Deep Learning-Based Entrapped Air Segmentation And Evaluation (EASE) for Plain Concrete Pavement Applications

AUTHORS: Yan, X.; Darnell, M.; Vandenbossche, J.M.; and Fascetti, A.

ABSTRACT: The quality of the consolidation achieved during the paving process affects the amount of entrapped air that will remain in the concrete and hence the service life of the pavement. Petrographic methods traditionally used for the analysis of entrained air systems could be used for characterizing entrapped air but requires time consuming specimen preparation by skilled technicians. Data-based methods, such as Convolutional Neural Networks (CNNs), have been proposed in recent years for the analysis of entrained air systems as an attempt to automate the process and standardize results. However, such methods still require time-consuming preprocessing of the digital images as well as large amounts of data to train the models. In this study, a novel entrapped air recognition toolset, called Entrapped Air Segmentation and Evaluation (EASE), is therefore proposed. The main tool used in the procedure is the recently developed Segment Anything Model, which is a pre-trained, data-centered machine learning model, which demonstrates superior accuracy in the application area of image recognition and segmentation. Subsequently, based on the spatial analysis of the segmented entrapped air, a novel statistical procedure, based on the nearest neighbor, and the principle of complete spatial randomness, is proposed to quantitatively characterize the spatial distribution of entrapped air in concrete pavements. Results demonstrate that EASE is a robust tool to quantify the spatial features of entrapped air voids in plain concrete pavements, and to formally identify potential correlation of the air voids structure with aggregate particles.