A New Multi-Modal Data Fusion Framework for Delamination Detection in Concrete Bridge Decks

Area of Science:

• Civil Engineering
• Structural Health Monitoring
• Non-Destructive Testing

Background:

• Bridge decks degrade due to environmental exposure, traffic, and aging, leading to delamination.
• Subsurface concrete delamination and steel corrosion compromise structural integrity and public safety.
• Existing non-destructive evaluation techniques like Ground-Penetrating Radar (GPR) and Infrared Thermography (IRT) have limitations in data fusion.

Purpose of the Study:

• To introduce a Physics-Enhanced Multi-Modal Fusion (PE-MMF) framework for improved delamination detection in reinforced concrete bridge decks.
• To address the challenges of integrating GPR and IRT data due to differences in sensing principles and resolution.
• To enhance the generalization capability of delamination detection across different bridge structures.

Main Methods:

• Developed a PE-MMF framework utilizing transfer learning, cross-modal attention, and gated fusion.
• Integrated a systematic feature selection protocol for identifying physically consistent indicators.
• Trained and validated the framework on the publicly available SDNET2021 dataset with co-registered GPR and IRT measurements.

Main Results:

• Achieved substantial performance improvements in delamination detection.
• Demonstrated average F1-score gains of up to 55% over IRT-based methods and 25% over GPR-based methods.
• Confirmed superior generalization capability of the multi-modal approach compared to single-modality methods.

Conclusions:

• The PE-MMF framework offers a scalable and data-efficient decision-support tool for infrastructure monitoring.
• Deep learning-based sensor fusion can effectively prioritize areas for detailed physical investigation.
• The study highlights the potential of integrated GPR and IRT for robust delamination detection in bridge decks.