Fatigue Crack Detection and Classification Based on Ultrasonic Guided Waves and Deep Learning Models

Fatigue cracking in aircraft aluminum alloy structures under cyclic loading poses a critical threat to aviation safety. To achieve efficient and precise monitoring, this paper investigates a crack detection and classification method integrating ultrasonic guided waves with hybrid deep learning. A sparse piezoelectric sensor network captures Lamb wave signals throughout the entire process from structural integrity to crack propagation, effectively characterizing damage evolution using differential processing and integral spectral energy (ISE) features. A TCN-LSTM-TransformerEncoder model is constructed to extract local multi-scale features, model temporal dependencies, and capture global contextual correlations. This model achieves 96.25% accuracy in four-level crack classification. Ablation experiments validated the performance contributions of each module and the importance of differential features. The study demonstrates that this method achieves high-accuracy fatigue crack identification even under limited sensor conditions, providing a reliable technical pathway for intelligent structural health monitoring in aerospace applications.