Effect of alkali metals working fluids on oscillating heat pipes performance: An experimental and machine learning approach

The heat transfer performance of oscillating heat pipes is significantly influenced by the thermophysical properties of the working fluids. However, research on these systems and their working fluids remains limited. This study investigated the effects of different operating conditions and working fluids on the heat transfer performance of high-temperature oscillating heat pipes through experiments and machine learning methods. The performance of pipes filled with three alkali metals (potassium, rubidium, and cesium) was experimentally evaluated. Additionally, seven machine learning algorithms were employed to predict thermal resistance based on the thermophysical properties of the working fluids. The results showed that, at low input powers, rubidiumand cesium-filled pipes exhibited similar and superior heat transfer performance compared to the potassiumfilled pipe. The lowest thermal resistance of 0.17 K/W was achieved under conditions of 3400 W heat input, a 60 degrees inclination angle, and cesium as the working fluid. The Random Forest algorithm provided the best prediction accuracy, achieving a coefficient of determination (R2) of 0.9947, a mean absolute error (MAE) of 2.739 x 10-3, and a mean squared error (MSE) of 2.581 x 10-5, when using thermophysical properties and input power as input features. Feature importance analysis and SHapley Additive exPlanations (SHAP) method highlighted the critical roles of (dP/dT)sat and dynamic viscosity in fluid motion. This study presents a novel approach for analyzing the heat transfer performance of high-temperature oscillating heat pipes and provides theoretical guidance for developing more efficient high-temperature working fluids.