Hardness Prediction Of AlCoCrCuFeNi High Entropy Alloys Based On Machine Learning

In recent years,high entropy alloys have attracted wide attention because of their unique excellent properties.The current research on high entropy alloy is usually experimental synthesis or complex theoretical calculation,but the former needs a lot of time and material cost,and even has high requirements for experimental equipment,while the calculation process of the latter is time-consuming,labor-consuming and has certain limitations.With the rapid development of artificial intelligence and computer technology,machine learning has gradually shown revolutionary advantages in material research and design,which has aroused great interest among researchers.Therefore,the high-entropy alloy composition design based on machine learning algorithms in order to achieve high hardness has very important research significance.Aiming at the common Al-Co-Cr-Cu-Fe-Ni system high-entropy alloys,this paper first collected six feature screening methods reported in the existing literature:Pearson correlation coefficient,univariate feature selection,stability selection,forward Sequence selection,backward sequence selection and genetic algorithm.Based on these 6 methods,the collected 20 physical features were screened to find the best combination of features,and the application time of each method and the accuracy of the results were compared and analyzed to summarize their respective features,including advantages and disadvantages as well as the scope of application.Through the above comparative research,it is found that the genetic algorithm has a relatively good comprehensive effect on the feature selection problem of high-entropy alloy hardness prediction.Secondly,aiming at the shortcomings of genetic algorithms discovered in the research process,this paper proposes four different improved genetic algorithms,which have been verified to be significantly better than the common genetic algorithms used before.Afterwards,a support vector regression model was used to construct a hardness prediction model based on the selected optimal feature combination,The regression coefficient of the model is 0.96,and the root mean square error of the prediction could reach 50.468672.The model was used to predict the hardness in the virtual space of the huge Al-Co-Cr-Cu-Fe-Ni system high-entropy alloy,and successfully searched out 6 high-entropy alloys with relatively high hardness.And the predicted value of the highest hardness is 791.5322 HV,which is a certain improvement compared to the highest hardness value in the original data set.Finally,the relationship between characteristics and hardness in feature combination is found through a lot of calculation and analysis.At the same time,the machine learning model and the improved genetic algorithm are verified to prove the rationality of the selected model and the improved genetic algorithm.