Anomaly detection using measurements from multiple positions is important for aero-engine inter-shaft bearings because different positions capture distinct responses of the coupled rotor-bearing system. In practice, these positions may be installed progressively because of constraints on installation space, wiring, acquisition capacity, maintenance access, and monitoring upgrades. Each newly deployed position must be evaluated to determine whether it provides discriminative information for bearing anomaly detection, while the detector must remain effective on all previously deployed positions because they continue to provide distinct measurements during operation. To avoid separately deploying and independently invoking a growing collection of position-specific models, a common framework is required to apply consistent position-aware sample routing, stage-wise updating, and cumulative evaluation across positions. Existing fixed-input detectors and sequential fine-tuning do not directly satisfy this requirement. We propose the Progressive Gated Continual Sensor-Expansion Anomaly Network (PG-CSAN), which assigns each arriving position a multi-scale temporal expert, applies position-specific logit scaling with a sensor gate, and freezes established branches during subsequent updates. Together, these components enable unified assessment of newly added and previously deployed sensor positions. Under a five-stage deployment protocol on a real aero-engine inter-shaft bearing dataset, PG-CSAN achieves 94.39% accuracy, exceeding the strongest baseline by 3.73 percentage points in accuracy.

Inter-shaft bearing, anomaly detection, continual learning, sensor expansion, multi-scale temporal convolution.