Accurate Prediction of Hybrid Nanofluids Viscosity: A Comparison of Soft Computational Approaches, Empirical, and Theoretical Models

Abstract

Hybrid nanofluids exhibit enhanced thermal properties compared to conventional nanofluids. Viscosity, critical for assessing heat transfer efficiency, influences pressure drop and pumping power. This study models hybrid nanofluid viscosity using Radial Basis Function (RBF), Multilayer Perceptron (MLP), and a Committee Machine Intelligent System (CMIS). A dataset of 584 viscosity data points is utilized. Particle Swarm Optimization (PSO) and Farmland Fertility Algorithm (FFA) are employed to train the RBF, while the MLP utilized Scaled Conjugate Gradient (SCG), Bayesian Regularization (BR), and Levenberg‐Marquardt (LM) algorithms. The CMIS is created by integrating MLP‐BR, RBF‐FFA, and RBF‐PSO networks. The AAPRE values for RBF‐PSO, RBF‐FFA, MLP‐LM, MLP‐SCG, MLP‐BR, and CMIS models are 1.7464, 1.6647, 2.6851, 2.1889, 2.1792, and 1.519, respectively. The R2 values are 0.9689, 0.9394, 0.4794, 0.9727, 0.9404, and 0.9688, respectively, which indicates that the CMIS model with the lowest Average Absolute Percent Relative Error (AAPRE) and the highest Determination Coefficient (R2) value is the most accurate model and outperforms other models in estimating viscosity, demonstrating greater accuracy than empirical and theoretical models. Sensitivity analysis showed that temperature has a significant positive impact on viscosity, while nanoparticle size has a negative effect. The CMIS model is reliable for predicting nanofluid viscosity, exhibiting a broad application range and minimal outlier data.