An online fault detection and isolation (FDI) method for several common sensor faults and even demagnetization of Permanent magnet synchronous machine (PMSM) is proposed by combining model-based and signal analysis technology. The flux residuals are transformed by multi sequence harmonic synchronous rotating transformation and inputted into low pass filters (LPF) in order to obtain the DC components. Last, offset and gain faults of the two phase current sensors, offset fault of the rotor angle sensor and permanent magnet (PM) demagnetization can be isolated by comparing the DC components and preset thresholds. The detection and isolation strategy of PMSM is validated by motor controller hardware in motor bench tests. This paper proposes two magnet flux linkage observers using inputs of phase current measurements and the control values of stator voltage respectively. The flux residuals are generated by comparing the outputs of the two flux observers and used to detect faults. Then, the residual signals are analyzed using multi sequence harmonic synchronous rotating transformation and low pass filtering, and the analysis results are used to isolate the current sensor faults, rotor position sensor fault and PM demagnetization. Last, the faults diagnosis method is validated by motor tests.