Thermo-physical properties estimation of an oil-based hybrid nanofluid: application of a new hybrid neurocomputing approach

Accurate estimation of thermophysical properties of hybrid nanofluids, such as thermal conductivity (THC), viscosity, and heat transfer performance (HTP), is crucial in energy applications and heat transfer. In this research, a comprehensive experimental and computational investigation has been performed to predict the heat transfer performance (HTP) and thermo-physical properties of a new hybrid engine oil-based nanofluid comprised of Mg(OH)₂ and multi-wall carbon nanotubes (MWCNTs) based on the temperature and the volume fraction of solids. Due to this aim, a new hybrid meta-heuristic data-intelligent method, called teaching–learning-based optimization integrated with an adaptive neuro-fuzzy inference system (ANFIS-TLBO), was developed. The proposed ANFIS-TLBO approach for the prediction of the HTP and thermo-physical properties of the engine was evaluated for robustness using the standalone ANFIS, multi-layer perceptron Artificial Neural Network (ANN), and multivariate adaptive regression spline (MARS) model. For the construction of the models, the temperature and the volume fraction of solids were utilized as the input features. At the same time, the output variables were HTC values for the internal laminar and turbulent flow regimen (HTPL and THPT), THC, and dynamic viscosity. The robustness and efficiency of the machine learning (ML) models were assessed using the root mean square error (RMSE), mean absolute percentage error (MAPE), and correlation coefficient (R). The outcomes of ML models reveal that the ANFIS-TLBO outperformed the other models. A piece-wise relationship for every target was extracted using the MARS model based on temperature and volume fraction. Besides, the outlier detection and statistical validity revealed the primary model's reliability and promising predictability potential for accurate estimation of thermophysical properties of hybrid nanofluids.