Recent Advances of Machine Learning in Fracture Mechanics of Quasi-Brittle Materials

The fracture mechanics of quasi-brittle materials, such as concrete, ceramics, and rocks, pose significant challenges due to their nonlinear stress-strain response, microstructural heterogeneity, and complex failure mechanisms. Traditional numerical and analytical methods often fall short in capturing the full intricacies of fracture propagation and damage evolution in such materials. However, recent advances in machine learning (ML) offer promising solutions to these limitations by enabling data-driven insights and enhanced computational performance. In this review paper, we explore the growing role of ML techniques in the fracture analysis of quasi-brittle materials. By leveraging large and diverse datasets from experiments and numerical simulations, ML models not only complement traditional fracture mechanics approaches but also introduce novel capabilities such as real-time damage prediction, adaptive modelling, and improved generalization across varying material conditions. The integration of data-driven models with physics-based frameworks, especially through hybrid techniques like physics-informed neural networks (PINNs), marks a significant shift in how fracture phenomena are modelled and understood. Key ML methods discussed include artificial neural networks (ANNs), convolutional neural networks (CNNs), and PINNs, with a focus on their respective advantages and implementation strategies. The review highlights how these approaches can enhance safety, efficiency, and predictive accuracy in engineering applications, ultimately making machine learning a transformative tool in the study of quasi-brittle fracture behaviour.