Explainable machine learning techniques for hybrid nanofluids transport characteristics: an evaluation of shapley additive and local interpretable model-agnostic explanations

• Main Authors: Kanti P.K., PrabhakarSharma, Wanatasanappan V.V., Said N.M.
• Other Authors: 57216493630
• Format: Article
• Published: Springer Science and Business Media B.V. 2025
• Subjects: Deep neural networks; Nanofluidics; Thermal conductivity of liquids; Extreme gradient boost; Graphene oxides; Hybrid nanofluid; Nanofluids; Neural-networks; Shapley; Shapley additive explanation; Thermal; Transport characteristics; Transport phenomenon; Titanium dioxide

Comprehending and managing the transport characteristics of nanofluids is critical for improving their efficacy in heat transfer applications, thereby improving thermal management systems. This research focuses on investigating the impact of varying concentrations (0.05?1 vol.%) and temperatures (30?60��C) on the thermal conductivity and viscosity of water-based nanofluids. These nanofluids contain graphene oxide, silicon dioxide, and titanium dioxide, as well as hybrid combinations thereof. The research revealed that nanofluids exhibit higher viscosity and thermal conductivity compared to water. The maximum thermal conductivity and viscosity of 1.52 and 2.77 are observed for GO for 1 vol% compared to the water at 60 and 30��C, respectively. Notably, graphene oxide nanofluid exhibits the highest thermal conductivity and viscosity among all the studied nanofluids. These findings imply that graphene oxide and its hybrid nanofluids hold promise for enhancing heat transfer and energy efficiency in various industrial applications. The modeling and simulation of hybrid nanofluids' thermophysical properties are difficult and time-consuming. Modern machine learning algorithms are capable of handling such complex data. As a result, in the current investigation, two distinct ensembles and deep learning-based techniques, deep neural networks and extreme�gradient boost, were used. The statistical examination of the viscosity model shows that the extreme�gradient boost-based model had an R2 value of 0.9122, while the deep neural network-based model had just 0.7371. The mean square error for the extreme�gradient boost-based model was just 0.010, whereas it climbed to 0.0329 for the deep neural network-based model. ? Akad�miai Kiad�, Budapest, Hungary 2024.