In-situ humidification performance evaluation of various membranes for proton exchange membrane fuel cell

Water management is a critical aspect of maximizing the performance and lifespan of proton exchange membrane fuel cells (PEMFCs). A membrane humidifier is an essential piece of equipment for maintaining PEMFC performance. This study focuses on assessing the in-situ humidification potential of three membranes - Nafion (NR-212), reverse osmosis (RO), and pervaporation (PV) membranes - for use in a planar membrane humidifier. The humidification potential was evaluated based on five different parameters: pressure drop, water recovery ratio, water flux, coefficient of performance, and dew point approach temperature. To create an accurate deep learning model, the study chose flow rate, temperature, humidity, and membrane type and material as the most significant input parameters. The data was first augmented using CTGAN, and the synthetic data was found to be well-correlated with real data, with a similarity score of 0.4805. The optimal deep neural network (DNN) model was created using Bayesian surrogate models, including random forest, Gaussian process regression, and gradient boosting regression trees. The model demonstrated a high level of accuracy, with a correlation coefficient of 0.986 and a mean absolute error of 0.077, 0.22, 0.265, 0.03, and 0.045 for pressure loss, DPAT, WRR, J, and COP, respectively. The model was also validated on unseen experimental data, with a correlation coefficient of 0.94. Finally, the predictions of the deep learning model were analyzed using explainable artificial intelligence (XAI) via the SHAP library. The analysis included dependence plots, embedding plots, partial dependence plots, summary plots, and force plots. When assessing the in-situ humidification performance of planar membrane humidifiers, temperature is the primary input variable that influences the outcome, followed by flow rate and humidity.