Modelling the thermal conductivity of nanofluids using a novel model of models approach

Abstract

In this study, a unique method for modelling the thermal conductivity of nanofluids is proposed using a "model of models" approach. Three distinct data streams are utilised to achieve this. The first stream uses experimental data to predict thermal conductivity, an input for the primary machine learning model. The other stream involves modelling correlations from previous studies and integrating them as an additional input. Lastly, theoretical data streams are modelled and included as a last stream. By training a model on these combined data streams, the study aims to overcome various challenges in modelling nanofluids' thermophysical properties. The research holds great significance as it can potentially reconcile and understand errors that come with various modelling methods. This could result in improved model performance that closely resembles experimental data. The presented model in the model of models’ approach achieves a remarkable coefficient of determination (R-squared) value of 0.999 on the test data set, showcasing its exceptional accuracy and effectiveness in handling complex data, particularly about the thermophysical properties of nanofluids. Furthermore, this implicit general model comprises of data models incorporating material properties and physical phenomena, offering broad applicability. It is recommended that this approach be extended to viscosity, enhancing the understanding and prediction of nanofluid properties.