Abstract: The gear transmission system is a critical subsystem for power delivery in high-end equipment, yet it is susceptible to diverse and unpredictable faults, which may compromise operational safety and service reliability. In practical engineering, the lack of labeled fault data and uncertainty in fault locations pose significant challenges to fault localization. To address the challenges posed by the lack of fault samples and unknown fault locations, this paper proposes an Unsupervised Hypersphere-based Fault Localization (UHFL) method empowered by the mode component energy feature. By extracting features that are not only related to the fault mechanisms of transmission components but also possess clear physical interpretability and fault location indication, the proposed method enhances both unsupervised anomaly detection and interpretable fault localization. Specifically, a data description model is constructed using only normal samples to detect anomalies, and attribution analysis is employed to quantify each feature’s contribution to the anomaly decision. Features with high contribution scores are then mapped to corresponding faulty components for localization. Experiments are conducted on a helicopter planetary gearbox test bench and an armored vehicle transmission system platform, covering both single-fault and compound-fault scenarios. The results demonstrate that the proposed UHFL method can accurately localize faults at the component level without requiring any labeled fault samples. This method provides an practical solution for fault localization in gear transmission systems under data-scarce conditions, showing strong potential for real-world engineering applications and industrial scalability.