Unlocking compositional design and mechanism for mechanical/thermal properties of high‐entropy rare‐earth disilicates

Abstract Designing effective thermal/environmental barrier coatings (T/EBCs) relies on multifunctional optimization through incorporating multiple rare earth (RE) elements into β‐ and γ‐RE 2 Si 2 O 7 . However, the vast compositional space poses significant challenges for the effective selection of multi‐component RE disilicates [( n RE xi ) 2 Si 2 O 7 ]. This study develops an autoencoder‐based multi‐task domain adaptation model to predict the mechanical and thermal properties of ( n RE xi ) 2 Si 2 O 7 . By leveraging self‐supervised learning on unlabeled data, the model reduces feature distribution discrepancies between source and target domains compared to previous models, resulting in high prediction accuracy and generalizability for both equimolar and non‐equimolar compositions without the need for labels. Model visualization identifies lattice constant c and average ionic radius of RE () as key factors. ( n RE xi ) 2 Si 2 O 7 with smaller values of c and shows reduced apparent bulk coefficient of thermal expansion, lower elastic constants, and increased minimum thermal conductivity. Guided by predictions, novel (Er 1/4 Y 1/4 Lu 1/4 Yb 1/4 ) 2 Si 2 O 7 and (Er 1/6 Tm 1/6 Y 1/15 Gd 1/15 Lu 4/15 Yb 4/15 ) 2 Si 2 O 7 are uncovered as high potential for T/EBC applications, whose superior elastic properties, enhanced damage tolerance, low thermal conductivity/residual stresses, and ceramic matrix composite‐matched thermal expansion are demonstrated by experiments with a small error (<10%) versus predictions. This deep learning framework integrates physical mechanisms into material design, accelerating the development of high‐performance ( n RE xi ) 2 Si 2 O 7 and providing valuable guidance for understanding the properties and mechanisms of T/EBC materials.