Short circuit faults in the stator winding are a common failure mode in Permanent Magnet Synchronous Machines (PMSM). Model-based fault detection has a potential benefit of being less sensitive to transient conditions compared to frequency-domain based methods for fault detection. However, accurate models of the machine being monitored are required in order to produce accurate fault detection whilst avoiding false alarms. This paper presents a model-based fault detection method based on the analysis of the residual between the measured currents and those predicted by an accurate real-time model of the machine. The model is nonlinear and based in pre-calculated flux linkages look up tables obtained through magneto static finite element analysis (FEA), accounting for magnetic saturation and spatial harmonics. The use of a residual provides a mean to amplify frequency components caused by a fault. Computer simulations, real-time Hardware-in-the-Loop (HIL) methods and experimental validations are provided.