of high entropy ceramic materials and construct an initial sample data set;

[0135] A feature calculation module, used to calculate input feature data based on an initial sample data set;

[0136] A feature screening module performs feature screening on the input feature data to obtain screened features, and constructs a training set and a test set based on the screened features and corresponding mechanical property parameters of high entropy ceramic materials;

[0137] A model training module, used for building a hardness prediction model based on the training set and the test set;

[0138] The output optimization module optimizes the composition of the hardness prediction model to obtain a target component element solution set, and determines the target component element composition of the high-hardness high-entropy ceramic according to the target component element solution set as a directional design result of the high-entropy ceramic.

[0139] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for directional design of high entropy ceramics, characterized in that: include: Obtaining initial sample data of high entropy ceramic materials and constructing an initial sample data set; Calculate input feature data based on the initial sample data set; Performing feature screening on the input feature data to obtain screened features, and constructing a training set and a test set based on the screened features and corresponding mechanical property parameters of high entropy ceramic materials; Building a hardness prediction model based on the training set and the test set; The hardness prediction model is subjected to composition optimization to obtain a target component element solution set, and the target component element composition of the high-hardness high-entropy ceramic is determined according to the target component element solution set as a directional design result of the high-entropy ceramic.

2. The method for directional design of high entropy ceramics according to claim 1, characterized in that: The initial sample data includes mechanical property parameters, component intrinsic property parameters, component element parameters, preparation method parameters and preparation process parameters of high entropy ceramic materials; Wherein, the mechanical property parameters of the high entropy ceramic material include one or more of the Vickers hardness, Rockwell hardness, Brinell hardness and nanoindentation hardness of the high entropy ceramic material; And / or, the obtaining of initial sample data of high-entropy ceramic materials and the construction of an initial sample data set include: obtaining material data of bulk high-entropy ceramic materials from an SCI literature library, wherein the bulk high-entropy ceramic materials include high-entropy ceramic systems of three different anions: high-entropy carbides, high-entropy nitrides and high-entropy carbonitrides.

3. The method for directional design of high entropy ceramics according to claim 2, characterized in that: The input characteristic data includes the inherent property characteristics of the components, the component composition characteristics and the preparation process characteristics; The step of calculating input feature data based on the initial sample data set includes: Obtaining the intrinsic property characteristics of the component based on the intrinsic property parameters of the component; Obtain component composition characteristics Com and carbon-nitrogen ratio C / N based on component element parameters; The preparation process characteristics Syn are obtained based on the preparation method parameters and the preparation process parameters.

4. The method for directional design of high entropy ceramics according to claim 3, characterized in that: The component intrinsic property characteristics are obtained based on the component intrinsic property parameters, wherein the component intrinsic property characteristics include the average melting temperature, melting temperature deviation, average atomic radius, atomic radius deviation, average density, density deviation, average mass, mass deviation, average electron affinity, electron affinity deviation, average Pauling electronegativity, Pauling electronegativity deviation, average ionization energy and fractional weighted average of ionization energy deviation of all constituent elements and variance , the calculation method is: in, It is the first of all the components The fractional weighted average of the intrinsic properties, For all the components The variance of the intrinsic property, Indicates The first component An inherent property, Indicates The mole fraction of an element, is the number of constituent elements; The intrinsic property characteristics of the components also include the average valence electron concentration of the corresponding precursor of the synthesized high entropy ceramic material. and valence electron concentration difference , the calculation method is: in , , is the mole fraction of carbon, is the mole fraction of nitrogen, Indicates The mole fraction of an element, is the number of cations in the high entropy ceramic material, For the The valence electron concentration of each constituent element, and They represent the valence electron concentrations of carbon and nitrogen in the corresponding precursors for synthesizing high entropy ceramic materials; And / or, the carbon-nitrogen ratio C / N has a value range of [0, 1], when the atomic percentage of nitrogen is 0, the carbon-nitrogen ratio is assigned a value of 1; when the atomic percentage of carbon is 0, the carbon-nitrogen ratio is assigned a value of 0; And / or, the obtaining of component composition characteristics Com based on component element parameters includes: Perform one-hot encoding on the constituent element parameters to obtain the binary component features; The binary component features are reduced to 1, 3, 5, 7 and 9 dimensions respectively to obtain Com of different dimensions; And / or, obtaining the preparation process characteristics Syn based on the preparation method parameters and the preparation process parameters includes: Performing target encoding on the preparation method parameters to obtain encoded preparation method parameters; Performing mode filling on the missing values ​​of the preparation process parameters to obtain the filled preparation process parameters; The encoded preparation method parameters and the filled preparation process parameters are reduced to 1 dimension to obtain the preparation process characteristics, which are denoted as Syn.

5. The method for directional design of high entropy ceramics according to claim 1, characterized in that: Performing feature screening on the input feature data to obtain screened features includes: Calculate the Pearson correlation coefficient between the input feature data pairs, and select the input feature data pairs with the Pearson correlation coefficient greater than 0.7; Calculate the Pearson correlation coefficient between two input feature data and the target feature in all input feature data pairs, delete the input feature data with the Pearson correlation coefficient between each pair of input feature data and the target feature, and obtain the input feature data for preliminary screening; The genetic algorithm is used to train the model and the random forest maximization model evaluation criteria are As the objective function, the iterative process is performed to obtain the selected features, wherein the genetic algorithm is iterated 100-500 times, the population value is set to 50-100, and the crossover factor and mutation factor range is 0.1-0.

5.

6. The method for directional design of high entropy ceramics according to claim 4, characterized in that: The step of constructing a hardness prediction model based on the training set and the test set includes: Calling the Support Vector Regression, Random Forest and eXtreme Gradient Boosting algorithms from the scikit-learn library in Python, and building the Vickers hardness prediction model under each algorithm based on the training set and the test set; The Scikit-learn library in Python is used to call GridSearchCV to perform hyperparameter tuning on the Vickers hardness prediction model under each algorithm, and the Vickers hardness prediction model under each algorithm after hyperparameter tuning is obtained; Calculate the determination coefficient of the Vickers hardness prediction model under each algorithm after the hyperparameter tuning With the root mean square error RMSE, choose The algorithm with the largest value and the smallest RMSE is used as the final algorithm of the Vickers hardness prediction model under each algorithm after hyperparameter tuning, and the optimal machine learning algorithm is obtained; Based on the preferred machine learning algorithm, Vickers hardness prediction models containing Com of different dimensions are constructed respectively; Calculation of the coefficient of determination for the Vickers hardness prediction model involving different dimensions of Com ,choose The dimension Com greater than 0.8 is taken as the target dimension Com, and the Vickers hardness prediction model including the target dimension Com is obtained, which is the hardness prediction model finally constructed.

7. The method for directional design of high entropy ceramics according to claim 6, characterized in that: The coefficient of determination The calculation formula is: ； The root mean square error RMSE calculation formula is: ； in, is the total number of samples, It is The true hardness value of the samples, Representative The hardness prediction value of samples, Indicates the true hardness average value.

8. The method for directional design of high entropy ceramics according to claim 7, characterized in that: The step of optimizing the composition of the hardness prediction model to obtain a target component element solution set and determining the target component element composition of the high-hardness high-entropy ceramic according to the target component element solution set includes: A genetic algorithm is used to perform target optimization on the Vickers hardness prediction model including the target dimension Com, so as to obtain a target solution set returned by the Vickers hardness prediction model including the target dimension Com; According to the size of the target solution set returned by different models, the principal component analysis and dimensionality reduction reconstruction of the component composition characteristics Com are performed to restore the constituent elements of the target high entropy ceramics. The system with the largest number of restored target high entropy ceramics is selected, and the model Models with values ​​greater than or equal to 0.8 are used as hardness prediction models for directional design; The Com in the target solution set of the hardness prediction model used for directional design is reconstructed to obtain the target element composition of high-hardness and high-entropy ceramics.

9. The method for directional design of high entropy ceramics according to claim 8, characterized in that: The genetic algorithm is used to perform target optimization on the Vickers hardness prediction model including the target dimension Com, including calling the genetic algorithm in the DEAP library in python to perform target optimization on the Vickers hardness prediction model including the target dimension Com, wherein the objective function is the predicted hardness value of the Vickers hardness prediction model, the constraint condition is that the predicted hardness value of the Vickers hardness prediction model is greater than the preset target hardness value, the population size is set to 10-100, the number of iterations is set to 100-1000, and the crossover factor and the variation factor are set to 0.1-0.

5.

10. A device for designing orientation of high entropy ceramics, characterized in that: A method for realizing the directional design of high entropy ceramics according to any one of claims 1 to 9, the device comprising: A data acquisition module is used to obtain initial sample data of high entropy ceramic materials and construct an initial sample data set; A feature calculation module, used to calculate input feature data based on an initial sample data set; A feature screening module performs feature screening on the input feature data to obtain screened features, and constructs a training set and a test set based on the screened features and corresponding mechanical property parameters of high entropy ceramic materials; A model training module, used for building a hardness prediction model based on the training set and the test set; The output optimization module optimizes the composition of the hardness prediction model to obtain a target component element solution set, and determines the target component element composition of the high-hardness high-entropy ceramic according to the target component element solution set as a directional design result of the high-entropy ceramic.

Images