| γ’phase strengthened high entropy alloys have been widely studied in recent years due to their excellent mechanical properties as well as high temperature resistance.However,due to the characteristics of elemental multiplicity and complexity of preparation process of high-entropy alloys,improving their properties by experimental methods can waste a lot of time.Phase diagrams and thermodynamic calculations also have limitations and cannot significantly accelerate the design process ofγ’precipitation phase reinforced high entropy alloys.Therefore,a new strategy is needed to predict the properties of the alloy quickly and accurately.Since the 21st century,the field of artificial intelligence has witnessed a new climax with the continuous maturation of data mining and data statistical techniques.Machine learning techniques have shown great potential in new material development as well as engineering applications,especially in property optimization showing the superiority of artificial intelligence and improving the efficiency of material design.In this study,a prediction system forγ’precipitation phase reinforced high entropy alloys with a multilayer structural model was developed by combining material genetic engineering machine learning techniques with experimental studies,and the method was applied to the design ofγ’phase reinforced multi-system high entropy alloys.The main research results obtained in this study are as follows:(1)In this study,different machine learning algorithms were used to establish the prediction model for the determination of the physical phase of the high-entropy alloy and the prediction model for the determination of the existence ofγ’phase.The algorithm with the highest accuracy for building the prediction model of physical phase of high entropy alloy is random forest(98.36%);the algorithm with the highest accuracy for building the prediction model ofγ’phase presence determination is random forest(AUC=0.99).(2)In this paper,three machine learning+deep learning algorithms,random forest,limit gradient boosting,and deep neural network,are used to build aγ’-phase reinforced high-entropy alloyγ’-phase volume fraction prediction model.by comparing the prediction performance of different models,the limit gradient boosting prediction model has the best performance with R~2=0.952 and RMSE=3.88.and the model is analyzed using SHAP algorithm.The general rule of precipitation ofγ’phase by different elements and heat treatment process was analyzed.With the increase of Ni,Al and Ti elements,the number of γ’phase will gradually increase,but when the Al of the alloy>10at%,the alloy will change from FCC single phase to FCC+BCC dual phase organization and the number ofγ’phase will gradually decrease by the physical phase prediction model;when Ti>10at%,other heterogeneous phases will precipitate in theγ’phase strengthened high entropy alloy.The Cr element increases gradually in the interval of 0-8at%,and the characteristic Shap Value also increases gradually,but when the Cr element exceeds 8at%,the characteristic shap value starts to decrease gradually and the number of precipitated phases also decreases.And with the increase of Fe element content,the characteristic shap value of Fe also decreases.(3)In this study,random forest,deep neural network and limit gradient boosting regression algorithms were used to train the prediction model using 90%of the data randomly as the training dataset to establish a high entropy alloy hardness prediction model.By comparing the prediction performance of different models,the extreme gradient boosting prediction model has the best performance:R~2=0.902 and RMSE=38.561.machine learning is used to establish a mathematical relationship model of strengthening mechanisms to study the effects of various strengthening mechanisms on high-entropy alloys.We found that in the precipitation-reinforced high-entropy alloy,precipitation strengthening is the most dominant strengthening mode of the alloy,deformation strengthening contributes less to the strength of the alloy than precipitation strengthening,and solid solution strengthening contributes the least to the strength of the alloy.(4)Combining the established high entropy alloy physical phase prediction model,γ’phase volume fraction prediction model,and hardness prediction model,a machine learning strategy for multi-layer structure optimization model was designed to establish a comprehensive prediction system for high entropy alloy with multi-layer structure.Based on the prediction and screening of 800000 sets of data sets with different compositions and processes by the integrated prediction system of multilayer structure,84 sets of composition and process data sets satisfying the requirements were obtained.Through experimental verification of the four candidate alloys,theγ’phase volume fraction and hardness of the four alloys are higher than the currently reported face-centered cubic lattice high-entropy alloys.A machine learning strategy using a multilayer structural optimization model can accelerate the efficiency as well as the accuracy of alloy design and ensure the calculation ofγ’phase reinforced high entropy alloys that meet the requirements.(5)In this study,random forest,deep neural network and extreme gradient boosting regression algorithms were used to train prediction models using 90%of the data randomly as the training data set to establish prediction models for tensile strength,yield strength,and elongation of high-entropy alloys in room-temperature tension.The best performing algorithms for the three prediction models were:random forest,extreme gradient boosting,and extreme gradient boosting.Room-temperature tensile tests were conducted on the candidate P1 and P2 alloys,and the P1 alloy maintained 1156 MPa tensile strength while still having 24%elongation;the P2 alloy had a yield strength of 1136 MPa and still maintained good plasticity.And by comparing the experimental real values with the predicted values,the errors of both tensile strength and yield strength models are less than 10%. |
PDF Full Text Request