Optimizing Pulsating Heat Pipes Using Artificial Neural Networks

Mira Ibrahim1^{1}, Majed Eddine Moustaid1^{1}
^{\star} : mira.ibrahim@capgemini.com
1^{1} Capgemini engineering
Mots clés : Pulsating heat pipes, Artificial neural networks, Regression, Two-phase flow
Résumé :

As thermal appliances become more compact, they generate excess heat which presents an ongoing challenge for engineers tasked with managing heat in various systems. Thus, effective heat management became a complex and challenging issue for thermal engineers. Pulsating heat pipes (PHPs), characterized by constant pulsating/oscillating motions of vapor bubbles and liquid slugs, are a potential option for dispersing high heat loads due to their superior thermal conductivity and adaptability to various orientations. They are used in various thermal applications, including those in the aerospace and electronics industry. However, several geometric and operational criteria have an impact on how well PHPs perform. As a result, designers need to optimize these factors to improve the performance of PHPs.

This research aims to develop adequate machine-learning models to predict the thermal performance of a Closed-Loop PHP based on several factors including the overall length, the lengths of the condenser and evaporator, the internal diameter, the number of turns, the heat load, the filling ratio, the inclination angles, the working fluid properties, and the condenser temperature. The constructed model will provide an effective tool to characterize and optimize Closed-Loop PHPs in various application synergies.

Thus, a database with different PHP geometries and multiple operating points was sourced from various works in the literature resulting in 612 experimental records. Artificial Neural Network (ANN) models were built to predict the thermal resistance of PHPs. The hyperparameters of these models are selected after a dedicated optimization was undertaken to achieve optimal accuracy. Validating the ANN predictions against experimental data is satisfactory within the total available data sets domain. The analysis of the correlation matrix confirms that the cold source temperature, the PHP’s inclination, and the condenser length are the most important independent parameters under consideration. The accuracy of the selected model is analyzed through the mean absolute error (MAE) of the thermal resistance resulting in an acceptable MAE value lower than 24%. The approach presented in this work seems promising in predicting and optimizing the thermal performance of Closed-loop PHP.

Work In Progress