Piezoelectric-enabled wave propagation method integrated with multi-channel wavelet packet spectrum transformer for fatigue damage detection in SFRC

Steel fiber reinforced concrete (SFRC) is widely used in civil infrastructure and is often subjected to progressive fatigue damage during long-term cyclic loading, necessitating reliable and effective structural health monitoring (SHM) techniques for ensuring structural safety and timely maintenance. This paper proposed a deep learning (DL)-assisted wave propagation method for the intelligent and accurate detection of fatigue damage in SFRC. First of all, stress wave signals were collected during the cyclic loading-unloading tests of SFRC using piezoelectric-based smart aggregates. To enhance the damage-relevant feature representation, the measured signals were decomposed using the wavelet packet decomposition (WPD), and each sub-band was subsequently transformed into a time-frequency spectrum via the continuous wavelet transform (CWT). Based on these multi-scale features, a multi-channel Transformer architecture, namely the multi-channel wavelet packet spectrum Transformer (MCWPS-Transformer), was developed. The proposed network explicitly fused multi-layer channels with the global dependency modeling capability of multi-head self-attention, enabling more accurate and robust quantification of fatigue damage progression in SFRC. Moreover, comparative studies against different channel amounts and DL models were systematically performed. The results show that the proposed method significantly outperforms the benchmark models in predicting fatigue damage evolution, achieving R² values ranging from 0.9605 to 0.9964 across different datasets, thereby demonstrating its strong potential for practical applications. Related limitations mainly concern the limited fatigue scenarios and small-scale dataset, which are expected to be enhanced in future research.