Abstract: Three-phase asynchronous motor (TPAM), as a key power component in industrial drive systems, directly affects equipment reliability and production system safety. However, TPAM fault samples are difficult to obtain, and signals from a single sensor are often insufficient to comprehensively characterize different fault states, which limits the recognition accuracy and stability of existing diagnostic methods in few-shot scenarios. To address these problems, we propose a few-shot TPAM fault diagnosis method based on multi-source six-view data and task-reliability fusion. First, synchronously acquired current and vibration signals are constructed into six complementary views, and multi-view fault feature representations are obtained through a shared multi-scale Ghost depthwise separable gated feature extractor. Then, a task-reliability descriptor is constructed using support-set features, and fusion weights for different views are adaptively generated to enhance more discriminative fault information in the current diagnostic task. Finally, residual feature calibration and a prototypical classifier are combined to achieve multi-class fault identification with only a few labeled samples. Experimental results show that the proposed method outperforms the comparison models in the 8-way 5-shot few-shot scenario, achieving a maximum accuracy of 99.63%. This research can provide an effective method for achieving high-precision intelligent diagnosis under the condition of scarce TPAM failure samples.