Newest Research Published in EAAI: Physics-Informed AI for Hydrocracking Kinetics Modeling
We are excited to share one of our newest research publications, recently published in Engineering Applications of Artificial Intelligence (EAAI). This work was led by Souvik, whose dedication, initiative, and hard work played a key role in bringing this study to completion. Congratulations to Souvik on this excellent achievement.
The paper, titled “Physics-Informed Neural Ordinary Differential Equations for Hydrocracking Kinetics Modeling and System Identification,” presents a new framework that integrates physical laws directly into neural ordinary differential equations for more accurate and interpretable kinetic modeling. By embedding reaction stoichiometry, rate positivity, and Arrhenius temperature dependence into the learning architecture, the proposed approach helps bridge the gap between mechanistic modeling and data-driven AI methods.
Our results show that progressively incorporating physical constraints can significantly improve prediction accuracy, with 40–55% RMSE reduction compared with an unconstrained baseline. The framework also demonstrated strong generalization beyond synthetic datasets and was able to recover physically meaningful kinetic parameters with statistical robustness. This study highlights the potential of physics-informed machine learning as a reliable grey-box modeling tool for refinery process optimization and other complex reaction systems.
Read the article here:
https://www.sciencedirect.com/science/article/pii/S0952197626000898
