Experimental analysis and deep learning prediction of nanofluid thermophysical properties under time-varying conditions

Jie Zhou, Xin Wang, Jie Xu, Zhenming Shi,
Experimental analysis and deep learning prediction of nanofluid thermophysical properties under time-varying conditions,
Construction and Building Materials,
Volume 489,
2025,
142226,
ISSN 0950-0618,
https://doi.org/10.1016/j.conbuildmat.2025.142226.
(https://www.sciencedirect.com/science/article/pii/S0950061825023773)
Abstract: As a stable and reliable renewable energy source, geothermal energy can effectively meet the urgent demand for low-carbon energy supply in the building sector amid the global energy transition. Enhancing the thermal conductivity of heat transfer fluids is one of the effective means for the efficient extraction of geothermal resources. The addition of nanoscale metal oxides can significantly improve the thermal conductivity of fluids. This study conducts time-varying experiments on the thermal conductivity and viscosity of nanofluids under multiple parameters to optimize suitable parameters for thermally conductive nanofluids. A time-varying prediction model for thermal conductivity is constructed using deep learning algorithms. The results indicate that smaller particle sizes, higher temperatures, and moderate mass fractions are beneficial for enhancing the thermal conductivity of nanofluids; however, the thermal conductivity decreases over time due to particle sedimentation and agglomeration. 20 nm CuO nanoparticles demonstrate good thermal conduction performance within a mass fraction range of 0.5–1 % at temperatures of 5°C and 30°C. The enhancement mechanism of nanofluid thermal conductivity involves changes in the base fluid structure, micro convection effects, and thermal conduction bridging, which are influenced by particle distribution, movement, and agglomeration. Prolonged standing can reduce both thermal conductivity performance and stability. A deep learning framework integrating Long Short-Term Memory (LSTM), Deep Neural Networks (DNN), K-Nearest Neighbors (KNN), and Gaussian Process Regression (GPR) models demonstrates that LSTM outperforms others, achieving an R² of 0.9955 and a mean absolute percentage deviation of 0.29 %, effectively capturing temporal dynamics. The research findings provide a basis for the application of nanofluids in geothermal heat exchangers to improve heat exchange efficiency and offer theoretical support for the selection of nanofluid parameters and the prediction of thermal conductivity.
Keywords: Nanofluid; Thermal Conductivity; Viscosity; Time-Varying Properties; Deep Learning Models