An integrated methodology and sensitivity analysis for the prediction and multiobjective optimization of the thermophysical properties of MWCNT/water nanofluids using SVR-GA and SPEA2

Abstract This study proposes an integrated methodology that utilizes intelligent computing to precisely predict and optimize the thermophysical properties of a three-dimensional multi-walled carbon nanotube (MWCNT)/water nanofluid. Two distinct Support Vector Regression (SVR) models, enhanced by hyperparameter optimization using a Genetic Algorithm (GA), were successfully trained to accurately estimate the dynamic viscosity (DV) and thermal conductivity (TC) based on experimental data across a range of volume concentrations (0–0.5%) and temperatures (30–65 °C). The resulting SVR-GA models demonstrated high predictive accuracy, evidenced by the test data’s mean and maximum relative percentage errors: 0.4546% and 0.9690% for DV, and 0.5950% and 0.9532% for TC, respectively. Moreover, the trained SVR models have been validated using K-fold cross-validation, and the network’s accuracy in estimating unseen data has been demonstrated. A sensitivity analysis confirmed that temperature exerts a greater influence on nanofluid properties than volume concentration. Leveraging these high-accuracy models, a multiobjective optimization problem was formulated to simultaneously minimize DV and maximize TC. This problem was solved using the Strength Pareto Evolutionary Algorithm 2 (SPEA2), which yielded a Pareto front of optimal trade-off solutions. The analysis of the Pareto front consistently identified 65 °C as the optimal operating temperature, providing useful insights for selecting conditions that maximize the nanofluid’s heat transfer efficiency.