Integrating machine learning and sensor technology to automate proton exchange membrane fuel cell stack disassembly

Abstract Hydrogen-based proton exchange membrane fuel cells (PEMFCs) are expected to be deployed in heavy-duty applications, resulting in a growing number of systems requiring effective end-of-life (EoL) strategies. Current PEMFC EoL strategies rely on manual disassembly to recover valuable resources such as platinum. Consequently, limited knowledge exists regarding information and sensor requirements for robust automation. This publication proposes and evaluates a sensor concept for PEMFC stacks to enable automated disassembly. Laser line triangulation (LLT) sensors are used to measure the height of stack components and to determine cut points for separation of stacks. With a measurement resolution of 17–26 $$\\mu $$ m at working distances of 70–130 mm, the system allows for a geometry-independent operation across varying stack designs. A camera system with dedicated lighting and machine-learning (ML)-based image analysis based on Cognex VisionPro Deep Learning is employed to classify bipolar plates (BPPs) and detect typical defects relevant for remanufacturing. The developed models achieve an F1-Score of 0.964 for seal defect detection and 0.914 for scratch detection on a representative dataset. The experimental results demonstrate an overall feasibility of the subsystems, providing key input data for automated EoL handling of PEMFC stacks and component classification for re-x strategies. With these developments, new opportunities emerge for the design of flexible industrial manufacturing and remanufacturing systems for hydrogen-based mobility. A future research pathway emerging from this work lies in the development of remaining useful life prediction (RUL) based on the structured condition information generated by the proposed geometry- and defect-inspection approach.