Unsupervised adaptive feature enhancement network for reliable damage alarm using guided waves under time-varying conditions

Abstract

Structural health monitoring technology is crucial for ensuring the safe operation and reducing maintenance cost of aircraft structures. But complex time-varying conditions, which contain multiple coupled service environment factors, will significantly reduce the accuracy and reliability of damage alarm. Aiming at this issue, this paper proposes a guided wave (GW) based unsupervised (damage‑label‑free) adaptive feature enhancement network (UAFEN), which is capable of achieving accurate and reliable structural damage alarm under coupled time-varying conditions. UAFEN introduces a feature distribution alignment mechanism based on a modified multi-kernel maximum mean discrepancy algorithm to enhance the model’s adaptability to environmental changes, and incorporates a sequence attention mechanism to reinforce the model’s response to localized, subtle anomalies in GW signals. Validated through crack and pit damage monitoring on an actual wing-box skin exposed to coupled temperature-load variations, the proposed method delivers accurate, reliable, region-specific damage alarms under coupled temperature and load time-varying conditions, underscoring its strong promise for practical engineering deployment.

Keywords

• Structural health monitoring
• coupled time-varying conditions
• guided wave
• unsupervised learning
• damage alarm