Machine learning aided first-principles studies of structure stability of Co3(Al, X) doped with transition metal elements

Abstract

To understand the alloying effects on the stability of Co3Al precipitate phase in Co-based superalloy, the energetic stability and structure of ternary alloy Co3(Al, X) doped with the thirty 3d, 4d, and 5d transition metal (TM) elements were studied in this work using first-principles (FP) computation and machine learning (ML) methods. Our FP computation indicated that Hf, Ta, and Ti doping were thermodynamically most stable. Based on the FP computation data, the ML models with three types of chemical composition (CC) and Center-Environment (CE) features, were developed to predict the formation energies and lattice constants of Co3(Al, X) (X = 3d, 4d, and 5d TM elements). The results show that these ML models all had good prediction accuracy with averaged mean absolute errors (<MAE > ) ~ 0.02 eV/atom for formation energies and ~ 0.01 Å for lattice constants. Then, the effects of important features were discussed on the energy and geometry properties. To study the alloying effects on the stability of Co3(Al, W) precipitate phase, the ML models of Co3(Al, X) were found to be equally accurate to predict the untrained structures of Co3(Al, WX3) where X = 3d, 4d, and 5d TM elements excluding Co and W. We found that the Co3(Al, WX3) structures became most stable when X are IVB and VB group TM elements. This work show that machine learning methods can efficiently extend the capabilities of first-principles predictions on the structure stabilities in multi-component alloy design.