Prediction of dielectric properties of ferroelectric materials based on deep neural networks.

Ferroelectric materials have emerged as significant research hotspots within the field of materials science and engineering, primarily due to their unique electrical properties. However, the electrical characteristics of these materials are influenced by various factors, including material composition, microstructure, and preparation processes, which introduce considerable randomness and uncertainty. Traditional experimental and simulation methods are often insufficient for capturing these complex interactions, thereby hindering the prediction and optimization of material performance. This paper presents a novel approach for predicting the electrical properties of ferroelectric materials by utilizing deep neural networks (DNNs). The DNNs are trained using experimental data and serve as a proxy model to predict critical electrical properties, such as the dielectric constant and dielectric peak. The (1-)NaBiTiO-SrZrO ceramics were synthesized via the solid-state reaction method, and both the phase structure and electrical properties of NBT-SZ were measured. The experimental results indicate that the DNN model effectively captures the intricate influences of factors such as material composition, preparation processes, and microstructure on electrical properties. The discrepancy between predicted values and experimental results remains within an acceptable range. By comparing the absolute error (<5) between measured and predicted data, alongside evaluation metrics such as MAPE, SMAPE, and ², the practicality and reliability of the DNN model are substantiated. The strong performance of this model not only accelerates the development of new materials but also enhances the optimization of the performance of existing materials.