Vibration-based damage detection in cable-stayed bridges using a novel 1D-ConvNeXt-LSTM network

Structural health monitoring (SHM) plays a crucial role in maintaining the safety and serviceability of civil infrastructure, such as cable-stayed bridges. However, simultaneously extracting detailed local signal features and long-range temporal dependencies from vibration data remains a significant challenge for standalone deep learning architectures like LSTM or ResNet1D. To address this limitation, this study proposes an advanced hybrid architecture, 1D-ConvNeXt-LSTM, for structural damage detection. This framework integrates the multi-scale feature extraction capabilities of 1D-ConvNeXt with the sequential modeling proficiency of Long Short-Term Memory (LSTM) networks. The proposed method was evaluated using vibration time-series data acquired from a scaled cable-stayed bridge model under five distinct simulated damage scenarios. Experimental results demonstrate that the 1D-ConvNeXt-LSTM model achieves superior classification performance, yielding a macro–Area Under the Curve (AUC) of 0.971 and a macro F1-score of 0.814, significantly outperforming baseline architectures including FCN, ResNet1D, and LSTM. Ultimately, the proposed architecture provides a robust, stable, and highly accurate solution for structural condition assessment, thereby enhancing the effectiveness of damage identification in practical SHM applications.