#South Korea #AI Technology #Hanbat University #Chungcheong Province #Thermal Energy

Innovative AI Technology for Thermal Energy Storage

As the global demand for energy continues to rise, decarbonizing the building sector is becoming increasingly vital. Researchers at Hanbat National University in Chungcheong Province, South Korea, have made a significant leap forward by introducing a physics-informed AI framework designed to optimize thermal energy storage systems.

The Importance of Thermal Energy Storage

Decarbonizing processes within the global building sector necessitates the development of innovative solutions to optimize heating and cooling demands. One of the promising technologies leading this change is Latent Heat Thermal Energy Storage (LHTES) systems, which utilize phase change materials. These materials not only have high thermal energy storage density but also maintain a near-constant temperature during the energy release process. Studies have indicated that LHTES systems can yield energy savings of up to 45%.

Overcoming Optimization Challenges

Despite their advantages, the optimization of LHTES systems has remained problematic, largely due to the limitations associated with traditional methods. Laboratory experiments typically operate at a smaller scale, and while Computational Fluid Dynamics (CFD) can model complex heat transfer and fluid flow, the cost and time required for simulations makes extensive design optimization cumbersome.

A New Approach with Physics-informed Neural Networks

In a groundbreaking study published on July 30, 2026, the research team headed by Assistant Professor Joo Hyun Moon from the Department of Building Systems Engineering unveiled their innovative physics-informed neural network (PINN) framework, aimed at rectifying the slow optimization processes. This framework integrates physical laws directly into an AI model, facilitating the rapid exploration of potential LHTES designs, thereby bypassing traditional simulation challenges.

Dr. Moon explained, "The model simulates the heat-release and solidification behavior of materials like wax, which function as thermal batteries. With this AI-based framework, we can efficiently examine and optimize thousands of design variations."

Creating a High-Fidelity Dataset

The researchers took a methodical approach to develop their PINN. By establishing a laboratory-scale LHTES setup, they validated their CFD model using empirical measurements, creating a robust dataset consisting of 15 high-fidelity simulations. This dataset was crucial for training the PINN, which incorporates a zero-dimensional physical model allowing it to learn specific heat-transfer coefficients tailored to various cases.

Furthermore, they employed a response surface model to capture geometric effects, thus forming a digital twin capable of predicting untested designs and conditions within a defined range. This digital rendition couples with Non-dominated Sorting Genetic Algorithm II (NSGA-II), optimizing designs according to key parameters: maximizing discharged heat and average power while minimizing pumping costs.

Promising Results and Wider Implications

The numerical experiments conducted through the PINN demonstrated its ability to accurately replicate CFD predictions, showcasing efficient, automated design optimization. The optimized designs not only achieved performance comparable to the best baseline cases but also succeeded in significantly reducing pumping power, with flatter pipe designs proving advantageous.

Beyond the realm of buildings, this physics-informed AI method has significant implications for various fields, including thermal management in electric vehicle batteries, data center cooling, cold-chain logistics, and solar thermal systems. Prior findings suggest enhanced control of latent heat storage could lower electricity bills by over 70%.

A Step Towards Sustainable Engineering

Dr. Moon emphasizes the transformative potential of their approach: "This framework enables engineers to autonomously explore the vast landscape of LHTES design, leading to the development of more energy-efficient systems that subsequently reduce energy consumption and carbon emissions."

The research carries significant weight in the ongoing push for sustainability in energy systems, illustrating the application of novel technologies to tackle some of the most pressing challenges of our time. For further detail, the research can be referenced in the Journal of Energy Storage, volume 167.

References

• - Original paper: Physics-informed neural networks for multi-objective design optimization of latent heat thermal energy storage systems
• - Journal: Journal of Energy Storage
• - DOI: 10.1016/j.est.2026.122514

To learn more about Hanbat National University and its research initiatives, visit their official website.