Prediction And Composition Design Of Single-Phase High-Entropy Nitride Ceramics Via Machine Learning

High-entropy ceramics have attracted much attention due to their excellent properties and broad application prospects,however,their huge component space has brought great challenges to researchers in the field of high-entropy ceramics.The complicated experimental process seriously hinders the discovery of new high-entropy systems.The field of high-entropy ceramics urgently needs new ways to explore.The proposal of material genetic engineering has created a new paradigm based on data drive,and the realization of this concept needs to be combined with artificial intelligence technology.Based on the concept of "Material design on demand",this work established a Gradient Boosting Decision Tree(GBDT)model with a goodness of fit R2 of 0.9953,and predicted the entropy formation and single-phase forming abilities of 70 different high-entropy systems.After that,based on this model,a(TiVZrNbHf)Nx high-entropy system was designed,and a single-phase(TiVZrNbHf)Nx ceramic was successfully prepared by hotpressing sintering,and the phase composition,microstructure,element distribution and mechanical properties of the ceramic were characterized.The main research content and results of this paper are as follows:(1)Through filtering the features of the established initial dataset by the Lasso regression method and Pearson correlation coefficient method,the initial dataset was optimized to find out the relationship between features and target values.The Lasso regression model is established,and the features are compressed by determining the optimal λ value to achieve the purpose of screening features.The screening results show that the regression algorithm is suitable for the research of this paper.The features that have a greater impact on the EFA were explored through the ranking of feature importance.Based on this,the high-entropy ceramic systems of the follow-up experiment part were designed.(2)Based on supervised learning,four machine learning models were established:K Nearest Neighbor model,Random Forest model,Support Vector Regression model and Gradient Boosting Decision Tree model.Through the error index analysis,it is found that the GBDT model performs best,and then it is used for prediction.In the prediction process,the GBDT model is used to predict the EFA of 70 different high-entropy systems.Among the predicted results,the(AlTiCrMoW)Nx system has a larger EFA,and the single-phase(AlTiCrMoW)Nx system has been reported in the paper,which verifies the accuracy of the GBDT model.In order to explore that the GBDT model established in this paper has good generalization ability as well,the enthalpy of formation and the standard deviation of bond length is selected as the target value training model.From the model fitting situation,the R square of the model has reached more than 0.9,which shows that the GBDT model also fits the two new target values very well.(3)Based on the GBDT model,the(TiVZrNbHf)Nx high-entropy system was designed,and related experiments proved that the system has the ability to form a single phase.The phase analysis and microstructure observation and characterization of the obtained(TiVZrNbHf)Nx ceramics show that the system can form a single-phase rock-salt structure.The mechanical properties of the sample show that the ceramic exhibits improved hardness and fracture toughness compared to its one-component nitride.