Fault detection for inverter-driven Permanent Magnet Synchronous Motor (PMSM) drives is critical for effective and reliable modern industrial and vehicle systems. A majority of classical supervised fault detection learning algorithms require significant amounts of labelled data, which can be extremely costly and time-consuming to acquire. Self-labelled strategies utilizing labelled as well as unlabelled data in the interest of improved classification performance offer a promising prospect. This paper offers a comparative evaluation of self-labelled approaches for PMSM fault detection, such as self-training, co-training and tri-training. We evaluate these approaches on a set with varied motor operating conditions and fault instances and compare them on the basis of classification accuracy. Experimental outcomes reveal that self-labelled approaches significantly reduce dependence on labelled samples with maintaining high fault detection accuracy. Besides, the study provides significant insights into their practical application in PMSM systems.