PolyU unveils multi-tier AI bridge inspection system deployed on 11 Hong Kong bridges

A research team at The Hong Kong Polytechnic University has developed a multi-tier intelligent bridge inspection system that detects surface cracks and hidden structural flaws that are not visible to the human eye. The project is led by Prof. Tarek Zayed of PolyU’s Department of Building and Real Estate. 

The system has been deployed across 11 local bridges. It integrates PolyU’s proprietary non-destructive testing technologies with advanced AI models, reducing inspection time by 50% and improving overall accuracy to 80%

The approach is designed to replace traditional visual inspection with three tools: drones for surface data capture, ground-penetrating radar to locate rebars and potential corrosion zones, and infrared thermography to identify deterioration such as spalling and delamination. 

For surface crack detection, the team uses a drone-enabled visual inspection workflow and a self-developed deep convolutional neural network model, Smart Bridge Deck Efficiency (SBDE). The model demonstrated superior crack identification in challenging conditions and reduced false positives compared with other leading object-detection models. 

For subsurface assessment, the fully automated GPR interpretation model locates rebars with over 98% precision and maps potential corrosion zones using amplitude analysis and clustering, streamlining corrosion assessment. 

For delamination and related deterioration, the team developed an optimum thermal gradient threshold (OTGT) system for infrared thermography, adjusting thresholds based on environmental conditions and automatically generating delamination maps. 

The system standardizes inspection results with a five-point severity scale to facilitate diagnosis and prioritization of repairs. The team is exploring collaboration with relevant government departments and industry partners for regular inspections in Hong Kong. 

The two-year project was supported by the Smart Traffic Fund. Findings have been published in journals including Construction and Building Materials, Automation in Construction, and Advanced Engineering Informatics.