Acceleration of optimizing the dielectric properties of (Ca0.7Nd0.2)1−x(Li0.5Nd0.5)xTi1−y(Mg1/3Nb2/3)yO3 with the aid of machine learning

Abstract The three key properties required for microwave dielectric ceramics are suitable dielectric constant ( ε r ), high quality factor ( Q × f ), and near‐zero temperature coefficient of resonant frequency ( τ f ). Due to the intricate coupling relationship among these three properties, regulating one often leads to the deterioration of the remaining two. In this study, the synergetic regulation of dielectric properties of Ca 0.7 Nd 0.2 TiO 3 was investigated to reduce τ f to near‐zero while maintaining its high ε r by A‐ and B‐site substitutions. To avoid the seesaw effect and improve the efficiency of composition optimization, machine learning method was introduced in this study. First, the models for predicting dielectric properties were fitted based on a small amount of high‐quality data gathered via a uniform experimental design. Then the dielectric properties of 1037 compositions of (Ca 0.7 Nd 0.2 ) 1− x (Li 0.5 Nd 0.5 ) x Ti 1− y (Mg 1/3 Nb 2/3 ) y O 3 were predicted, from which 69 microwave dielectric ceramics with near‐zero τ f and adjustable ε r of 95–125 were quickly picked out, and 5 of them were proved by experiments with very high prediction accuracy. This work was finished within a period of 1 month, which proves the obvious acceleration of composition designing process of microwave dielectric ceramics with the aid of machine learning.