Towards intelligent processing: Machine learning-based prediction of flow curves and hot workability maps for Ti71Fe25.15Sn3.85 alloy

Abstract

To identify the best conditions for hot deformation, it is necessary to design innovative alloy systems. Data on flow stress and strain under various hot working scenarios are critical for creating processing (Hot Workability) maps. The deformation characteristics of Ti<inf>71</inf>Fe<inf>25.15</inf>Sn<inf>3.85</inf> ternary alloys were explored through high-temperature compression experiments by a Gleeble® simulator at different temperature (700 °C, 800 °C, 900 °C, and 950 °C) with strain rates of 0.01 s−1, 0.1 s−1, and 10 s−1. The alloy exhibited a fine eutectic structure composed of β-Ti and FeTi phases, alongside coarse dendritic Ti<inf>3</inf>Sn and FeTi phases. Five machine learning (ML) models were employed for predicting the flow curve for another strain rate 1 s−1 and generating processing maps. The random forest (RF) model shows exceptional accuracy an R2 (coefficient of determination) of 96.7 %, RMSE (root mean square error) of 9.6 %, and MAE (mean absolute error) of 6.4 %. © 2025 The Authors.