In this paper, an online model-based fault detection approach based on residual analysis in synchronous generators (SGs) is presented. Two types of faults are studied in this paper: (i) reduction of the cross-sectional area of windings wires in SGs, and (ii) air-gap eccentricity. The residual vector based on the equivalent circuit (EC) parameters or state-space model of the SG is employed for fault detection. The introduced fault detection approach employs the stator and field currents and voltages, and rotor rotational speed. The main advantage of the presented method is able to be used for linear and nonlinear loads in the presence of uncertainty in EC parameters. The effectiveness of the proposed method is shown using the experimental data of five SGs in diesel-electric locomotives.