Predictive modelling of thermal conductivity in single-material nanofluids: a novel approach

Bulletin of the National Research Centre

Research

Ekene Onyiriuka¹ 

[1] School of Mechanical Engineering, University of Leeds, LS2 9JT, Leeds, UK;

关键词: Single material;    Nanofluids;    Modelling;    Predict;    Thermal conductivity;    Feature selection;   

DOI  :  10.1186/s42269-023-01115-9

 received in 2023-06-27, accepted in 2023-09-11,  发布年份 2023

来源: Springer

【 摘 要 】

BackgroundThis research introduces a novel approach for modelling single-material nanofluids, considering the constituents and characteristics of the fluids under investigation. The primary focus of this study was to develop models for predicting the thermal conductivity of nanofluids using a range of machine learning algorithms, including ensembles, trees, neural networks, linear regression, Gaussian process regressors, and support vector machines.The main body of the abstractTo identify the most relevant features for accurate thermal conductivity prediction, the study compared the performance of established feature selection algorithms, such as minimum redundancy maximum relevance, Ftest, and RReliefF, a newly proposed feature selection algorithm. The novel algorithm eliminated features lacking direct implications for fluid thermal conductivity. The selected features included temperature as a thermal property of the fluid itself, multiphase features such as volume fraction and particle size, and material features including nanoparticle material and base fluid material, which could be fixed based on any two intensive properties. Statistical methods were employed to select the features accordingly.ResultsThe results demonstrated that the novel feature selection algorithm outperformed the established approaches in predicting the thermal conductivity of nanofluids. The models were evaluated using fivefold cross-validation, and the best model was the model based on the proposed feature selection algorithm that exhibited a root-mean-squared error of validation of 1.83 and an R-squared value of 0.94 on validation set. The model achieved a root-mean-squared error of 1.46 and an R-squared value of 0.97 for the test set.ConclusionsThe developed predictive model holds practical significance by enabling nanofluids' numerical study and optimisation before their creation. This model facilitates the customisation of conventional fluids to attain desired fluid properties, particularly their thermal properties. Additionally, the model permits the exploration of numerous nanofluid variations based on permutations of their features. Consequently, this research contributes valuable insights to the design and optimisation of nanofluid systems, advancing our understanding and application of thermal conductivity in nanofluids and introducing a novel and methodological approach for feature selection in machine learning.

【 授权许可】

CC BY   
© National Research Centre 2023

【 图 表 】

13690_2023_1170_Article_IEq79.gif

Fig. 1

Fig. 2

40360_2023_687_Article_IEq1.gif

13690_2023_1170_Article_IEq203.gif

13690_2023_1170_Article_IEq201.gif

Fig. 2

13690_2023_1170_Article_IEq198.gif

12888_2023_5142_Article_IEq5.gif

12888_2023_5142_Article_IEq3.gif

13690_2023_1170_Article_IEq188.gif

13690_2023_1170_Article_IEq75.gif

13690_2023_1170_Article_IEq187.gif

13690_2023_1170_Article_IEq74.gif

Fig. 5

Fig.3

Fig. 3