The work carried out aims to diagnose fuel cells with reduced instrumentation and computation times. The article describes a non-intrusive, application-oriented diagnostic tool, based on the sole measurement of the stack voltage and requiring no specific external excitation of the electrochemical generator. The adopted data-driven method relies on well-suited signal analysis techniques (fast calculations of relevant fault signatures based on envelopes and instantaneous frequencies) and information processing (pattern recognition). A wide range of operating regimes can be identified (variations in flow rates, pressures, temperatures; combinations of simultaneous faults), even when they correspond to small deviations from nominal conditions. The portability of the method has been studied on two PEMFC stacks, designed for different applications: transport and stationary. Correct classification rates close to 98% are obtained in both cases.