In the era of renewable and clean energies, the demand for less polluting energy generation technologies has increased rapidly. Among these technologies, the Proton Exchange Membrane Fuel Cell (PEMFC) receives much attention, as it can convert the hydrogen chemical energy into electricity with high efficiency, and also produce water and heat. However, to make this technology commercially viable, some challenges still remain. Especially the extension of the fuel cell lifespan and reliability are identified as major concerns in the research and industry sectors. The lifetime and reliability objectives can notably be achieved by implementing a diagnosis tool capable of high performances, whatever the stack design and the operating environment. In this context, we propose a new tool based on the investigation of singularity measurements stamped in fuel cell stack voltage signals. Indeed, measuring local singularities on voltage signals provides suitable information about the evolving dynamics of non-stationary and non-linear processes involved in fuel cell systems. In our study, two PEMFC stacks are experimented to evaluate the portability of our diagnosis tool. The first one is an 8 cell stack designed for automotive applications and manufactured by CEA LITEN, France. The second one is an 12 cell stack dedicated to stationary application (micro combined heat and power - µCHP application). It is designed and marketed by Riesaer Brennstoffzellentechnik GmbH and Inhouse Engineering GmbH, Germany. The steps of our diagnosis strategy are the following ones: - Two PEMFC stacks are operated under a variety of conditions (nominal, and faults i.e. more or less severe deviations from the nominal conditions) using characterization testbenches developed in lab. The deviations from the nominal conditions refer either to single fault types or to combinations of different faults. - The recorded stack voltages are analyzed using a Wavelet Leader based Multifractal Analysis (WLMA) in order to identify their singularity spectra as fault signatures. - A feature selection method is used to select the most relevant singularity features and to remove the redundant ones. - The selected singularity features are classified using Support Vector Machine (SVM) classifier according to the considered operating situations (faults and combinations of faults). The obtained results show that the proposed PEMFC diagnosis tool allows identifying simple operating failures and even more complicated situations that contain several failure types, for different stack sizes, powers and technologies for different power application environments. Framework of the 'Decentralized energy production' project, directed by EFFICACITY, the French R&D Institute for urban energy transition.