High-entropy alloy hardness prediction method fusing auto-encoder dimensionality reduction and clustering modeling

Abstract

The invention relates to the technical field of high-entropy alloy hardness prediction, in particular to a high-entropy alloy hardness prediction method based on dimension reduction and clustering modeling of an auto-encoder, and the prediction method comprises the following steps: constructing an initial data set of high-entropy alloy hardness and physical characteristics; learning component data in the initial data set through multiple rounds of training by using an auto-encoder, and performing dimensionality reduction to obtain a hardness distribution two-dimensional hidden space of the high-entropy alloy; constructing a Gaussian mixture model in the two-dimensional space, and obtaining the optimal clustering number and clustering mode of the Gaussian mixture model by using an elbow rule; feature screening is carried out, a plurality of machine learning models are constructed and evaluated in each cluster, and the optimal models in each cluster are selected to form an integrated model with the optimal prediction effect; high-entropy alloy sample points needing to be predicted are input into the integrated model for prediction. The hardness prediction model obtained through the method can effectively improve the accuracy of high-entropy alloy hardness prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of high entropy alloy hardness prediction, and more specifically to a high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling. Background Art

[0002] High entropy alloys, also known as multi-principal alloys, are a type of alloy composed of five or more main elements with an atomic percentage of 5%-35%. Due to their unique high entropy effect, high entropy alloys exhibit many properties that are superior to traditional alloys, such as high hardness, high corrosion resistance and high temperature resistance, making their application space very broad. Although high entropy alloys have many advantages, there are still some challenges in alloy design. In the composition space of high entropy alloys with five or more elements, the traditional trial and error method requires the preparation of the alloy and then testing and analyzing the performance when designing high entropy alloys. This is not only time-consuming and costly, but also difficult to systematically explore and optimize all high entropy alloy composition points in the composition space.

[0003] Machine learning has shown superior application advantages in the design of high-entropy alloys. Compared with traditional empirical methods, machine learning can quickly screen a large number of components and ratios and discover the design space of new high-entropy alloys. The optimization method based on machine learning can quickly find alloy formulas that meet specific performance requirements, greatly shortening the development cycle of high-entropy alloys.

[0004] The Chinese patent announcement number is CN112216356A, which proposes a high entropy alloy hardness prediction method based on machine learning. This method screens the feature data, obtains the optimal feature combination, and uses the support vector regression algorithm to train the model to predict the hardness of unknown high entropy alloys. However, the current machine learning methods in materials science usually model the original data set as a whole. This method ignores the heterogeneity of data features and cannot fully explore the complex internal laws contained in the data, thus affecting the prediction accuracy and generalization ability of the model. The clustering modeling method can capture the potential distribution characteristics in the data, improve the model's fitting and generalization capabilities for heterogeneous data, and thus obtain better prediction results than the one-size-fits-all overall modeling method.

[0005] In summary, the present invention develops a high entropy alloy hardness prediction method that integrates autoencoder dimensionality reduction and clustering modeling, which has important theoretical significance and application value, and plays a vital role in promoting the development of the field of materials science and realizing the accurate prediction of the hardness of high entropy alloys. Summary of the invention

[0006] The present invention provides a high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling, wherein a two-dimensional latent space of the high entropy alloy hardness distribution is constructed by an autoencoder, and a Gaussian mixture model is used in combination with the elbow rule to cluster the latent space; then, a plurality of machine learning models are constructed for each cluster respectively, and the models are trained and evaluated, and the best model in each cluster is screened out to form an integrated model, thereby finally achieving the best prediction of the high entropy alloy hardness, so as to solve the problem of low prediction accuracy of the overall modeling method.

[0007] The specific technical solutions of the present invention are as follows:

[0008] A high entropy alloy hardness prediction method that integrates autoencoder dimensionality reduction and clustering modeling is designed as follows:

[0009] 1. A high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling, characterized in that the prediction method steps are as follows:

[0010] Step S1, construct and collect hardness data based on the Al-Co-Cr-Cu-Fe-Ni high entropy alloy system, select physical characteristics other than composition in combination with knowledge in the field of materials science, and construct a data set for further analysis;

[0011] Step S2, using an autoencoder to reduce the dimension of the component data in the data set and construct a two-dimensional latent space of the hardness distribution of the high entropy alloy;

[0012] Step S3, constructing a Gaussian mixture distribution model in a two-dimensional space, clustering the Gaussian mixture model using the elbow rule, and clustering the samples;

[0013] Step S4, feature screening of the original data set, training using multiple machine learning algorithms in each cluster, building multiple models, and evaluating all models to screen out the optimal model to form an integrated model;

[0014] Step S5, inputting unknown high entropy alloy sample points into the integrated model for prediction to obtain hardness prediction results.

[0015] As a technical solution of the present invention, in step S1, based on the Al-Co-Cr-Cu-Fe-Ni system, the composition and its corresponding hardness data are collected, combined with the atomic radius difference (δr), electronegativity difference (Δχ), valence electron concentration (VEC), mixing enthalpy (ΔH), configuration entropy (ΔS config ), vibration entropy (ΔS vib ), electron entropy (ΔS elec), Ω parameter (Ω), Λ parameter (Λ), γ parameter (γ), cohesive energy (Ec), modulus mismatch (η), local size mismatch (Dr), work function (w), shear modulus (G), shear modulus difference (δG), local modulus mismatch (DG), lattice distortion energy (μ), free energy (ΔG), unit cell volume (V), band gap (E g ), elastic constant (C), dielectric constant (ε r ) physical characteristics of high entropy alloys to form the initial data set.

[0016] As a technical solution of the present invention, in step S2, the autoencoder is divided into two parts, an encoder and a decoder, and is implemented by two neural networks. The encoder reduces the data from 128 dimensions to 2 dimensions, and the decoder restores the 2-dimensional encoded data to its original 128 dimensions.

[0017] As a technical solution of the present invention, in step S3, three different clusters are obtained when the Gaussian mixture model is clustered according to the elbow rule.

[0018] As a technical solution of the present invention, in step S4, a random forest-based sequence floating forward selection method is used in the feature screening step to screen out valence electron concentration (VEC), mixing enthalpy (ΔH), average cohesive energy (Ec) and work function (w) features for subsequent machine learning training.

[0019] As a technical solution of the present invention, in step S4, the multiple machine learning algorithms used include but are not limited to support vector regression, categorical feature boosting, decision tree, lightweight gradient boosting machine, back propagation neural network, random forest, extreme gradient boosting, adaptive boosting, and K nearest neighbor.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The present invention obtains the latent space of the hardness distribution of high entropy alloys by adopting an autoencoder algorithm, and clusters the latent space using a Gaussian mixture model combined with the elbow rule. Then, multiple machine learning models are constructed for each cluster, and the models are evaluated. The best model in each cluster is screened out and then formed into an integrated model, ultimately achieving higher model prediction accuracy than the overall modeling method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart for predicting hardness of high entropy alloys of the present invention;

[0023] Figure 2 It is a two-dimensional latent space clustering effect diagram of the high entropy alloy composition data set of the present invention;

[0024] Figure 3It is a Pearson correlation coefficient diagram of four characteristics screened by the present invention;

[0025] Figure 4 It is a prediction effect diagram of the integrated learning model constructed by the present invention.

Claims

1. A high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling, characterized in that: The prediction method includes the following steps: Step S1, construct and collect hardness data based on the Al-Co-Cr-Cu-Fe-Ni high entropy alloy system, select physical characteristics other than composition in combination with knowledge in the field of materials science, and construct a data set for further analysis; Step S2, using an autoencoder to reduce the dimension of the component data in the data set and construct a two-dimensional latent space of the hardness distribution of the high entropy alloy; Step S3, constructing a Gaussian mixture distribution model in a two-dimensional space, clustering the model using the elbow rule, and clustering the samples; Step S4, feature screening of the original data set, training using multiple machine learning algorithms in each cluster, building multiple models, and evaluating all models to screen out the optimal model to form an integrated model; Step S5, inputting unknown high entropy alloy sample points into the integrated model for prediction to obtain hardness prediction results.

2. According to claim 1, a high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling is characterized in that: In S1, based on the Al-Co-Cr-Cu-Fe-Ni system, the composition and its corresponding hardness data are collected, combined with the atomic radius difference (δr), electronegativity difference (Δχ), valence electron concentration (VEC), mixing enthalpy (ΔH), configuration entropy (ΔS config ), vibration entropy (ΔS vib ), electron entropy (ΔS elec ), Ω parameter (Ω), Λ parameter (Λ), γ parameter (γ), cohesive energy (Ec), modulus mismatch (η), local size mismatch (Dr), work function (w), shear modulus (G), shear modulus difference (δG), local modulus mismatch (DG), lattice distortion energy (μ), free energy (ΔG), unit cell volume (V), band gap (E g ), elastic constant (C), dielectric constant (ε r ) physical characteristics of high entropy alloys to form the initial data set.

3. The high entropy alloy hardness prediction method according to claim 1, characterized in that: In step S2, the autoencoder is divided into two parts, an encoder and a decoder, which are implemented by two neural networks. The encoder reduces the data from 128 dimensions to 2 dimensions, and the decoder restores the 2-dimensional encoded data to its original 128 dimensions.

4. The high entropy alloy hardness prediction method according to claim 1, characterized in that: In step S3, 2-5 different clusters are obtained when the Gaussian mixture model is clustered using the elbow rule.

5. The high entropy alloy hardness prediction method according to claim 1, characterized in that: In step S4, in the feature screening step, a random forest-based sequential floating forward selection method is used to screen out the valence electron concentration (VEC), mixing enthalpy (ΔH), average cohesive energy (Ec), and work function (w) features for subsequent machine learning training.

6. A high entropy alloy hardness prediction method integrating autoencoder dimensionality reduction and clustering modeling according to claim 1, characterized in that: In step S4, the various machine learning algorithms used include but are not limited to support vector regression, categorical feature boosting, decision tree, lightweight gradient boosting machine, back propagation neural network, random forest, extreme gradient boosting, adaptive boosting, and K nearest neighbor.

Images