Study On The Design, Fabrication And Property Of Novel Amorphous And Nanocrystalline Alloys By Machine Learning

Due to their excellent mechanical properties and corrosion resistance,amorphous/nanocrystalline alloy obtained much attention in the flied of wireless charging,aerospace and amorphous motor.In which,Fe–based amorphous/nanocrystalline alloy has advantages of low price and good soft magnetic properties,is a classic amorphous alloy.With the development of techology,design and preparation of novel amorphous alloy with excellent glass formation ability（GFA）and Fe–based amorphous/nanocrystalline alloy with good soft magnetic properties are still desired.Traditional“trial and error”method is too inefficient to design amorphous/nanocrystalline alloy.Fortunately,machine learning（ML）can accelerate design of novel materials,so as to achieve purposes of cost saving and efficient design.In this work,database for glass formation ability（GFA）and soft magnetic properties are estabilished,based on the dataset,ML models are established.Predictive ability of ML models is compared,and the ML model with best predictive ability is chose to predict the GFA and magnetic properties of alloys.Next,Python is employed to establish virtual datasets,atomic ratio of target samples with desired GFA and magnetic properties are selected from the virtual datasets to prepare.Finally,X–ray diffraction（XRD）,tansmission electronic microscopy（TEM）,differential scanning calorimeter（DSC）,physical property measurement system（PPMS）and vibrating sampke magnetometer（VSM）are used to analysis the phase structure and properties of samples,and verify the feasibility of ML models.The work of this paper is list as follows:（1）Machine learning（ML）classifiers including k–Nearest Neighbor（kNN）,Support Vector Machines（SVM）,Decision Tree（DT）,Random Forest（RF）,e Xtreme Gradient boosting Trees（XGBT）,Artificial Neural Network（ANN）and Logistic Regression（LR）are employed to design of novel M–based（M=Fe,Co,Ni,Ti,Zr and Rare Earth metal（RE））and X（X=2,3,4,5,6,and>6）components alloy with excellent GFA.With the assistance of feature selection analysis,it is found that the critical features of GFA include average melting of alloy（Tm）,electronegativity of alloys（X）,electronegativity difference of alloys（ΔX）,valence electron concentration of the alloy（VEC）,configurational entropy（ΔS）and atomic precent.Then,the GFA of alloy ribbon systems Fe₈₅Si₂B_xP_yC_z,Fe80.4Si0.2Nb0.5Bx Py Cz,Fe81.6Si0.2Nb0.5Bx Py Cz,Fe81.8Si0.2Nb0.5Bx Py Cz,Cox Siy Bz,Ni_xTi_yFe_zB₁₀,Zr_xCu_yB_z,Ti₂₆Zr_xSi_yMn_z,Ti₂₂B₂₄Ni_xCu_yCo_z and Y_xFe_yCu_z are predicted by ML.Finally,10 kinds of different metal–based alloys with good GFA designed by machine learning are successfully prepared,verifying the feasibility of established ML models.The present work provides an effective way to design multiple different metal–based amorphous alloys with high GFA.（2）LR,SVR,DTR,ANN and RFR are employed to build prediction models of soft magnetic properties.It is found that ANN has the excellent fitting ability with largest coefficient of determination（R²）to predict the soft magnetic properties of new designed alloys.Python screening is used to find the alloy compositions with best soft magnetic properties of Fe–B–P–C–Nb system form 5586000 virtual data.Fe₈₃B₉P₃C₄Nb₁ alloy with good soft magnetic properties has been designed and prepared to verify.The B_s of Fe₈₃B₉P₃C₄Nb₁ amorphous ribbon is 1.64T,and the B_s,T_c of Fe₈₃B₉P₃C₄Nb₁nanocrystalline ribbon is 1.71T and 702K.The measured results in accordance with the predicted results.（3）Gradient Boosting Decision Tree（GBDT）,e Xtreme Gradient boosting Trees（XGBT）,Artificial Neural Network（ANN）and k–Nearest Neighbor（kNN）are employed to predict the B_s,H_c,T_c,（initial permeability）μ_i,grain size,core loss and crystallization volume fraction of Fe–based amorphous alloy.In order to enhance predictive ability of ML models,normalization method is used to pre–process raw data and grid–search method is used to search the most proper parameters of ML models.Compared with other ML models,it is found XGBT has excellent predictive ability for predicting various properties,indicates established ML model provides a more precise way for accelerating design Fe–based amorphous/nanocrystalline alloy with desired properties.