The increasing application of permanent magnet (PM) machines in emerging safety critical sectors such as electric vehicles and "more electric" aircrafts due to their high torque density and high efficiency has raised the demand for high availability (or reliability). Health monitoring for the machines is becoming an important tool for improving drive availability by facilitating intelligent maintenance and avoiding catastrophic failure. This thesis is concerned with the diagnostics of the turn-to-turn fault, also known as turn fault, in PM machine stator windings by enhancing the detectability, robustness against transient states, and reducing the sensitivity towards other faults especially high resistance connection (HRC) fault. Firstly, it proposes a turn fault detection method based on current residual and angular integration. The fault feature is enhanced in the current residual while the impacts of transient states are minimized by the angular integration. Secondly, the fault features at the switching sideband harmonics are also studied and a fault indicator based on the PWM ripple currents is proposed and tested on different permanent magnet machines. Thirdly, to distinguish the HRC fault from the turn fault, both the impedance around switching sideband frequencies and high frequency voltage injection are investigated, which explore the different behaviours of resistive and inductive asymmetry at high frequency. The robustness against transient states is achieved by the signal processing in the designed circuit board. Finally, to improve the fault detectability at high frequency, an artificial neutral point is established to provide zero sequence voltage measurements at both fundamental and high frequencies, and the results allow for a better classification of HRC and turn faults with less impact from operating conditions. The investigation and implementation of turn fault detection techniques for a commercial permanent magnet alternator with a unique structure is also performed.