This paper deals with an on-line method for turn-to-turn short-circuit fault detection in low-voltage permanent-magnet synchronous machine drives. Due to the closed-loop control, the fault effects are reflected in the voltage. Therefore, an appropriate diagnostic index is proposed, which is derived from the positive- and negative-sequences of the voltage references. These sequences are obtained in the time domain via adaptive filters, which require only a few calculations. To increase the sensitivity to the fault, the algorithm is only applied to a part of the voltage references, i.e. the output of the proportional-integral controllers. Further, the cumulative-sum algorithm is introduced to cope with changes of small magnitudes. This algorithm allows a change of a fault index to be detected and can be used as a decision system. The resulting fault detection scheme is computationally cheap and can be embedded in the control unit. Simulations and experimental results validate the proposed method in steady state and the performances under non-stationary operating conditions are also investigated.