Revolutionizing thermal energy systems: Researchers unveil advanced physics-informed digital twin framework

Shanghai Jiao Tong University Journal Center

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

A closed-loop feedback system between a physical system and its DT models, enabled by MPC

view more 

Credit: Sadegh Ataee & Mehran Ameri.

A new comprehensive review introduces a pioneering exergy-based loss function for Physics-Informed Neural Network-Digital Twins (PINN-DT), promising unprecedented accuracy and real-time optimization for industrial thermal energy systems.

 

Thermal energy systems (TES) are the essential backbone of modern industry, playing a critical role in everything from power plants to advanced manufacturing. However, accurately predicting their performance under complex, real-world conditions has long been a challenge. Traditional simulation methods rely heavily on empirically derived formulas, which often suffer from limited predictive accuracy, narrow operational ranges, and reduced flexibility when handling geometrically complex configurations.

 

A groundbreaking review article published in ENGINEERING Energy systematically examines a powerful solution to these industrial bottlenecks: Physics-Informed Neural Network-Digital Twin (PINN-DT) technology. Conducted by researchers Sadegh Ataee and Mehran Ameri from Shahid Bahonar University of Kerman, the study provides a comprehensive taxonomy for applying PINN-DTs to industrial thermal challenges.

 

To highlight the transformative potential of this technology, the researchers detailed several core advancements:

• Solving Ill-Posed Thermal Problems: PINN-DTs successfully resolve complex, non-linear thermal problems that are completely inaccessible to conventional computational methods.
• Transcending the AI "Black Box": By embedding fundamental physical laws—such as energy conservation and fluid dynamics—directly into the neural network's training process, the system achieves high prediction accuracy and physical interpretability even with scarce or noisy observational data.
• Real-Time Predictive Control: Integrating PINN-DTs with Model Predictive Control (MPC) algorithms allows the digital twin to anticipate future system states, explicitly incorporate operational constraints, and send optimized, real-time control signals back to the physical entity.
• Pioneering Exergy-Informed Loss Functions: The researchers identified the absence of exergy analysis in loss function formulation as a major research gap. They propose a novel physics-informed loss function derived from exergy analysis (combining the first and second laws of thermodynamics) to drastically improve model fidelity and predictive accuracy.
• Broad Industrial Scalability: The PINN-DT framework offers vital decision-making support across diverse sectors, including supercritical CO2 Brayton cycles, smart power grids, food processing refrigeration, and dynamic HVAC control for GPU-centric data centers.

 

"The development of robust physics-informed machine learning frameworks fundamentally depends on embedding appropriate physical principles through carefully designed constraint terms," the authors state.

 

For industries seeking to minimize energy consumption while maximizing output, this systematic review serves as a definitive roadmap for implementing digital-physical synchronization in the Industry 4.0 era.

 

Journal: ENGINEERING Energy

Read the full article for free: https://rdcu.be/frBqf

Cite this article: Ataee, S., Ameri, M. Physics-informed neural network-based digital twins for thermal energy systems: A review of solvability and loss function design. ENG. Energy 20, 10491 (2026). https://doi.org/10.1007/s11708-026-1049-1

---

Journal

ENGINEERING Energy

DOI

10.1007/s11708-026-1049-1 

Method of Research

News article

Article Title

Physics-informed neural network-based digital twins for thermal energy systems: A review of solvability and loss function design