Key Takeaways
- Researchers at Hanbat National University in South Korea developed a hybrid physics-informed neural network (PINN) framework to speed design optimization for latent heat thermal energy storage (LHTES) systems.
- The framework used just 15 high-fidelity simulations to train a fast model, then applied a genetic algorithm to balance discharged heat, power output and pumping power.
- The optimized design matched baseline thermal performance while cutting pumping power, with flatter heat-exchanger pipe geometries outperforming conventional round designs.
- Lead researcher Joo Hyun Moon said the framework “overcom[es] the limitations of slow and computationally expensive conventional simulations.”
- The study was published online in the Journal of Energy Storage on May 5, 2026, and appeared in print in Volume 167 on July 30, 2026.
Hanbat National University Builds AI Model for Thermal Storage Design
Researchers at Hanbat National University in South Korea developed a hybrid physics-informed neural network (PINN) framework to speed up design optimization for latent heat thermal energy storage (LHTES) systems used in buildings. The research addresses a bottleneck in designing these systems: physical experiments only work at lab scale, while computational fluid dynamics simulations are too slow and expensive to test large numbers of design variations.
Physics-Informed AI Works as a Fast Digital Twin
The PINN framework embeds the governing laws of physics directly into the AI model rather than relying on data alone, allowing it to function as a fast digital twin capable of evaluating thousands of potential LHTES designs while preserving physical accuracy. The team trained the model using just 15 high-fidelity computational fluid dynamics simulations, then applied a genetic algorithm known as NSGA-II to search for designs that simultaneously maximized discharged heat and average power output while minimizing pumping power.
“Our framework integrates the governing laws of physics into an AI model, overcoming the limitations of slow and computationally expensive conventional simulations,” said Joo Hyun Moon, Assistant Professor in the Department of Building Systems Engineering at Hanbat National University, who led the research team. The team said the approach “enables autonomous exploration of the continuous LHTES design space,” helping engineers develop more efficient systems while reducing energy use and emissions.
Study Targets Buildings’ Decarbonization Push
Latent heat thermal energy storage systems store and release heat for building heating and cooling, and are seen as one path to cutting the energy demands of the building sector. The resulting optimized design matched baseline thermal performance while substantially cutting pumping power, and the study found that flatter heat-exchanger pipe geometries outperformed conventional round designs. The researchers said the approach could extend to other applications, including electric-vehicle battery thermal management, data center cooling, cold-chain logistics and solar thermal systems.
Hanbat National University Research Published in Journal of Energy Storage
The study was published online in the Journal of Energy Storage on May 5, 2026, and appeared in print in Volume 167 on July 30, 2026.
