RUB » Bibliotheksportal

Tehaseen Mujawar

• The evaluation and inspection of structures, which use asphalt mix on decks like roads, bridges, parking lots, and driveways involve a comprehensive assessment by engineers, inspectors, and professionals. Air voids in these asphalt mixtures are a critical factor in measuring their strength and performance as they play a significant role in determining the resistance of the mixtures towards major pavement distresses such as rutting, fatigue cracking, and low-temperature cracking [1]. This study proposes a novel approach for detecting air voids in asphalt mixtures using state-of-the-art techniques for instance segmentation of CT scans. The proposed technique can automate this, can reduce the efforts of manual annotation, and may increase the quality of asphalt mixtures produced. The significance of this study lies in the potential for automated quality checks and early detection of asphalt structure deterioration. An automated process is used for data annotation. The annotated data is used to train Deep Learning models, specifically the DeepLabV3+ and the Mask R-CNN model. Optimization of hyper-parameters is performed using a grid search approach to detect air voids in CT scans. The trained model is able to differentiate between instances of the same class, which is necessary for detecting air voids.