Abstract:

BiScO₃-PbTiO₃ (BSPT)-based piezoelectric ceramics have emerged as one of the most promising candidates for high-temperature piezoelectric applications above 350 ℃ due to their high Curie temperature (T_C) and large piezoelectric coefficient (d₃₃). However, the conventional trial-and-error approach is inefficient for the rapid design of high-temperature piezoelectric ceramics across a wide compositional space. In this work, we developed a machine learning model trained on a small dataset and integrated it with experimental knowledge to accelerate the design of BSPT-based ceramics with simultaneously large d₃₃ and high T_C. Guided by the trained model, we designed Ga-W ion-pair co-doped 0.36BiScO₃-0.64PbTi_1-x(Ga_2/3W_1/3)_xO₃ (BSPTGW1000x) ceramics. The results demonstrated that this doping strategy significantly modified the lattice distortion and domain structures of BSPT-based ceramics, leading to enhanced piezoelectric performance. Among the compositions, BSPTGW10 (x=0.010) exhibited the best overall properties (d₃₃=525 pC/N, T_C=423 ℃), which were in close agreement with the predicted values. Moreover, its piezoelectric coefficient variation remained within ±15% up to 365 ℃, indicating excellent thermal stability. This study not only provides an effective approach for the rapid discovery of BSPT-based ceramics with dual high-performance characteristics, but also yields a promising piezoelectric ceramic material suitable for high-temperature applications. The datasets in this article are listed in Science Data Bank at https://www.doi.org/10.57760/sciencedb.27980.

Key words: machine learning, BiScO₃-PbTiO₃, high-temperature piezoelectric ceramic, data-driven design, ion- pair co-doping