Researchers have developed a modified constitutive model and artificial neural network frameworks to predict the flow behavior of a specific titanium alloy during hot deformation. The study, published in the Journal of Materials Science, focuses on the Ti–0.4O–4Fe alloy, a material where accurate prediction of mechanical response is essential for optimizing thermomechanical processing. The team compared a modified Fields–Backofen (MFB) constitutive model against the original Fields–Backofen (FB) model and two artificial neural network (ANN) models.
The experimental foundation of the study involved isothermal compression tests conducted under controlled thermomechanical conditions. The tests were performed at temperatures ranging from 700 to 850 degrees Celsius and strain rates from 0.001 to 0.1 per second. The microstructural evolution observed during these tests was strongly governed by the specific thermomechanical conditions applied. Elevated temperatures and lower strain rates favored dynamic restoration and microstructural homogenization, while higher strain rates promoted deformation heterogeneity and the development of pronounced substructures.
To ensure rigorous validation, the researchers reserved a 25% holdout dataset consisting of data from 850 degrees Celsius across all strain rates. The modified Fields–Backofen model substantially outperformed the original FB model on this withheld validation data. The MFB model achieved a correlation coefficient of 0.988 and a mean error of 6.772%, whereas the original FB model yielded a correlation coefficient of 0.959 and a significantly higher mean error of 29.89%. On the calibration data, the MFB model achieved a correlation coefficient of 0.993 and a mean error of −0.291%, indicating a strong fit to the training set.
The artificial neural network models demonstrated superior predictive accuracy compared to the constitutive models. The LM-ANN achieved the lowest RMSE and MAE among all tested models. Specifically, the LM-ANN recorded a correlation coefficient of 0.99998, a mean absolute error of 0.137 MPa, and a root mean square error of 0.226 MPa. The mean error for the LM-ANN was 0.0015%, with individual errors ranging from −1.574% to 5.175%.
For comparative evaluation, each of the four models was implemented in finite element (FE) simulations. The resulting predictions were compared with the same experimental reference data used for model calibration and validation. The study explicitly states that FE was used strictly as a common benchmarking platform rather than as an independent validation method or to predict deformation outcomes. The integrated experimental–constitutive–ANN-FE framework provides a reliable strategy for flow stress prediction and microstructural control.
Subject of Research: Constitutive and artificial neural network modeling of hot deformation in Ti–0.4O–4Fe alloy
Article Title: Flow behavior and microstructure evolution of Ti–0.4O–4Fe alloy during hot deformation: constitutive and artificial neural network modeling
Article References: Youssef, S. S., Wang, X., Shi, X., Gowid, S., Shokry, A., & Shen, J. (2026). Flow behavior and microstructure evolution of Ti–0.4O–4Fe alloy during hot deformation: constitutive and artificial neural network modeling. Journal of Materials Science. https://doi.org/10.1007/s10853-026-13971-7
Image Credits: AI Generated
DOI: 10.1007/s10853-026-13971-7
Keywords: titanium alloy, hot deformation, constitutive modeling, artificial neural network, flow stress, microstructure, materials science, finite element simulation, Flow, behavior, evolution, alloy
News Source: Neil Sanderson. (November 9, 2026). Neural Networks Improve Hot Deformation Prediction for Titanium Alloy. Scienmag.



