Surface cracks in reinforced concrete (RC) bridge structural elements pose significant concerns regarding durability, strength, and serviceability. Micro-cracks, if left unchecked, can propagate into macro-cracks due to several factors leading to structural health deteriorations and increased maintenance costs. Generative Adversarial Networks (GANs), which leverage Convolutional Neural Networks (CNNs), provide a powerful alternative by learning from unstructured image datasets and generating realistic visual outputs. This study presents a novel GAN-based mechanism referred to as CrackGAN for predicting the propagation patterns of cracks in RC bridge elements. The model synthesizes realistic crack growth patterns and allows for predictive visualization of their evolution. Validation of predicted results were validated by comparing them with real time crack propagation recorded in laboratory tests on RC beams and slabs. The results were evaluated using Structural Similarity Index Measure (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS), demonstrating strong agreement between predicted and experimentally observed crack propagation.
Khan et al. (Sat,) studied this question.