.png){kind=icon}
.png){kind=icon}
Return
Revealing loss mechanisms through interpretable machine learning and accelerated discovery of ultra-low-loss dielectric ceramics in the Li2TiO3-Li3NbO4-MgO system
.png){kind=icon}
PRE
AI
X
L
J
J
M
Y
Z
W
B
H
L
DOI：10.1016/j.jmst.2025.12.025.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Developed an ensemble machine leaning (ML) model (R² = 0.8285) with SHAP and permutation importance, and successfully identified that internal strain and configurational entropy as dominant factors reducing dielectric loss (high Q × f) from twelve descriptors. • Used Gaussian Process Regression (GPR) with Expected Improvement to identify optimal composition x = 0.85 in only 6 experiments in the (1-x)Li3MgNbO5-xLi2TiO3 system (out of 21). • Mapped complex interactions between structure and dielectric loss using minimal yet physically grounded descriptors. Demonstrated that peak Q × f results from a synergy of medium-level entropy, low internal strain, and high densification, moving beyond single-factor models to a validated multivariable explanation. • All model predictions were rigorously validated through experimental synthesis and characterization, and the optimal composition (x=0.85) delivered excellent performance with εr = 16.17, Q × f = 104,300 GHz at 9.2 GHz, and τf = -3.47 ppm/°C.
Journal
IF:
14.3
Papers:
8.0K
Citations:
6.1W
Organization
Cited Papers