Long-Span Bridge Structural Health Monitoring & Digital Twin

Background

Operation and maintenance of long-span suspension bridges is a major engineering challenge. Traditional methods rely on periodic manual inspection and cannot capture damage evolution in real time. A digital twin combining physics models with measurements enables real-time health assessment and warning.

Approach

Physics model

• Elasticity equations (linear / geometrically nonlinear)
• Modal analysis (natural frequencies, mode shapes)
• Fatigue accumulation (Rainflow + S-N curve)

Data-driven layer

• Measured strain/acceleration (SHM sensor network)
• PINN physics constraints: loss = PDE residual + measurement residual + BC
• Graph Neural Network (GNN): bridge spatial topology encoding

Key results

• Damage localization accuracy: 92% at 5% noise
• Remaining-life prediction error: < 15%
• Inference: real-time (~1 Hz), suited to online warning

Use cases

• SHM systems for long bridges
• Rapid damage assessment after extreme events (typhoon/earthquake)
• Digital-twin platform integration (with BIM)

Resources

• Kimda Bridge dataset (Korea, 10-year monitoring)
• IASC-ASCE 6-story steel-frame SHM benchmark