Accelerated design of AgNbO3-based ceramics with high energy storage performance via machine learning

Abstract

Silver niobate based lead-free antiferroelectric (AFE) ceramics exhibit tremendous potential on energy storage applications, but the composition space is too huge for experimental investigation. In this work, we employed machine learning method to accelerate the development of high performance AgNbO3-based AFE ceramics, by using the component system (Ag1−x−3y-4z-3m-3n-3tNaxLayCezNdmSmnTbt)(Nb1-sTas)O3 as a representative, in which the supporting vector regression with radial basis function kernel model with lowest cross-validation error is assigned as the preferred machine learning model. Based on the predicted outcomes, the highest recoverable energy storage density of 7.0 J/cm3 was successfully achieved in (Ag0.94Sm0.02)(Nb0.6Ta0.4)O3 ceramic in experiment, which was close to its prediction value of 6.76 ± 0.55 J/cm3, indicating good reliability of machine learning technique. Moreover, the (Ag0.94Sm0.02)(Nb0.6Ta0.4)O3 ceramic exhibited excellent temperature, frequency and cycling stabilities, possessing outstanding practical application prospective. This study exhibits a paradigm for speedily achieving good energy storage performance in AgNbO3-based and other related materials through the employment of machine learning.