Deep convolutional transformer network for quantifying crack width in tunnel lining structures using distributed fiber optic sensing data

Abstract Distributed fiber optic sensing (DFOS) technique provides distinct advantages for crack monitoring in infrastructure by measuring strain distribution. However, deriving crack width from measured strain data is challenging due to their complex nonlinear mapping relationship. To address this issue, this paper proposes a deep learning (DL)-based method for crack width quantification in tunnel lining structures using strain data measured by DFOS. First, simplified lining segments were cast and subjected to destructive eccentric loading tests, during which strain distributions were collected using DFOS sensors. Afterward, the collected strain sequences were appropriately segmented and labeled with corresponding crack width values to form the sample dataset. Importantly, this paper developed a novel DL framework called deep convolutional transformer network (DCT-Net), which is capable of extracting local and global sensitive features from strain data for crack width quantification. The effectiveness, noise robustness and generalization ability of the proposed DCT-Net were extensively validated. Experimental results demonstrate that the proposed approach can accurately quantify crack widths in tunnel lining segments and exhibits strong generalization. In addition, the DCT-Net outperforms current five state-of-the-art DL models, particularly under strong noisy conditions. This study will pave the way for future application of DFOS technique for intelligent monitoring and quantification of cracks in tunnel lining structures in in-situ engineering projects.