Applications of physics-informed deep-learning in polymer rheology: Prediction and optimization for rheological molding process through integrated physical and data-driven learning

The review explores the application of Physics-Informed Deep Learning (PIDL) in polymer rheology, highlighting its potential to address limitations of traditional rheological models. By integrating physical prior knowledge with data-driven methods, PIDL enhances the accuracy and reliability of fluid behavior predictions, fostering a more refined, intelligent, and interdisciplinary development in rheological research. The article outlines the fundamental principles, common architectures, and evaluation methods of PIDL, and presents practical examples of its applications in polymer rheology, including constitutive relationship modeling, fluid behavior prediction, experimental data analysis, multiscale simulation and optimization, and process parameter optimization. Challenges faced by PIDL in polymer rheology, such as data quality, model generalization capabilities, and multiscale problem-solving, are also discussed, along with current solutions.