[Objective] Inter-turn short-circuit faults are a common and latent early-stage failures in circuit and motor drive systems. Although the underlying physical mechanisms of such faults are essentially identical across different phases, significant discrepancies in data distributions arise due to phase shifts and measurement conditions, which limits the cross-phase applicability of data-driven diagnostic methods. This paper proposes a unified diagnosis method for inter-turn short-circuit faults in different phases under limited annotation conditions. [Methods] The phase-to-phase difference was regarded as a specific form of data distribution shift, so the core criterion for inter-turn short-circuit was independent of the specific phase. Based on this, a feasible method was proposed in this paper from the perspective of cross-phase transfer, which utilized only single-phase inter-turn short-circuit fault data for model training and directly applied it to fault diagnosis of other phases. This method achieved multi-phase inter-turn short-circuit fault identification under single-phase training conditions by constructing a physically consistent feature representation and employing a unified lightweight diagnostic model. [Results] The results indicated that, without the introduction of complex transfer structures or additional labeled data, the proposed method demonstrated high diagnostic accuracy on the training phase and maintained stable performance in different phase-to-phase inter-turn short-circuit fault diagnosis tasks, significantly enhancing the cross-phase generalization ability of fault diagnosis. [Conclusion] This study provides a concise and effective solution for reducing the dependence of the inter-turn short-circuit fault diagnosis model on multi-phase labeled data and enhancing the model’s generalizability in practical engineering scenarios, thus possessing high engineering practical value.