Blades in aero-engines and other high-speed rotating machinery are continuously subjected to cyclic loads and complex aerodynamic forces. Damage such as fracture, wear, and bending may jeopardize operational safety, making noncontact condition monitoring and early fault identification essential. Continuous-wave (CW) radar can acquire blade micromotion echoes without contacting the rotating component and thus offers a new means of condition sensing in challenging environments. However, speed variations substantially alter the time--frequency distribution of blade echoes, causing a domain shift when a diagnostic model is transferred between operating conditions, while fault labels are generally unavailable at the target speed. To address this problem, this paper develops a domain-adversarial diagnostic framework based on a spiking neural network (SNN). Its input is a multiband representation formed by applying the short-time Fourier transform (STFT) separately to three modes obtained through complex variational mode decomposition (VMD). Direct input encoding and a spiking convolutional feature extractor learn time--frequency representations, while a gradient reversal layer and a domain discriminator align features from the source and target domains. The model thereby suppresses speed-dependent differences while retaining fault-discriminative information. Experiments on four leave-one-speed-out transfer tasks show that the proposed method achieved a mean target-domain accuracy of 95.60 percent , outperforming the Source Only baseline with the same SNN backbone by 13.90 percentage points. These results demonstrate the feasibility of combining spiking feature extraction with domain-adversarial learning for cross-speed CW-radar blade diagnosis.

CW radar,rotating blade,fault diagnosis,spiking neural network,unsupervised domain adaptation,VMD