Synergistic approach to quantifying information on a crack-based network in loess/water material composites using deep learning and network science

• Published in: Computational materials science Vol. 166; pp. 240 - 250
• Main Authors: Hwang, Heesu, Oh, Jiwon, Lee, Keon-Hee, Cha, Jung-Hwan, Choi, Eunsoo, Yoon, Young, Hwang, Jin-Ha
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01-08-2019
• Subjects: Complex networks; Convolutional neural networks; Crack detection; Deep learning; Deep learning; Crack detection; Convolutional neural networks; Complex networks
• ISSN: 0927-0256; 1879-0801
• DOI: 10.1016/j.commatsci.2019.04.014

[Display omitted]
•Deep learning and network science are synergistically applied to the crack networks.•Network science is employed to quantify the interconnected features between nodes.•CNN and YOLO algorithms are proven to be effective in detecting the nodes and edges.
Deep learning and network science are applied in a synergistic manner to address structural crack issues with the aim of providing the characteristic features of crack generation and a quantitative description of crack networks in natural materials. Loess/water mixtures were chosen as a model system due to the facile formation of cracks resulting from water evaporation. Deep learning algorithms are applied to the detection and classification of edges and nodes in cracks forming in the drying stage of the loess/water mixture system. Deep learning is shown to effectively detect and classify cracks in terms of nodes and edges. Based on the guided information on nodes and edges, cracks were subject to a connectivity analysis with network science.The combined deep learning/network science approach is proven to be suitable for understanding crack formation and propagation in both qualitative and quantitative aspects.