Composition Design of a Novel High-Temperature Titanium Alloy Based on Data Augmentation Machine Learning.

Area of Science:

• Materials Science and Engineering
• Metallurgy
• Computational Materials Design

Background:

• High-temperature titanium alloys are vital for aerospace and defense, enabling lighter and more resilient components.
• Traditional alloys with numerous elements complicate manufacturing processes like casting and deformation.
• There is an urgent need for simplified, low-component high-temperature titanium alloys suitable for hot processing.

Purpose of the Study:

• To accelerate the development of novel, low-component high-temperature titanium alloys using machine learning.
• To design and validate a new quinary high-temperature titanium alloy with improved manufacturability and performance.

Main Methods:

• Employed data augmentation (Gaussian noise) to enhance the generalization capabilities of machine learning models.
• Evaluated four machine learning models (XGBoost, RF, AdaBoost, Lasso) for predicting alloy properties.
• Designed a novel quinary alloy (Ti-7.2Al-1.8Mo-2.0Nb-0.4Si) based on the best-performing XGBoost model.

Main Results:

• The XGBoost model achieved high prediction accuracy (R²=0.94) after data augmentation.
• The designed quinary alloy exhibited a high ultimate tensile strength (UTS) of 629 MPa at 600 °C, closely matching predictions.
• The novel alloy demonstrated competitive high-temperature mechanical properties compared to traditional six-element alloys, with fewer components.

Conclusions:

• Machine learning, enhanced by data augmentation, effectively aids in designing advanced, low-component high-temperature titanium alloys.
• The novel quinary alloy offers a promising alternative for aerospace applications due to its balanced properties and simplified composition.
• Microstructure analysis confirmed an α+β type alloy with a Widmanstätten structure, exhibiting mixed fracture behavior.