Machine Learning Based Surface Crack Detection

Abstract

Crack is one among crucial damages on concrete surface. To assess the
performance and structural health of a concrete structure, it is important to determine the
presence, location, and geometry of cracks. Currently surface cracks are evaluated through
manual inspection, which totally depends on the skill and experience of trained personnel. As
well as manual inspection for surface cracks is time-consuming. Due to these shortcomings,
recent research activity regarding crack detection concentrates primarily on automated crack
detection techniques. This paper proposed a model for crack detection using machine
learning. The process involves AutoML Vision Edge to train the model. For which the
datasets were prepared, then separated into two parts (85% for training and 15% for tests) and
labeled as according to type of crack. Then trained model were tested using test dataset,
which were not used during training. The result indicates that the introduced model is able of
detecting crack.