Accelerating the design of multivariate Nb-Si based alloys with excellent strength and toughness matching through machine learning

Abstract

The applications of Nb-Si-based alloys for new-generation engines are heavily influenced by their strength and toughness. Therefore, accurately and effectively predicting fracture toughness and compressive strength is crucial for accelerating the research and development of Nb-Si-based alloys. To achieve this, this work established a database based on Nb-Si-based alloy data fabricated by arc melting. It contains 259 samples with fracture toughness, of which 124 groups have compressive strength information. Machine learning methods were then employed to construct the prediction models between alloying elements, fracture toughness, and compressive strength. The result indicated that Random Forest (RF) and K-Nearest Neighbors (KNN) regression models were suitable for predicting Nb-Si-based alloys' fracture toughness and compressive strength, respectively, and the generalization ability was proved by the Nb-16Si-20Ti-xB4C (x= 0, 0.25, 0.50, 0.75, 1.0 at%) alloys fabricated in previous work. Therefore, the RF and KNN prediction models were used for the alloying design of five-element Nb-(16−20) Si-(20−24) Ti-(0−6) Al-(0−6) Cr alloys (at%) with compositions ranging from hypoeutectic to hypereutectic zones. Through these models and high-throughput optimization algorithms, Nb-16Si-24Ti-3Al-6Cr alloy with excellent strength and toughness matching was selected as the optimal material from 1225 candidate materials. Experimental results showed that the actual fracture toughness (11.9 MPa·m1/2) and compressive strength (2333.5 MPa) of Nb-16Si-24Ti-3Al-6Cr alloy matched the predicted values with less than 15 % errors.