The fully encapsulated fuze electronic safety and arming device (ESAD) does not allow direct inspection of internal components without damaging the package. This makes it difficult to determine whether a device still meets service requirements after long-term storage. The capacitor is a key component of the ESAD high-voltage conversion module. This study examines samples representing eight common abnormal capacitor states after long-term storage. Firstly, an interferometric weak-FBG system records steady-state vibration responses transmitted to the metal housing after capacitor charging is completed. Then, spectral kurtosis enhanced filtering (SKFE) suppresses noise-dominated components, and mel-frequency cepstral coefficients (MFCC) and their first- and second-order differences form a 60×40 feature matrix. Finally, a dual-attention residual convolutional neural network (DA-ResCNN) classifies the capacitor states and achieves 96.28% accuracy. The results support non-destructive condition assessment of long-stored, fully encapsulated ESADs.

non-destructive testing,spectral kurtosis,dual attention network,weak fiber Bragg grating (wFBG）