MCMN Deep Learning Model for Precise Microcrack Detection in Various Materials

Damage in metals, composites, and cemented porous solids, in the form of cracks, inclusions, and voids, is a nontrivial problem. Many experimental, numerical, and analytical methods have been proposed in the past, with some recent models deploying neural networks. However, past methods often lack the accuracy and precision needed to identify microcracks. This paper presents the MicroCracksMetaNet50E (MCMN) deep learning model, inspired by Meta's Segment Anything Model (SAM). MCMN is trained with numerical data produced by an advanced mesoscale numerical model for spatial crack detection inside various materials. MicroCracksMetaNet50E achieves an accuracy of 0.867% and a precision of 0.906% in identifying microcracks. The robust performance of MCMN is highlighted, showcasing a notable advancement that its capabilities and propels the field into uncharted territories by expanding oppor-tunities for the comprehensive exploration of additional datasets. The method could be adopted for damage detection in metals and composites in manufacturing as well as structural health monitoring.