Machine learning-accelerated peridynamics model for mechanical and failure behaviors of materials

Computational mechanics is essential for understanding and predicting complex material behaviors, particularly in areas such as material fracture mechanics and structural engineering. However, the high computational costs associated with traditional methods, especially for large-scale simulations, present significant challenges. Peridynamics (PD) offers a compelling alternative to classical continuum mechanics by effectively modeling discontinuities such as cracks. Despite its strengths, PD is computationally intensive, limiting its broader application. To address these challenges, we introduce a machine learning-accelerated PD model that significantly reduces computational time while maintaining high accuracy. Our method integrates a machine learning-based surrogate model trained on displacement field data, which efficiently approximates the behaviors of material points, bypassing the iterative processes of conventional PD simulations. This approach is validated through a series of benchmark tests, ranging from one-dimensional bars to three-dimensional beams, demonstrating speedups of over six times compared to traditional methods. The integration of machine learning with PD not only enhances computational efficiency but also expands the practical applicability of PD to large-scale engineering problems, making it a viable tool for a wide range of scientific and industrial applications.