Keimyung University team publishes AI model for predicting hydrogen output from plastic waste

A research team at Keimyung University has published a study on optimizing plastic recycling supply chain networks in the international journal Engineering Applications of Artificial Intelligence (ISSN: 0952-1976, top 2.79% in JCR), which covers environmental and sustainability topics, the university announced Monday.

The paper is titled "Physics-informed surrogate modeling for hydrogen production in plastic waste gasification."

The study presents a physics-informed neural network (PINN) framework that integrates physical knowledge into the learning process to predict hydrogen output from plastic waste gasification.

To address the limitations of conventional artificial neural networks (ANN), the team developed three PINN models — kinetic, equilibrium and integrated — by incorporating reaction kinetics and thermodynamic equilibrium conditions into the loss function as soft constraints, then compared their performance.

The results showed that conventional ANN models suffered a significant drop in predictive accuracy under data-scarce conditions, while the PINN models maintained high accuracy even when trained on just 1 percent of the full dataset.

Gradient-based residual analysis further confirmed that the models' predictions aligned with thermodynamic and reaction-kinetic trends.

"This is a proof-of-concept study based on simplified process data, but it demonstrates the potential of surrogate modeling combined with physical knowledge," said Professor Ahn Yu-chan, the corresponding author. "It could be extended in the future to real-time optimization and control of hydrogen production processes based on plastic waste."

Son Jun-hyeok served as first author and Jeong Hyo-jin as second author, both master's researchers in the Department of Chemical Engineering at Keimyung University. The research was supported by the National Research Foundation of Korea's Basic Research Program and its Master's Student Research Incentive Support Program.