A data-driven based design method for super wear-resistant and tough nitride high-entropy ceramics

Abstract

The application discloses a kind of based on data-driven super wear-resistant tough nitride high-entropy ceramic design method, comprising: collecting and collating nitride high-entropy ceramic wear resistance and fracture toughness historical experimental data information, according to nitride high-entropy ceramic element ratio and element inherent physical and chemical properties, input feature is structured and calculated;The input feature is screened to delete redundant features, and the original data set of nitride high-entropy ceramic wear resistance and fracture toughness prediction is obtained;The original data set is imported into the generative adversarial network model built to carry out data enhancement, and the enhanced data set with expanded sample capacity is obtained;The selected different machine learning algorithm model is trained using enhanced data set, and the optimal model is selected by evaluation;According to the optimal model, feature importance analysis is carried out, and the internal correlation between nitride high-entropy ceramic performance and input feature is established, so as to design a new type of super wear-resistant tough nitride high-entropy ceramic, and the application relates to the technical field of ceramic material design.The method solves the problems of long research and development cycle, low efficiency and high cost of multi-main-element ceramic materials, and the problems of overfitting and poor generalization ability of artificial intelligence data mining methods such as machine learning under small sample, and provides a high-precision, performance-oriented efficient design method based on data-driven for new super wear-resistant tough nitride high-entropy ceramic.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of ceramic material design, in particular to a design method of super-wear-resistant and strong-tough nitride high-entropy ceramics based on data driving. BACKGROUND

[0002] High-entropy ceramics are multi-principal element ceramic materials composed of five or more metal cations in equal or nearly equal amounts. With the unique high-entropy effect and lattice distortion effect, high-entropy ceramics often have higher mechanical properties, oxidation resistance and corrosion resistance, and have broad development prospects in the fields of aerospace, marine vessels and other national defense and military industries. Therefore, it is of great strategic significance to develop new wear-resistant and strong-tough high-entropy ceramic materials.

[0003] As a branch of high-entropy ceramics, nitride high-entropy ceramics are single solid solution phase ceramics composed of five or more elements and nitrogen. Due to the excellent phase stability and the hysteresis diffusion effect caused by the synergistic effect of elements, nitride high-entropy ceramics have higher hardness, toughness, thermal stability and wear resistance compared with mainstream transition metal nitrides such as titanium nitride, aluminum titanium nitride and chromium aluminum titanium nitride, and are expected to become the next generation of high-performance protective coatings. However, due to the nearly infinite element ratio combination and complex element component space, the design of nitride high-entropy ceramic systems is quite difficult. At present, the design of nitride high-entropy ceramics mainly relies on experimental trial and error methods. For example, a nitride high-entropy ceramic fiber and its preparation method and application are disclosed in Chinese Patent Publication No. CN111592361A. The invention prepares a single-phase nitride high-entropy ceramic fiber containing Ti, Hf, Ta, Nb and Mo elements, which can be applied to the process of photocatalytic carbon dioxide to prepare methane. Chinese Patent Publication No. CN113004047A discloses a (CrZrTiNbV)N high-entropy ceramic block and a preparation method thereof. The invention realizes the hot-pressing sintering of (CrZrTiNbV)N high-entropy ceramic block, obtains high-entropy ceramic with single-phase face-centered cubic structure, and significantly improves the fracture toughness. However, this method mainly relies on human experience and requires a large number of trial and error experiments, which has the problems of long research and development cycle, low efficiency, high cost and the like, and it is difficult to accurately design the components of nitride high-entropy ceramics with desired performance. In addition, due to the limited number of experimental samples of nitride high-entropy ceramics, machine learning and other artificial intelligence data mining methods are prone to overfitting and poor generalization, and cannot be effectively applied to the design of new nitride high-entropy ceramics.

[0004] Therefore, there is an urgent need for a new ceramic material design method based on data driving, high precision and performance demand, to solve the problems of difficult design, low efficiency and poor precision of new high-performance nitride high-entropy ceramic materials. SUMMARY

[0005] The application aims to provide a data-driven super-wear-resistant strong and tough nitride high-entropy ceramic design method to solve the problems of difficult design, long research and development cycle, low efficiency, high cost, poor precision and the like of new high-performance nitride high-entropy ceramic materials.

[0006] To achieve the above object, the application adopts the following technical scheme:

[0007] A data-driven super-wear-resistant strong and tough nitride high-entropy ceramic design method comprises the following steps:

[0008] Historical experimental data information of wear resistance and fracture toughness of nitride high-entropy ceramics is collected and sorted, input features are constructed and calculated according to element ratio of the nitride high-entropy ceramics and inherent physical and chemical properties of the elements;

[0009] The input features are subjected to feature screening to delete redundant features, and an original data set for wear resistance and fracture toughness prediction of the nitride high-entropy ceramics is obtained;

[0010] The original data set is imported into a generated adversarial network model to be built to perform data enhancement, and an enhanced data set with expanded sample capacity is obtained;

[0011] The generated adversarial network model comprises a generator and a discriminator, wherein the generator is responsible for converting random input into false samples simulating real sample data distribution, and the discriminator is responsible for identifying whether a mixed sample composed of false samples and part of real samples is a real sample; through continuous iterative training, the discriminator will become more and more skilled in distinguishing whether the sample generated by the generator is a real sample, while the generator will create samples more consistent with the real sample data distribution to try to deceive the discriminator, and finally reach Nash equilibrium; through the well-trained generator, the original data set with small sample size can be expanded into an enhanced data set with sufficient samples;

[0012] Different machine learning algorithm models selected are trained by using the enhanced data set, and the model with the best performance is evaluated and selected;

[0013] According to the optimal model, feature importance analysis is performed, and the internal correlation between the performance of the nitride high-entropy ceramics and the input features is established, so that a new super-wear-resistant strong and tough nitride high-entropy ceramic is designed.

[0014] Further, the elements include eleven elements of Al, Si, Ti, V, Cr, Zr, Nb, Mo, Hf, Ta and W, and the nitride high-entropy ceramic is a multi-principal-element high-entropy ceramic system composed of any four, five or six elements of the elements.

[0015] Further, the inherent physical and chemical properties of the elements include valence electron concentration, electron affinity, lattice constant, total energy, atomic radius, molar heat capacity, Mendeleev number, space group, first ionization energy, Pauling electronegativity, density and mixing entropy, and the input features are weighted averages and deviations of the inherent properties of the elements.

[0016] Further, the feature screening uses a stability selection algorithm to calculate the stability coefficients of each feature, and the principle of feature selection is to retain features with a stability coefficient greater than 0.9 as important features and delete other redundant features.

[0017] Further, the optimal feature combination of the feature screening includes atomic radius deviation, total energy deviation, valence electron concentration deviation, Mendeleev number deviation, electron affinity deviation and mixing entropy.

[0018] Further, the generator includes 1 input layer, 4 hidden layers and 1 output layer, the number of neurons contained in the layers is 40, 128, 256, 128, 64 and 30 respectively, the hidden layer activation function uses ReLU function, and the output layer activation function uses Tanh function; the discriminator includes 1 input layer, 4 hidden layers and 1 output layer, the number of neurons contained in the layers is 30, 64, 128, 64, 32 and 1 respectively, the hidden layer activation function uses LeakyReLU function, and the output layer activation function uses Sigmoid function; the generator and the discriminator each hidden layer is processed by dropout regularization.

[0019] Further, the machine learning algorithm includes random forest, K-nearest neighbor, gradient boosting decision tree, support vector machine, extreme gradient boosting and artificial neural network, and the optimal machine learning model is a support vector machine model.

[0020] Further, the kernel function of the support vector machine model uses a Gaussian kernel function, the regularization parameter is set to 192, and the kernel coefficient is set to 0.195.

[0021] Further, the feature importance analysis uses a permutation importance algorithm, and the number of iterations is set to 10.

[0022] The application provides a data-driven super-wear-resistant and strong-tough nitride high-entropy ceramic design method, which can obtain an optimal feature combination for predicting the wear resistance and fracture toughness of the nitride high-entropy ceramic by retaining important features and deleting redundant features through feature screening, significantly improves the training efficiency of a machine learning model and reduces the overfitting risk of the model; a generative adversarial network is built to perform data enhancement on an original data set, so that the small sample data set can be expanded, the data set quality is improved, and the prediction accuracy of the machine learning model is improved; the potential errors caused by single model prediction can be reduced through collaborative optimization and evaluation of multiple machine learning algorithms; the wear resistance and fracture toughness of all five-element or six-element nitride high-entropy ceramic systems composed of elements in a specified element interval can be predicted through the optimized support vector machine model, so that the low-cost, high-precision prediction and efficient design of the super-wear-resistant and strong-tough nitride high-entropy ceramic are realized. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A data-driven super-wear-resistant and strong-tough nitride high-entropy ceramic design method flowchart is provided for the specific embodiment of the application.

[0024] Figure 2 A generative adversarial network generator and discriminator training loss diagram is provided for the specific embodiment of the application.

[0025] Figure 3 A training and evaluation diagram of different machine learning models is provided for the specific embodiment of the application.

[0026] Figure 4 A feature importance analysis diagram of the optimal feature combination is provided for the specific embodiment of the application.

[0027] Figure 5 A five-element nitride high-entropy ceramic wear resistance and fracture toughness prediction diagram of the specific embodiment 1 of the application is provided.

[0028] Figure 6 A six-element nitride high-entropy ceramic wear resistance and fracture toughness prediction diagram of the specific embodiment 2 of the application is provided. DETAILED DESCRIPTION

[0029] The technical solutions of the application will be further described below with reference to the drawings and through specific embodiments.

[0030] Please refer to Figures 1-4 The specific embodiments of the application provide a technical solution: a data-driven super-wear-resistant and strong-tough nitride high-entropy ceramic design method, which comprises:

[0031] The historical experimental data information of the wear resistance and fracture toughness of the nitride high-entropy ceramic is collected and sorted, and the input features are constructed and calculated according to the element ratio of the nitride high-entropy ceramic and the inherent physical and chemical properties of the elements.

[0032] Feature screening is performed on the input features to delete redundant features, to obtain an original dataset for predicting the wear resistance and fracture toughness of nitride high-entropy ceramics;

[0033] The original dataset is imported into the established generative adversarial network model for data enhancement, to obtain an enhanced dataset with expanded sample capacity;

[0034] The generative adversarial network model includes a generator and a discriminator, wherein the generator is responsible for converting random inputs into false samples simulating the data distribution of real samples, and the discriminator is responsible for identifying whether the mixed samples composed of false samples and part of real samples are real samples; through continuous iterative training, the discriminator will become more and more skilled in distinguishing whether the samples generated by the generator are real samples, while the generator will create samples more consistent with the data distribution of real samples to try to deceive the discriminator, and ultimately reach Nash equilibrium; through well-trained generator, the original dataset with small sample size can be expanded into an enhanced dataset with sufficient samples;

[0035] The selected different machine learning algorithm models are trained using the enhanced dataset, and the model with the best performance is selected;

[0036] According to the optimal model, feature importance analysis is performed to establish the internal relationship between the performance of nitride high-entropy ceramics and the input features, so as to design a new type of super-wear-resistant and strong-tough nitride high-entropy ceramic.

[0037] Specifically, the elements include Al, Si, Ti, V, Cr, Zr, Nb, Mo, Hf, Ta, and W eleven elements, and the nitride high-entropy ceramic is a multi-principal-element high-entropy ceramic system composed of any four, five or six elements among the elements; the inherent physical and chemical properties of the elements include valence electron concentration, electron affinity, lattice constant, total energy, atomic radius, molar heat capacity, Mendeleev number, space group, first ionization energy, Pauling electronegativity, density and mixing entropy, the input features are weighted average and deviation of the inherent properties of the elements; the feature screening adopts a stability selection algorithm to calculate the stability coefficients of the features, and the feature selection principle is to retain the features with a stability coefficient greater than 0.9 as important features and delete the other redundant features; the optimal feature combination of the feature screening includes atomic radius deviation, total energy deviation, valence electron concentration deviation, Mendeleev number deviation, electron affinity deviation and mixing entropy; the generator includes 1 input layer, 4 hidden layers and 1 output layer, the number of neurons contained in the layers is 40, 128, 256, 128, 64 and 30 respectively, the hidden layer activation function adopts the ReLU function, and the output layer activation function adopts the Tanh function; the discriminator includes 1 input layer, 4 hidden layers and 1 output layer, the number of neurons contained in the layers is 30, 64, 128, 64, 32 and 1 respectively, the hidden layer activation function adopts the LeakyReLU function, and the output layer activation function adopts the Sigmoid function; the generator and the discriminator are both subjected to dropout regularization processing after each hidden layer; the machine learning algorithm includes random forest, K nearest neighbor, gradient boosting decision tree, support vector machine, extreme gradient boosting and artificial neural network, wherein the optimal machine learning model is a support vector machine model; the kernel function of the support vector machine model adopts a Gaussian kernel function, the regularization parameter is set to 192, and the kernel coefficient is set to 0.195; the feature importance analysis adopts the permutation importance algorithm, wherein the number of iterations is set to 10.

[0038] Figure 5 The five-element nitride high-entropy ceramic wear resistance and fracture toughness prediction map provided for the specific embodiment 1 of the present application, this specific embodiment is to design a super wear-resistant and strong five-element nitride high-entropy ceramic, and the design method is specifically described as follows:

[0039] The historical experimental data information of the wear resistance and fracture toughness of the nitride high-entropy ceramic is collected and arranged, a total of 98 groups of data, according to the element ratio and the inherent physical and chemical properties of the elements of the nitride high-entropy ceramic, the input features are constructed and calculated;

[0040] The input features are screened, the stability coefficients of the features are calculated by using a stability selection algorithm, the features with a stability coefficient greater than 0.9 are retained, and the remaining redundant features are deleted, to obtain an optimal input feature combination, and obtain the original data set for predicting the wear resistance and fracture toughness of the nitride high-entropy ceramic;

[0041] The original data set is imported into the built generative adversarial network model for data enhancement, wherein the generative adversarial network model comprises a generator and a discriminator; the generator is responsible for converting random input into false samples simulating the data distribution of real samples; the discriminator is responsible for identifying whether the mixed sample composed of false samples and part of real samples is a real sample; through continuous iterative training, the discriminator will become more and more skilled in distinguishing whether the sample generated by the generator is a real sample, and the generator will create samples more consistent with the data distribution of real samples to try to deceive the discriminator; as shown in Figure 2 , the generator and the discriminator have been iterated for 5000 times and reached Nash equilibrium, so as to obtain an enhanced data set with expanded sample capacity, and the enhanced data set comprises 298 groups of data;

[0042] The selected different machine learning algorithm models are trained by using the enhanced data set, and the model with the best performance is selected, as shown in Figure 3 , wherein the prediction accuracy of the support vector machine is the highest, reaching 93.9%;

[0043] According to the optimal support vector machine model, the feature importance analysis is performed, and the internal correlation between the performance of nitride high-entropy ceramics and the input features is established, as shown in Figure 4 , wherein the atomic radius deviation has the greatest influence on the wear resistance and fracture toughness of the nitride high-entropy ceramics, and the total energy deviation is second;

[0044] A five-element nitride high-entropy ceramic system composed of Al, Si, Ti, V, Cr, Zr, Nb, Mo, Hf, Ta and W is established by using the support vector machine model, and the wear resistance and fracture toughness are predicted, and the component space size is , as shown in Figure 5 , the model can design a new type of super wear-resistant and strong-tough five-element nitride high-entropy ceramic which has not been reported by experiments, and the specific component is (CrMoTaVZr)N5, the fracture toughness evaluation index H 3 / E 2 is about 1.61 GPa, and the wear resistance evaluation index H / E is about 0.25; in addition, H 3 / E 2 of more than 1 GPa and H / E of more than 0.2 also includes (CrMoTiVZr)N5 and (CrHfMoTaV)N5.

[0045] Figure 6 The six-element nitride high-entropy ceramic wear resistance and fracture toughness prediction diagram provided for the specific embodiment 2 of the present application, this specific embodiment is to design a super wear-resistant and strong-tough five-element nitride high-entropy ceramic, and the design method is described as follows:

[0046] The data set construction, data set enhancement and model training are the same as Embodiment 1; according to the optimal support vector machine model, a six-element nitride high-entropy ceramic system composed of Al, Si, Ti, V, Cr, Zr, Nb, Mo, Hf, Ta and W is established, and the wear resistance and fracture toughness are predicted, and the component space size is As shown in Figure 6 The model can design a new super wear-resistant and strong six-element nitride high-entropy ceramic which has not been reported by experiments, and the specific components are (CrMoNbTaVZr)N6, the fracture toughness evaluation index H 3 / E 2 is about 1.71 GPa, and the wear resistance evaluation index H / E is about 0.28; in addition, H 3 / E 2 The new six-element nitride high-entropy ceramic with H / E higher than 1 GPa and H / E higher than 0.2 also includes (CrMoNbTiVZr)N6 and (CrMoNbVWZr)N6.

[0047] The above is an embodiment given in combination with the drawings, which is only a preferred scheme for realizing the present application but not a limitation thereof, and any modification to the specific embodiments of the present application or equivalent replacement to part of the technical features without departing from the spirit of the technical scheme of the present application shall be covered in the technical scope of the present application claimed for protection. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative labor shall fall within the protection scope of the present application.

Claims

1. A data-driven based super wear-resistant refractory nitride high-entropy ceramic design method, characterized in that, The method comprises: Collecting historical experimental data information of nitride high-entropy ceramic wear resistance and fracture toughness, constructing and calculating input features according to element ratio of the nitride high-entropy ceramic and inherent physical and chemical properties of the elements; Performing feature screening on the input features to delete redundant features, and obtaining an original data set for predicting the wear resistance and fracture toughness of the nitride high-entropy ceramic; Importing the original data set into a generated adversarial network model to perform data enhancement, and obtaining an enhanced data set with expanded sample capacity; The generated adversarial network model comprises a generator and a discriminator, wherein the generator is responsible for converting random input into false samples simulating the data distribution of real samples, and the discriminator is responsible for identifying whether the mixed samples composed of false samples and part of real samples are real samples; through continuous iterative training, the discriminator will become more and more skilled in distinguishing whether the samples generated by the generator are real samples, while the generator will create samples more consistent with the data distribution of real samples in an attempt to deceive the discriminator, and finally reach Nash equilibrium; through the well-trained generator, the original data set with small sample size can be expanded into an enhanced data set with sufficient samples; Training selected different machine learning algorithm models using the enhanced data set, evaluating and selecting the model with the best performance; Performing feature importance analysis according to the optimal model to establish the internal correlation between the performance of the nitride high-entropy ceramic and the input features, and designing a new type of super-wear-resistant and strong-tough nitride high-entropy ceramic.

2. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The elements include Al, Si, Ti, V, Cr, Zr, Nb, Mo, Hf, Ta and W, and the nitride high-entropy ceramic is a multi-principal-element high-entropy ceramic system composed of any four, five or six of the elements.

3. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The inherent physical and chemical properties of the elements include valence electron concentration, electron affinity, lattice constant, total energy, atomic radius, molar heat capacity, Mendeleev number, space group, first ionization energy, Pauling electronegativity, density and mixing entropy, and the input features are weighted average and deviation of the inherent properties of the elements.

4. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The feature screening adopts a stability selection algorithm to calculate the stability coefficients of each feature, and the feature selection principle is to retain the features with a stability coefficient greater than 0.9 as important features and delete other redundant features.

5. A data-driven based super wear-resistant tough nitride high-entropy ceramic design method according to claim 4, characterized in that: The optimal feature combination of the feature screening includes atomic radius deviation, total energy deviation, valence electron concentration deviation, Mendeleev number deviation, electron affinity deviation and mixing entropy.

6. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The generator comprises one input layer, four hidden layers and one output layer, the number of neurons contained in the layers is 40, 128, 256, 128, 64 and 30 respectively, the activation function of the hidden layers adopts ReLU function, and the activation function of the output layer adopts Tanh function; the discriminator comprises one input layer, four hidden layers and one output layer, the number of neurons contained in the layers is 30, 64, 128, 64, 32 and 1 respectively, the activation function of the hidden layers adopts LeakyReLU function, and the activation function of the output layer adopts Sigmoid function; the generator and the discriminator are subjected to dropout regularization processing after each hidden layer.

7. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The machine learning algorithm includes random forest, K-nearest neighbor, gradient boosting decision tree, support vector machine, extreme gradient boosting, and artificial neural network, wherein the optimal machine learning model is a support vector machine model.

8. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 7, characterized in that: The kernel function of the support vector machine model adopts a Gaussian kernel function, the regularization parameter is set to 192, and the kernel coefficient is set to 0.

195.

9. The data-driven based super wear-resistant and tough nitride high-entropy ceramic design method according to claim 1, characterized in that: The feature importance analysis adopts a permutation importance algorithm, wherein the number of iterations is set to 10.

Images