Physics-Guided Spatiotemporal DAS Representation for Leakage Aperture Identification and Unseen Intermediate-Aperture Prediction

Distributed acoustic sensing (DAS) based on phase-sensitive optical time domain reflectometry provides continuous vibration sensing along optical fibers and a scalable sensing layer for Internet of Things (IoT)-enabled infrastructure monitoring. However, existing DAS-based pipeline monitoring mainly addresses leakage detection, localization, or classification under predefined conditions, while quantitative severity assessment and prediction of intermediate aperture states excluded from model development remain limited. This study proposes a physics-guided spatiotemporal DAS framework for leakage-aperture identification and unseen intermediate-aperture prediction. A finite-element model with aperture-dependent localized excitation is first developed to reveal leakage-induced spatial responses and identify a leakage-sensitive region. Spatiotemporal statistical representations are then constructed from distributed DAS phase responses, and an anchor-based evaluation protocol is introduced by excluding intermediate apertures from model development. Fourteen leakage apertures from 0.5 to 4.0 mm are experimentally investigated in a water-circulation pipeline. The proposed representation improves the Macro-F1 score from 0.76 to approximately 0.94 and reduces severe cross-level misclassifications to approximately 1%. For six held-out apertures, the overlapping-window radial basis function support vector regression (RBF-SVR) model achieves an R2of 0.902, a mean absolute error (MAE) of 0.207 mm, and a root-mean-square error (RMSE) of 0.247 mm over 82 held-out 10 s DAS acquisition records. These results demonstrate the potential of the proposed framework for fine-grained leakage-aperture identification, unseen intermediate-aperture prediction, and IoT-enabled pipeline integrity monitoring.