Digital twin-based fatigue life assessment of orthotropic steel bridge decks using inspection robot and deep learning

Abstract

Fatigue cracks are a major issue affecting the lifespan and operation and maintenance (O&M) costs of bridges with orthotropic steel decks (OSDs), while current practices for detecting fatigue cracks often rely on manual inspection with time inefficiency. This paper presents a digital twin framework that employs robots equipped with nondestructive testing devices for data collection and deep learning algorithms for data analytics, aiming to enable automatic detection of cracks and assessment of fatigue life. Inspected crack are fed into a finite element model constructed via ABAQUS-FRANC3D co-simulation to conduct fatigue life analysis, and an MLE-PCE-Kriging surrogate modeling technique is developed to facilitate rapid assessment of fatigue life. The deep learning-based crack detection achieves accuracy and recall of 95.6 % and 92.2 %, respectively, while the MLR-PCE-Kriging model exhibits an MPAE of 2 %, demonstrating high accuracy. The proposed digital twin framework can guide automated bridge inspection, thereby promoting intelligent O&M management for bridges.