Prediction of nanofluid thermal conductivity and viscosity with machine learning and molecular dynamics

Abstract

It is well-known that nanofluids differ significantly from traditional heat transfer fluids in terms of their thermal and transfer characteristics. Two of CO2 transfer characteristics, its thermal conductivity and its viscosity, are crucial to improved oil retrieval methods and industries refrigeration. By combining molecular modelling with various machine learning algorithms, this study predicts the conduction characteristics of iron oxide CO2 nanofluids. It is possible to evaluate the accuracy of these transfer parameter estimates by applying machine learning methods such as decision tree, K-nearest neighbors, and linear regression. Predicting these transfer qualities requires knowing the size, fraction of nanoparticle volume, and temperature. To determine the characteristics, molecular dynamics simulations are run using the large-scale atom Vastly equivalent simulant. An inter- and intra-variable Pearson correlation was established to confirm that the input variables were reliant on m and thermal conductivity. The results were finally confirmed by using statistical coefficients of determination. For a variety of temperature ranges, volume fractions, and nanoparticle sizes, the study found that the decision tree model was the best at predicting the transport parameters of nanofluids. It has a 99% success rate.