Experimental studies and machine learning approaches for thermal parameters prediction and data analysis in closed-loop pulsating heat pipes with Al2O3-DI water nanofluid

A closed-loop pulsating heat pipe (CLPHP) can provide effective and adaptable thermal solutions for various applications. This work presents extensive experimental studies on CLPHP to enhance thermal performance using nanofluid. The experimental studies are conducted using two different heat transfer fluids: deionized (DI) water and a nanofluid (Al₂O₃-DI water with 0.1 mass/% nanoparticles). Parametric studies are performed with different combinations of filling ratios (FR) and heat input values. To analyze the experimental data, an in-house Python library named PyPulseHeatPipe is developed, which facilitates statistical analysis, data visualization, and process data for machine learning from raw experimental data. Furthermore, the experimental datasets are used to train various machine learning (ML) models, including random forest regressor (RFR), extreme gradient boosting regressor, gradient boosting regressor, support vector machine, and K-nearest neighbors (KNN) to determine the thermal parameters for a given CLPHP. These models precisely predict the thermal performance of CLPHP using two novel approaches. The first approach predicts thermal resistance under given thermal properties such as evaporator temperature, pressure, FR, heat input, and heat transfer fluid, while the second approach predicts thermal parameters such as evaporator temperature, pressure, and heat input to achieve the desired thermal resistance. For the first approach, the RFR model performs the best among the trained ML models, with the lowest root mean square error (RMSE) of 0.0175 and the highest goodness of fit, with R² score and R²-adjusted (R²-adj.) of 0.9873 and 0.9872, respectively. For the second approach, the KNN model achieves the highest goodness of fit (R²-adj.) for evaporator temperature, pressure, and heat input values of around 0.9889, 0.9524, and 0.8149, respectively. This study establishes a foundation for the more efficient thermal design of CLPHP in various engineering systems by integrating experimental research with data-driven solutions through ML.