COMPLAS 2027

IS046 - Deep Learning and Phase-Field Methods for Fracture, Damage and Failure Mechanics

Organized by: J. Ghaffari Motlagh (Leeds Beckett University, United Kingdom)
Keywords: (Multi)phase-field method, Damage and Fracture, Fracture and failure, physics-informed modeling, Scientific machine learning
Phase-field modelling has emerged as a powerful computational framework for simulating complex fracture processes without explicit tracking of evolving crack surfaces. Despite its versatility in capturing crack initiation, propagation, branching and coalescence, conventional phase-field approaches remain computationally demanding, particularly for three-dimensional, multiscale and fatigue-fracture problems. At the same time, rapid developments in deep learning and scientific machine learning are creating new opportunities to enhance the efficiency, scalability and predictive capability of phase-field fracture models. This invited minisymposium focuses on recent developments at the intersection of deep learning and phase-field modelling for fracture, damage and failure mechanics. Particular attention will be given to deep neural network formulations for solving phase-field fracture problems, including deep energy methods, physics-informed neural networks (PINNs), neural operators and other physics-informed learning strategies. Contributions exploring hybrid approaches in which machine learning is integrated with conventional finite-element phase-field formulations are also particularly encouraged. The minisymposium will address both brittle and cohesive fracture and welcomes extensions to ductile fracture, fatigue, heterogeneous and composite materials, and coupled multiphysics failure. Topics of interest include deep-learning-based solution of coupled displacement–phase-field equations, data-driven identification of phase-field parameters and constitutive behaviour, surrogate and reduced-order models, neural operators for rapid fracture prediction, adaptive learning strategies, and computational acceleration of large-scale phase-field simulations. A central objective is to examine how deep learning can overcome some of the computational limitations of conventional phase-field methods while retaining physical consistency, robustness and generalisability. Contributions addressing accuracy, convergence, interpretability, training efficiency and comparisons with established finite-element solutions and experimental observations are especially welcome. By bringing together researchers in phase-field fracture, computational mechanics and scientific machine learning, the minisymposium aims to promote new computational paradigms for efficient and physically informed prediction of fracture, damage and structural failure.