Application of neural networks to high-throughput screening of ferroelectric materials

Abstract

Combinatorial & high-throughput experimentation is a kind of R&D trend in the 21st century. Rapid progress of high-throughput experimental tools requires the development of data mining software. In this work, two different neural networks were applied to HTS (High-throughput screening) of ferroelectric materials including $Bi_{4-x}La_xTi_3O_{12}$ (BLT) and $Bi_{4-x}Ce_xTi_3O_{12}$ (BCT). And we can find that the necessary calculation time of generalized regression neural network (GRNN) models was much shorter, but multilayer perceptron (MLP) models were more accurate than GRNN models. Once a neural network model with high accuracy was established, we could find the optimal reaction conditions with the best characteristics. The highest gradient value obtained by the neural network model is higher than the maximum value found by experiments in the case of both BCT and BLT.