Image-based learning and experimental verification of crack propagation in random multi-fractures rock

Fractures and the rock matrix are fundamental components of rock masses, with their random distribution being a common characteristic. Previous studies often focus on regularly fractured rock samples to facilitate experimental and numerical analysis. However, the limited number of samples used in these studies hinders a comprehensive understanding of the mechanical properties and failure characteristics of fractured rock masses. In this paper, we implement batch numerical simulations using PFC (Particle Flow Code) with an original automatic control code, resulting in a dataset of 400 numerical simulation results. The crack propagation characteristics and failure parameters of rock mass with random multi-fractures have been studied by using GANs (Generative Adversarial Networks) and other neural network models. Randomly fractured granite samples were subjected to uniaxial compression loadings, and the evolution of the strain field was analyzed by applying digital image processing technology. The testing results were then compared with the training model results. After verifying the model’s accuracy, the obtained CNN (Convolutional Neural Networks) model can be used to predict the UCS (Uniaxial Compressive Strength) of the real experimental samples. Additionally, we analyzed and discussed the influence of various parameters of random fractured rock mass on its bearing capacity.