The automatic diagnosis of systems is essential in several industries such as aeronautics. This paper introduces a method to diagnose systems with respect to the constraints of the aeronautics field: robustness and low computation costs. The proposed methodology is based on the combination of Support Vector Machine and Fuzzy Membership Functions (SVM-MBF). The distances, which are computed by the SVM, are fuzzified in order to give a degree of confidence in the classification. Besides, using SVM-MBF allows estimating the severity of a fault. The architecture of the proposed diagnosis system consists in putting in series one classifier to detect faults, with a set of classifiers, one per fault, to assess the severity. The method is applied to the diagnosis of inter-turn short-circuits of a Permanent Magnet Synchronous Machine (PMSM). The data come from measurements performed on a machine designed for aeronautics applications. The method is evaluated in terms of robustness and computation time by using cross validation. The results show the suitability of the methodology for aeronautics applications and to the path of onboard diagnosis algorithm.