Automated Unmanned Aerial Vehicle-Based Bridge Deck Delamination Detection and Quantification

Delamination is one of the most critical defects assessed during bridge deck inspections. Recently, infrared (IR) thermography has gained more attention for delamination detection since it provides fast and effective inspections with reasonable accuracy. However, point-by-point inspections with handheld IR cameras and manual data interpretation are still time consuming. In addition, manual data interpretation is highly dependent on the inspectors' experiences. To tackle these concerns, this study conducted investigations from two perspectives to improve IR-based delamination detection: (1) data collection and (2) data interpretation. In this study, unmanned aerial vehicles (UAVs) equipped with IR sensors have been deployed to perform automated inspection data collection. Various factors have been considered to develop a preliminary UAV-based IR data collection plan. The developed data collection plan has been implemented on a full-scale bridge deck specimen at the Bridge Evaluation and Accelerated Structural Testing (BEAST) facility. Discussions and suggestions based on the results have been provided. A pixel-level deep learning method is developed for automatic delamination detection and quantification of the bridge deck. IR data collected from four real bridge decks are pixel-wise labeled and used for model calibration. The accuracy and mean intersection over union achieve 99.36%, 97.96%, 97.83% and 0.98, 0.96, 0.95 for training, validation, and testing datasets, respectively. Furthermore, an easy-to-use tool is developed based on the proposed method for practical implementation. The developed tool is validated using the BEAST specimen data. The fast and accurate implementation of the developed tool makes it a promising option for autonomous bridge deck inspection.

Automated summary

This study improves the IR-based delamination detection method by focusing on data collection and data interpretation. Automated inspection data collection is achieved using UAVs equipped with IR sensors, while a pixel-level deep learning method is developed for automatic delamination detection. The results show that the proposed method is highly accurate and efficient.