A systematic 3D simulation method for geomaterials with block inclusions from image recognition to fracturing modelling

A systematic hybrid modelling approach for heterogeneous geomaterials with irregular block inclusions is creatively developed based on a deep learning technique, computational geometry algorithms, and a 3D finite discrete (or discrete-finite) element method; the approach includes the following three major steps: (1) the deep learning-based image identification technique and the computational geometry algorithm are employed to establish a 2D geometry library of realistic rock blocks; (2) 3D block inclusions with desired block shapes are regenerated by a surface morphing technique and then randomly allocated to the specimen domain based on the overlapping detection algorithm; and (3) the finite-discrete element method is developed by integrating cohesive elements with a solid mesh based on a finite element code to simulate the progressive fracture and interface behaviours of heterogeneous geomaterials. To validate the proposed hybrid approach, a series of synthetic specimens with Brazilian split tests are prepared and implemented from 2D to 3D. The results verified that the finite-discrete model can be easily established through images, and the consequent simulation performance is validated through comparisons between observations and numerical results regarding failure patterns and stress strain relations. Using the calibrated and verified approach, we further numerically discuss the influence of the block-matrix strength ratio and interface strength on the mechanical responses of bimrocks. All results demonstrate that the proposed hybrid approach has a powerful ability to dealing with heterogeneous composite materials that maintain the characteristics of both continuity and discontinuity.