The online diagnosis of an electrical drive is the first step towards the application of an effective predictive maintenance plan on such electrical machines. The analysis of residual features corresponding to instrumentation faults of an induction motor lead to the design of a faults classification system. The training set is obtained by a faulted machine dynamical model as simulator. Two neuro-fuzzy structures will be conceived to learn the exact input-output relation of the fault detection process for induction motor, using measured data. The first neuro-fuzzy architecture maps the residuals into two classes: one of fixed direction residuals and another of faults belonging to velocity sensor. The second adaptive neuro-fuzzy network will provide updated membership functions of the sets of fixed oriented residuals that better describe the fault diagnosis map.