Field-Validated Digital Twin–Enabled Structural Health Monitoring for Offshore High-Pile Wharves

Abstract

Offshore high-pile wharves operate under harsh, variable loads that can mask the earliest signs of deterioration. We present a closed-loop structural health monitoring (SHM) framework that couples multimodal machine/deep learning with a physics-based digital twin (DT). A year-long field deployment at an operating high-pile wharf in Western Asia instrumented six piles with triaxial accelerometers and strain gauges (GPS-synchronized, 100 Hz). After quality control and environmental/operational variability normalization, an unsupervised VAE produces an anomaly score that is fused with a CNN–GNN classifier to yield window-level damage probabilities and pile-by-elevation localization. The DT assimilates these diagnostics through weighted least squares and an EKF, with measurement covariance derived from model uncertainty; “what-if” simulations quantify the structural consequence of candidate repairs. Under chronological splits with leave-one-pile-out, the framework achieves F1 = 0.95 and AUC-ROC = 0.98, with false-alarm rate ≈ 2–5 % and median detection latency ≈ 45 s. Spatial heatmaps consistently pinpoint the affected elevation, while online updating reduces mean frequency bias from ≈ 4.8 % to ≈ 0.9 % and increases mode-shape MAC from ≈ 0.81 to ≥ 0.92. First-mode shifts of ≈ 3–4 % observed in the field are reproduced by DT “stiffness-restoration” scenarios, enabling risk-informed maintenance planning. Compared on identical normalized windows, the proposed method outperforms classical vibration-based indices in both discrimination and false-alarm control. The results demonstrate a scalable, explainable pathway to predictive asset management for critical maritime infrastructure. Beyond structural safety, the proposed framework contributes to sustainable port operation by enabling condition-based maintenance, reducing unnecessary inspections, extending asset service life, and lowering the embodied carbon associated with premature repair or replacement.