Title:
Generative Artificial Intelligence for Smart Characterization of Complex Cracks in Strain-Hardening Cementitious Composites (SHCCs)
Author(s):
Meng
Publication:
Web Session
Volume:
ws_S25_Meng.pdf
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
3/30/2025
Abstract:
The accurate identification of cracks is crucial in the research and practical use of strain-hardening cementitious composites (SHCC). The rise of deep learning techniques in computer vision has introduced efficient ways to automate crack detection processes. However, creating dataset for training these deep learning models demands a lot of effort and time, a situation worsened by intricate crack patterns. This study introduces a novel method using a hybrid generative adversarial network (HGAN) to simplify the task of detecting complex cracks. HGAN combines the strengths of deep convolutional generative adversarial network (DCGAN) and conditional generative adversarial network (CGAN), offering a solution for evaluating SHCC characterized by dense microcracks and conventional concrete with simpler cracks. Our findings demonstrate the method's effectiveness for SHCC with dense and microcracks, leading to enhanced precision in crack characterization, with an F1 score and intersection-over-union (IOU) for SHCC crack segmentation at 0.982 and 0.980, respectively.