Abstract
In recent years, the rapid advancement of materials science has underscored the significance and complexity of understanding and optimizing the properties of piezoelectric materials. Traditional analytical methods have encountered difficulties in addressing the challenges posed by variable material compositions and intricate processing conditions. However, deep neural networks (DNNs) present new opportunities in this field due to their robust capabilities in nonlinear data processing and pattern recognition. This study employs DNN to model the relationship between the composition and processing conditions of piezoelectric materials and their electrical properties. Through quenching experiments, it was observed that the electrical properties of Na(0.5)Bi(0.5)TiO(3)-based ceramics are positively correlated with local polar heterogeneity. This phenomenon was further validated by manipulating the local polar heterogeneity through Bi or Sr nonstoichiometric modifications. Subsequently, a DNN model characterized by local polarity heterogeneity was developed. After training on extensive high-quality experimental data, the DNN model demonstrated exceptional prediction accuracy and effectively elucidated the underlying complex mechanisms governing material properties. The findings indicate that DNN can accelerate the development and optimization of new piezoelectric materials, provide guidance for experimental design, and significantly enhance research and development efficiency. The successful application of this research strongly supports the expansion of deep learning technology in materials science.