Mesh generation and optimization from digital rock fractures based on neural style transfer

The complex geometric features of subsurface fractures at different scales makes mesh generation challenging and/or expensive. In this paper, we make use neural style transfer (NST), a machine learning technique, to generate from rock fracture images. new approach, digital multiple that represent ‘content’ and define uniformly shaped sized triangles ‘style’. 19-layer convolutional network (CNN) learns the content image, including lower-level (such as edges corners) higher-level rock, fractures, or other mineral fillings), triangular grids. By optimizing cost function achieve approximation both style, numerical meshes can be generated optimized. We utilize NST for rough with asperities formed in embedded sand aggregate grains. Based on examples, show technique optimization much more efficient by achieving good balance between density presentation features. Finally, discuss future applications approach perspectives applying bridge gaps modeling experiments.