Physics-informed explainable machine learning and thermodynamic optimisation of CNT-based nanolubricants for enhanced vapour compression refrigeration system performance

The significance of refrigeration lies in decreasing electricity consumption. However, carbon nanotube (CNT) nanolubricants have nonlinear effects that are challenging to predict through traditional analysis. This research develops a comprehensive experimental, thermodynamic, explainable machine learning (ML), and physics-based optimisation framework for the refrigeration system with R134a working fluid and CNT/polyolester (POE) nanolubricant. Experimental data for different concentrations of CNT ranging from 0 to 1 wt% are analysed without data pre-processing, whereas states, refrigeration effect, compressor work, coefficient of performance (COP), and exergy efficiency are determined through CoolProp and TESPy simulations. Random forest and XGBoost models are created using SHAP analysis with thermodynamics and energy balance conditions imposed on optimisation. The optimal concentration is 0.075 wt%, leading to COP equal to 2.031, evaporator refrigeration capacity of 0.640 kW and compressor work of 0.315 kW. The refrigeration effectiveness, compressor work per unit mass, mass flow rate and exergy effectiveness are 148.50 kJ kg −1 , 73.10 kJ kg −1 , 0.00431 kg s −1 and 41.10% respectively. The R 2 values for COP, heat extraction rate and work input obtained by XGBoost are 0.994, 0.992 and 0.989, respectively.