TY - JOUR
T1 - Data-Driven Analysis of Dominant Factors for High Impact Toughness in High-Strength Titanium Alloys
AU - Lü, Qiyang
AU - Su, Tiejian
AU - Fan, Qunbo
AU - Gong, Haichao
AU - Xu, Shun
AU - Yang, Lin
AU - Pan, Shiwei
AU - Cheng, Xingwang
N1 - Publisher Copyright:
© 2026, Beijing Institute of Technology. All rights reserved.
PY - 2026
Y1 - 2026
N2 - Impact toughness is a critical indicator for the engineering application of titanium alloys, yet it often exhibits a trade-off relationship with strength enhancement. To uncover the dominant factors governing high impact toughness while maintaining strength, a dataset comprising 188 samples of composition, heat treatment processes, and mechanical properties was compiled from publicly available literature. A three-layer ensemble model architecture centered on "expert models" was proposed. This architecture effectively addressed the problem of low accuracy of impact toughness prediction, mostly caused by mixed-type data training, and achieved a coefficient of determination of 0.88. Three dominant factors were identified based on SHAP values, the phase stability ratio (aluminum equivalent/molybdenum equivalent), Mo content, and phase stability ratio's interaction effect with the average atomic radius. Quantitatively, a Mo content trap range (2 – 4) and two optimal design windows for the phase stability ratio and average atomic radius ((2 – 3/1.35 – 1.40) × 10^-10 m, and (0 – 2/1.45 – 1.50) × 10^-10 m) were delineated. Literature validation based on a confusion matrix further verified the clear boundary of these dominant factors.
AB - Impact toughness is a critical indicator for the engineering application of titanium alloys, yet it often exhibits a trade-off relationship with strength enhancement. To uncover the dominant factors governing high impact toughness while maintaining strength, a dataset comprising 188 samples of composition, heat treatment processes, and mechanical properties was compiled from publicly available literature. A three-layer ensemble model architecture centered on "expert models" was proposed. This architecture effectively addressed the problem of low accuracy of impact toughness prediction, mostly caused by mixed-type data training, and achieved a coefficient of determination of 0.88. Three dominant factors were identified based on SHAP values, the phase stability ratio (aluminum equivalent/molybdenum equivalent), Mo content, and phase stability ratio's interaction effect with the average atomic radius. Quantitatively, a Mo content trap range (2 – 4) and two optimal design windows for the phase stability ratio and average atomic radius ((2 – 3/1.35 – 1.40) × 10^-10 m, and (0 – 2/1.45 – 1.50) × 10^-10 m) were delineated. Literature validation based on a confusion matrix further verified the clear boundary of these dominant factors.
KW - dominant factors
KW - impact toughness
KW - machine learning
KW - titanium alloys
UR - https://www.scopus.com/pages/publications/105047686194
U2 - 10.15918/j.tbit1001-0645.2026.049
DO - 10.15918/j.tbit1001-0645.2026.049
M3 - Article
AN - SCOPUS:105047686194
SN - 1001-0645
VL - 46
SP - 906
EP - 917
JO - Beijing Ligong Daxue Xuebao/Transaction of Beijing Institute of Technology
JF - Beijing Ligong Daxue Xuebao/Transaction of Beijing Institute of Technology
IS - 8
ER -