EXTENDED ABSTRACT: Potassium sodium niobate (KNN)-based ceramics are one of the most potential environmentfriendly optoelectronic functional ceramics with distinguished transparency and piezoelectric properties. This speech first introduces a database and corresponding machine learning models for doped KNN based on high-throughput density functional theory calculations. By extracting key features, we identify physically meaningful descriptors and analyzed the effects of dopants and vacancy defects on the performance of KNN-based ceramics. Furthermore, by taking the light transmittance of KNN-based transparent ceramics as the target property, we establish a traditional machine learning regression model to predict the transmittance of un-synthesized KNNbased ceramics and forecast their optimal sintering temperature. The predicting results align well with experimental data, achieving mutual validation between machine learning and experimental research. Moreover, using the piezoelectric coefficient d33 as the target property, a descriptor framework with three-layer dataset structure is established. A visualizable and interpretable symbolic regression model is utilized to propose the descriptor for d33. We not only enable rapid estimation of the piezoelectric coefficient before experiments but also provide a reasonable explanation for the high piezoelectricity induced by the coexistence of multiple phases at the interfaces of KNN-based ceramics.
Keywords: potassium sodium niobate; data driven; machine learning; transparent ceramics; piezoelectric ceramics

REFERENCES:
[1] B. Ma, F. Yu, P. Zhou, X. Wu*, C. Zhao, C. Lin, M. Gao, T. Lin, B. Sa*, J. Mater. Inf., 3, (2023) 13.
[2] B. Ma, X. Wu*, C. Zhao, C. Lin, M. Gao, B. Sa*, Z. Sun*, npj Comput. Mater., 9, (2023) 229.