Machine learning-based precise identification and quantitative characterization of discontinuous phase transformation macrozones in titanium alloys

Texture macrozones in titanium alloys significantly influence mechanical properties, particularly fatigue and dwell fatigue behaviors. However, conventional identification methods exhibit limitations in processing complex-morphology macrozones, such as phase transformation macrozones, due to their discontinuous and discrete spatial distributions. To address this issue, an automated identification and quantitative characterization framework based on machine learning is developed herein. This framework integrates an improved K-means clustering algorithm with an adaptive region-growing mechanism, allowing high-precision identification and quantitative visualization of texture macrozones to be achieved via EBSD data of near-α titanium alloys. Mechanistically, the core technical advancement relies on an "eight-neighborhood + extended neighborhood" region-growing mechanism coupled with the "gap tolerance" criterion. This configuration successfully prevents identification interruptions and artifacts induced by macrozone discontinuities. Simultaneously, the framework continuously updates macrozone orientation data within an interactive mode, permitting both the adaptive tracking of complex boundaries and the detailed quantitative extraction of individual macrozone features. Evaluation on the Ti150 alloy dataset shows that the identification accuracy of a complex macrozone is 93.7%, representing an increase of 13.7% and 33.7% compared to traditional K-means (<80%) and fixed-threshold manual identification (<60%), respectively. Consequently, this method establishes a reliable microstructural characterization foundation for predicting and evaluating anisotropic mechanical properties.