Designing γ-TiAl Alloys With Refined Twin-Lamellar Microstructures via Machine Learning-Driven Phase-Field Methods

Abstract

Dual phase γ-TiAl alloys, characterized by fine lamellar and twin structures, offer an impressive balance between strength and elongation. However, predicting optimal design conditions, including content and heat treatment, for such microstructures poses significant challenges. We establish a microstructure-stress-content database by integrating phase-field simulations with machine learning optimization algorithms. The database reveals an initial increase followed by a decrease in twin structure density as both external stress and Al content rise. Concurrently, the average lamella size shows an initial decrease and subsequent increase. Machine learning is used to expand the phase-field database and positioned the refined twin-lamellar structure as our optimization objective. This allowed for an expedited search for the ideal design conditions as guided by optimization theory. Notably, phase-field simulations demonstrate that the optimal design condition results in a presence of refined twin-lamellar structures, characterized by an average twin size of ∼13.5 nm, twin fraction of approximately 92%, and γ-phase volume fraction of ∼48%. We conclude by detailing the physical factors governing the linear relationship observed between Al content and stress. This study ultimately provides valuable insights for the design of two-phase γ-TiAl alloys with superior lamellar and twin structures.