PISA-Net: A physics-informed structure-aware neural network for multiphysics field reconstruction in liquid cooling systems

Zhang, Chaobo, Jiang, Lizhe, Yang, Zhao, Yan, Kun and Yan, Jun 2026. PISA-Net: A physics-informed structure-aware neural network for multiphysics field reconstruction in liquid cooling systems. Applied Thermal Engineering 286 , 129347. 10.1016/j.applthermaleng.2025.129347

Full text not available from this repository.

Abstract

Efficient thermal management in liquid cooling systems relies heavily on the accurate reconstruction of temperature and velocity fields. However, obtaining full-field information under sparse sensor deployment remains a critical challenge. To address this issue, this study proposes a Physics-Informed Structure-Aware Network (PISA-Net) for adaptive and high-fidelity reconstruction of coupled thermal-fluid fields in liquid-cooled environments with limited measurements. The proposed framework integrates sparse temperature and velocity data with geometric information of heat sources and flow channels, enabling structure-aware representation of varying thermal configurations. A physics-informed loss term, derived from the steady-state energy conservation equation, is incorporated to enforce physical consistency during training. This hybrid learning strategy effectively combines data-driven approximation with physical constraints, improving both predictive accuracy and generalizability. Numerical validation on a representative cold plate configuration demonstrates that PISA-Net achieves a normalized mean absolute error of 0.98% for temperature and velocity field reconstruction using only eight sensor measurements. In addition, the physics residual, quantified by the energy equation deviation, is reduced by approximately 80% compared to purely data-driven models. These results highlight the potential of PISA-Net as a robust and interpretable approach for real-time field reconstruction, anomaly detection, and sensor optimization in complex thermal-fluid systems.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	1359-4311
Date of Acceptance:	1 December 2025
Last Modified:	15 Dec 2025 14:00
URI:	https://orca.cardiff.ac.uk/id/eprint/183239

Actions (repository staff only)

Edit Item