The study summarized in this paper proposes a new tool for PEMFC non-intrusive diagnosis based on voltage singularity measurement and classification. The method takes advantage of the non-linearities associated with discontinuities introduced in the dynamic response data resulting from various failure modes. Continuous wavelets and multifractal formalism, named WTMM (Wavelet Transform Modulus Maxima), are used together to quantify the singularity strength of the signal. The singularities signature of poor PEMFC operating conditions (faults) is first revealed through multifractal spectra. Then, these ones are classified using SVM (Support Vector Machine). The good classification rates obtained demonstrate that the multifractal spectrum based on WTMM is effective to extract the incipient fault features during PEMFC operation. The proposed method leads to a promising non-intrusive and low cost diagnostic tool to achieve on-line characterizations of dynamical PEMFC behaviors.