AI-Driven Digital Twin for Predictive Lifecycle Management of Bridge Infrastructure: A Review, Conceptual Framework, and Proof-of-Concept Evaluation
Abstract
Bridge infrastructure is increasingly impacted by age, growth in traffic volume and loading, environmental variability (i.e. climate change), extreme weather events occurrence, and limited maintenance funds. Standard practice for bridge-management systems primarily relies on periodic inspection and large-scale condition ratings, while ignoring the need for detailed models capable of predicting sudden structural state changes or the uncertainty in future deterioration. ProposalWORK presents: a systematic review, a six-level artificial intelligence operational digital twin modelling framework and is evaluated via proof concept for predictive bridge life cycle management. The framework connects the physical bridge with sensing, inspection imagery, traffic and environmental observations, bridge information modelling, geographic information systems, finite-element analysis, semantic data integration, artificial intelligence, lifecycle-risk assessment, and governed maintenance workflows. A hybrid physics-data approach is proposed for quality control, anomaly screening, damage diagnosis, virtual sensing, remaining-useful-life forecasting, maintenance optimization, uncertainty quantification, and explainable human decision support. The experimental demonstration used an ageing-bridge structural-health-monitoring time-series dataset containing 1,340 observations and 53 engineered numerical predictors. The 30-day degradation forecast achieved MAE = 8.9210, RMSE = 10.8718, and R² = 0.7851, whereas current-state degradation prediction produced R² = -0.0480 and structural-condition classification achieved accuracy = 0.3433 and weighted F1 = 0.2751, with no Severe-class detections.
