Transformer-Driven Inverse Learning for AI-Powered Ceramic Material Innovation With Advanced Data Preprocessing

In the advanced landscape of materials science, particularly in the development of ceramic materials, artificial intelligence (AI) emerged as a transformative tool for accelerating innovation. This study proposed a comprehensive analysis of the Transformer-based Inverse Learning model to optimize component and process recommendations. K-Nearest Neighbors (KNN) imputation was first applied, improving data accuracy and completeness to address data gaps. Subsequently, Variational Autoencoders (VAE) were used for data augmentation, enriching the dataset's diversity. The Transformer model, leveraging this enhanced data, demonstrated strong predictive performance, achieving an R2 score of 0.966 for component analysis and an outstanding R2 score of 0.982 for process analysis in Barium Titanate (BaTiO3) material data. These results show the effectiveness of combining imputation, augmentation, and advanced AI modeling in capturing complex material properties. The study highlights the potential of AI-driven methodologies to significantly improve prediction accuracy in material discovery, offering valuable insights for developing future ceramic materials.