Sonication impact on thermal conductivity of f-MWCNT nanofluids using XGBoost and Gaussian process regression

Background: Previous research has revealed that nanofluids are capable of improving the heat transfer performance of energy systems. Researchers devote a considerable deal of attention to multi-walled carbon nanotubes owing to their exceptional features, like superior thermal conductivity. By adding COOH functional groups to multi wall carbon nanotubes (MWCNTs), the carbon nanotubes become hydrophilic, hence enhancing the stability of nanofluids. Methods: In this investigation, nanoparticles were characterized using X-Ray Diffraction (XRD) and Brunauer–Emmett–Teller (BET). The durational effects of sonication on three distinct nanofluid concentrations (0.3, 0.1, and 0.05 vol.%) generated with functionalized-multi walled carbon nanotubes (f-MWCNTs/water) on thermal conductivity were also examined. After 80 minutes of sonication, a volume concentration of 0.10 vol% exhibits the best stable long-term value (39.84 mV) based on the findings of studies. As sonication duration increases, particle size decreases, and the 0.30 vol% concentration changes significantly. At a temperature of 50 °C and a sonication period of 20 minutes, the largest increase in thermal conductivity was 5.90% at a concentration of 0.10 vol %. The experimental data acquired through extensive lab-based testing was employed for the development of the meta-model. A modern machine learning technique i.e., extreme gradient boosting (XGBoost) has been employed to model-predict the effects of sonication on thermal conductivity. A contemporary machine learning method Gaussian process regression (GPR) was employed for comparative analysis. Significant findings: The optimal sonication time and long-term stability were determined to be 0.10 vol.% concentration and 80 minutes. All samples remained stable for up to two months during the examination. With proper sonication time and at 0.10 vol.%, improved dispersions and characteristics were obtained. The XGBoost-based model outperformed the GPR-based framework in terms of both reduced errors and greater correlation values. The coefficient of determination (R²) of the XGBoost-based model was observed to be 5.45% higher than that of the GPR-based model.