An approach for stack-integrated fuel cell Cooling in aerospace; Machine learning-based performance prediction of Pulsating Heat Pipes

Kurzfassung

The transition towards more sustainable aviation has intensified research into novel concepts for aircraft propulsion. In this context, hydrogen-powered proton exchange membrane fuel cells (PEMFCs) represent a promising technology for aerospace applications, but require effective cooling to maintain high efficiency and minimize degradation. Conventional liquid cooling systems require parasitic power and additional components, increasing system complexity and weight. As an alternative, pulsating heat pipes (PHPs) could be integrated into the residual space between bipolar plates, providing a passive, stack-integrated cooling solution. For the efficient design of future PHPs, accurate performance prediction is essential. Therefore, this work expands a machine learning model based on the extreme gradient boosting (XGBoost) algorithm to predict PHP thermal resistance and employs game theory based Shapley values for model interpretation. The final model achieves a mean absolute percentage error (MAPE) of 13.6% and an R2 value of 0.979, supporting the development of new PHP designs and their integration into fuel-cell stack cooling concepts for aerospace applications.