Visualization of water transport behavior and distribution characteristics in PEMFC flow channels based on deep learning image recognition

Proton exchange membrane fuel cells (PEMFCs) suffer from water management issues that restrict their performance and durability. This study combines an improved YOLOv8s-P2 deep learning model with optical visualization to quantitatively analyze liquid water transport in flow channels. The improved model achieves an F1 score of 0.93 and a recall of 0.98, enabling accurate identification of droplet, film, and slug patterns. The results demonstrate that the perforated gas diffusion layer (GDL) significantly enhances water removal efficiency compared with the normal GDL. Specifically, in ex-situ tests, the perforated GDL reduced the P95 indicator from 4.17% to 0.63% in the single-injection setup, and from 12.14% to 1.25% in the multi-injection setup. This improvement is achieved by creating preferential drainage pathways to reduce the breakthrough pressure. The perforated GDL promotes intermittent slugs but suppresses film flow growth. In-situ tests, perforated GDL shows slightly higher P 95 but achieves lower water saturation in GDL, resulting in a “wetter channel but drier GDL” state. Overall, perforated GDLs provide more efficient and stable water management, enhancing PEMFC performance under high current density operation. This work provides a powerful tool for liquid water analysis and offers valuable guidance for PEMFC water management optimization.