The stability and thermophysical properties of Al2O3-graphene oxide hybrid nanofluids for solar energy applications: Application of robust autoregressive modern machine learning technique

• Published in: Solar energy materials and solar cells 2023-05, Vol.253, p.112207, Article 112207
• Main Authors: Kanti, Praveen Kumar, Sharma, Prabhakar, Maiya, Manoor Prakash, Sharma, Korada Viswanatha
• Format: Article
• Language: English
• Subjects: Al2O3; Graphene oxide; Machine learning; Mixture ratio; PER; Thermal conductivity
• ISSN: 0927-0248; 1879-3398
• DOI: 10.1016/j.solmat.2023.112207

This paper investigates the dispersion stability and thermophysical characteristics of water-based alumina (Al2O3), graphene oxide (GO) and their hybrid nanofluids (HNF) at different mixing ratios. Initially, the sol-gel and Hummer's method was employed for the synthesis of Al2O3 and GO nanoparticles (NPs) and they were characterized with X-ray diffraction analysis (XRD), ultraviolet–visible spectroscopy (UV–visible) and field emission scanning electron microscopy (FESEM). The effect of three different surfactants was analyzed on the stability of nanofluids (NFs). The properties such as thermal conductivity (TC) and viscosity (VST) were measured at different volume concentrations and temperatures ranging from 0.1 to 1 vol% and 30–60 °C, respectively. The maximum TC enhancement of GO is 43.9% higher than Al2O3 NF at 1 vol% at a temperature of 60 °C. The addition of GO content increases the TC and VST of HNF. The regression equations were developed to forecast the VST and TC of HNFs. Finally, two modern novel machine learning approaches, a Bayesian optimized support vector machine and a wide neural network, were used to model-predict the thermophysical properties of HNFs with a robust prognostic efficiency of 97.15–99.91%. •Al2O3 and GO nanoparticles were prepared using the sol-gel and Hummer's method.•SDBS provides better stability for Al2O3 based HNFs.•The addition of GO content enhances the thermal conductivity and viscosity of HNF.•Bayesian optimized support vector machine and wide neural network was used.