Page 238 - Read Online

Liu et al. J. Mater. Inf. 2026, 6, 18 Page 3 of 29

Figure 1. Roadmap of knowledge-enabled data-driven smart design of ultra-strong ductile near-α titanium alloys under extreme conditions.
R : The coefficient of determination; RMSE: root mean square error; Bagging: bootstrap aggregating; RF: random forest; GBRT: gradient
2
boosting regression tree; ExtraTree: extremely randomized tree; AdaBoost: adaptive boosting; LGBM: light gradient boosting machine;
XGBoost: eXtreme gradient boosting; KNN: K-nearest neighbor; DT: decision tree; ANN: artificial neural network; SVR: support vector
regression; NSGA-II: nondominated sorting genetic algorithm II; TA: titanium alpha; TC: titanium composite; ASB: adiabatic shear band.

MATERIALS AND METHODS
A knowledge-enabled data-driven multi-objective optimization framework
This work aims to establish a domain-knowledge-enhanced data-driven multi-objective optimization
framework for the design of titanium alloys under extreme conditions. Here, “domain knowledge” refers to
the use of physically motivated descriptors and metallurgical constraints in both feature construction and
optimization, rather than purely empirical feature selection. The research roadmap is illustrated in Figure 1.
First, a mechanical property database is established, encompassing 10 alloying elements, strain rate, dynamic
compressive strength, and fracture strain. All data in this work are either derived from or correspond to
high-strain-rate impact loading conditions. Meanwhile, a material feature pool is constructed, including
intrinsic physical properties, electronic and surface characteristics, mechanical properties, and structural
features. Based on this feature pool, key feature parameters influencing the strength and ductility of titanium
alloys are identified by employing correlation analysis, feature importance analysis, and feature subset search
methods. Second, by evaluating twelve ML models, the optimal models for strength and ductility prediction
tasks are identified to validate the fitting performance of the selected key feature parameters. By coupling
composition with these key feature parameters, the optimal ML models for strength and ductility are
combined with multi-objective optimization algorithms to design near α-titanium alloys satisfying [Mo]eq ≤
2.5 and [Al]eq ≤ 8.5. Finally, the titanium alloys within the designed Pareto front subset are validated by
using constitutive equations and thermodynamic calculations via Thermo-Calc to demonstrate the
application potential of the new alloys.

Analyze the database and determine key characteristic parameters
Due to the substantial diversity and compositional complexity of titanium alloys, available datasets are often
limited in scope, leading to underrepresentation of critical performance features and hindering the accurate
development of ML models [50,51] . To overcome this limitation, a comprehensive database of titanium alloy
mechanical properties was established, comprising 175 data points for dynamic compressive strength and