Prognostic modeling of polydisperse SiO2/Aqueous glycerol nanofluids' thermophysical profile using an explainable artificial intelligence (XAI) approach

• Main Authors: Sharma, K.V., Talpa Sai, P.H.V.S., Sharma, P., Kanti, P.K., Bhramara, P., Akilu, S.
• Format: Article
• Published: Elsevier Ltd 2023
• Online Access: http://scholars.utp.edu.my/id/eprint/37301/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85167809881&doi=10.1016%2fj.engappai.2023.106967&partnerID=40&md5=e658726409ceadf1be85bccce4b74934

Ceramic nanoparticles have become increasingly popular owing to their wide range of engineering applications in the industry. Silica is one of the most promising nanomaterials for heat transfer applications due to its favorable thermophysical characteristics and dispersion stability. This work investigates the thermophysical properties of polydisperse SiO2 nanoparticles in a glycerol and water mixture. Aqueous glycerol nanofluid comprised 30 glycerol (30 GW) by weight, and SiO2 with mean particle sizes of 15, 50, and 100 nm were prepared for 0.5 and 1.0 volume concentrations using a two-step method. Measurements of properties are made in the temperature range of 30�100oC. The nanofluids behaved as a single-phase liquid remaining stable for four weeks. The viscosity and density of the base liquid and nanofluid decreased while the thermal conductivity increased with temperature. The base liquid-specific heat increased slightly while that of nanofluid decreased significantly with temperature. For 0.5 SiO2 concentration, the thermal conductivity and viscosity varied by 11.1 and 32, respectively at 60oC compared to the base liquid. Correlations are developed for the estimation of nanofluid properties. An Explainable Artificial Intelligence (XAI) technique called Bayesian approach optimized Gaussian Process Regression was employed to develop a prognostic model for the thermophysical properties of test nanofluids. In addition, the model's predictability and explainability are both enhanced by the use of kernel functions, which makes it possible to include historical data in the representation of a phenomenon. The test XAI approaches were shown as robust one because of the high correlation values, which ranged from 99.68 to 99.99, along with minimal modeling errors. © 2023 Elsevier Ltd