Modeling of ferroelectric properties of KNN-based ceramics

Replacement of highly noxious led zirconate titanate based materials in MEMS applications for environmentally safe materials remains an important task. Complex multicomponent doping of KNN ceramics provides for a significant improvement of their ferroelectric properties. The efficiency of different machine learning (ML) methods in the prediction of piezoelectric coefficient of KNN-based ceramics has been analyzed. ML integration allows one to accelerate the development of new materials and identify latent interdependences between their composition, structure, and properties. To understand the mechanisms of relationship between the composition and properties, we used the Sure Independence Screening and Sparsifying Operator (SISSO) method in combination with regression ML methods, such as Support Vector Regression, Ridge, Lasso, Elastic Net, Xtreme Gradient Boosting, Random Forest, Gradient Boosting Regression, and Extreme Tree Regression. The Gradient Boosting Regression (GBR) method proved to be the most efficient in combination with SISSO. The test sampling accuracy reached 81 %, and GBR validation for experimental data delivered a mean absolute error of 26 pC/N.