Thermal conductivity prediction of metallic and metallic oxide nano-enhanced phase change materials using machine learning

Phase change materials (PCMs) have found potential applications in energy storage and thermal management systems; however, their low thermal conductivity reduces their efficiency. To address this issue, metallic and metal oxide nanoparticles are integrated with PCMs to produce Nano-Enhanced Phase Change Materials (NEPCMs) with higher thermal conductivity. In this study, an optimized artificial neural network (ANN) is proposed to predict the thermal conductivity of various NEPCMs, which consist of different types of nanoparticles, including silver, iron, copper, aluminum, zinc, copper oxide, zinc oxide, titanium oxide, silicon oxide, aluminum oxide, iron oxide, and cerium(IV) oxide, as well as different types of PCM including paraffins, ethyl cinnamate, myristic acid, magnesium nitrate hexahydrate, PEG2000, capric acid, capric–myristic acid, palmitic acid, lauric–stearic acid, and PEG6000. The dataset was collected from experimental investigations reported in the literature. The dataset was preprocessed and tested with bayesian surrogate optimization with different surrogate models including random forest, gaussian process, and gradient boosting regression tree to select optimal hyperparameters for training the ANN. Results showed that applying an KNN method for filling the missing data with bayesian surrogate gaussian processes as an optimization method produced the best hyperparameters for training the ANN. The trained ANN achieved prediction accuracy, with an R2 value of 0.729 and MSE and MAE values of 0.0111 and 0.0552, respectively, on the testing dataset. Moreover, sensitivity analysis demonstrated that the output parameter is most sensitive to loading ratio (WT%), and least sensitive to liquid density (PCM). on the other hand, SHAP analysis demonstrated that the model's output is highly dependent on loading ratio (WT%), Thermal conductivity (NP), diameter (NP), and is least dependent on the liquid density of the PCM.