Machine learning-augmented finite element modeling for fracture mechanics: A systematic review and future research directions

Predicting crack initiation and growth in safety‑critical structures remains a core challenge in computational fracture mechanics. Conventional finite element methods (FEM), while foundational, are hindered by prohibitive computational cost, mesh sensitivity, and difficulty with multiscale damage simulation. Emerging machine learning-finite element method integration (ML-FEM coupling) offers a new approach, delivering rapid surrogate modeling for crack propagation, physics‑informed neural networks (PINNs), and hybrid digital twin for fracture prediction frameworks that retain physical fidelity while achieving substantial speed‑up. Quantitative cross‑comparison reveals high performance for physics‑informed hybrids and distinct trends across model families. This systematic review provides the first combined scientometric and thematic synthesis of ML‑augmented FEM in fracture mechanics, categorizing advances into surrogate modeling, hybrid modeling in engineering mechanics, neural operator learning, and autonomous frameworks. Applications span structural health monitoring (SHM), composite manufacturing optimization, and biomechanical/geo‑mechanical fracture prediction. Ongoing challenges, data scarcity, uncertainty quantification in fracture analysis, scalability, are examined, and a research roadmap is proposed toward interpretable, transferable, and real‑time ML-FEM systems for damage‑tolerant design, predictive maintenance, and engineering safety.