Sub-second X-ray tomographic microscopy of liquid water dynamics in polymer electrolyte fuel cells

Data courtesy: J. Eller (PSI)

The global energy system relies strongly on fossil fuels. Their expected reduced availability and detrimental environmental impact call for alternative energy solutions. Polymer Electrolyte Fuel Cells (PEFC) are a promising technology for future energy sources, especially in decarbonizing the mobility sector. To push this technology into the markets as a viable alternative to fossil fuels, improvements in efficiency, performance, durability and cost are still needed. Suboptimal water management during cell operation is one of the major limiting factors for increasing the performance at high current density operations. Dynamic sub-second X-ray tomographic microscopy is an invaluable technique to investigate these liquid water dynamics during cell operation [1,2].

The aim of this project is to develop sub-second tomographic microscopy for PEFC in order to detect the liquid water dynamics in the gas diffusion layers (GDL), the key component regulating water management during cell operation. Tomographic challenges arise from the sensitivity of PEFC to X-ray radiation and limited signal-to-noise ratio at the short scanning times required for the investigation of PEFC during transient operation. Advanced imaging procedures, including hardware and software developments, are therefore required. In a first step, the imaging hardware has been upgraded with a novel, high-numerical-aperture macroscope optics, especially designed for fast, high-spatial-resolution imaging experiments. The spatial-resolution of the original time-resolved imaging setup has been increased by a factor of 6 and, at the same time, the system is 4 times more efficient [3]. In a second step, an iterative reconstruction algorithm incorporating time-regularization has been developed to fully automatically reconstruct and segment the time-resolved fuel cell tomographic datasets, typically strongly undersampled and noisy [4]. Deep learning approach are currently being integrated within the proposed reconstruction scheme, so to significantly shorten the computational time and enable up-scaling of the reconstruction and segmentation process to efficiently handle large data volumes, extensively boosting the possibilities in future dynamic X-ray tomographic experiments.