| The physical and mechanical properties and seepage characteristics of the rock mass are greatly influenced by the number,occurrence,surface features and distribution of the fracture.In the rock mass,a large number of fractures are partially invisible or completely invisible and they form a complex network,leading to it difficult to study the complex fracture system in the nature.To solve this problem,many researchers choose to study single fracture and learn the influence laws of various single factors and their combinations on the rock properties.Among them,one of the key steps is to establish the model of the single fracture and quantify the roughness of the fracture surface.However,it usually requires a lot of manpower and material resources to finish this step based on manual mapping,mold turning,laser scanning and photogrammetry and other modelling methods.Meanwhile,it can only obtain the exposed local surface information and engineering parameters during the field sampling of the fracture,which cannot meet the needs of practical engineering analysis.Fortunately,with the development of the study on fracture surface roughness parameters and machine learning technology,it is possible to deduce the complete fracture surface information and quantity the roughness of the fracture surface based on a small amount of information.Therefore,the aim of this study is to put forward a new 3D fracture surface generation algorithm.The characteristic indexes of the fracture surface are then summarized based on the evaluation criteria of the permeability,which can be used to improve the algorithm.This algorithm can be applied to predict the roughness of the whole fracture surface based on local fracture surface roughness and generate a large number of fracture surface model with designated permeability.The specific content and main achievements of this paper are as follows:(1)A basic 3D fracture surface generation algorithm inherits the local features of sample model is established based on the theory of Markov Random Field and single fracture sample model with pixel accuracy.Specifically,a granite single fracture digitized model is created and its surface elevation matrix is extracted as a sample for reference and a large number of fracture surface models with similar local features to the sample model are generated based on Markov Random Field.(2)The factors named feature indexes influence the permeability of the single fracture are summarized at the mesoscale and macroscale,respectively,which can provide theoretical guidance for the optimization of the basic algorithm.The seepage numerical simulation of the sample model and new models generated by the algorithm is carried out.The models with the permeability similar to that of sample model are then screened out to study their common features at different scales.At mesoscale,the influence of roughness indexes of the qualified models on permeability is studied by regression analysis.The results show that the power spectral density fractal dimension and center-line average asperity height are defined as mesoscopic roughness feature indexes,which affects the permeability of the single fracture.At macroscale,SIFT algorithm is used to calculate the feature points of each block region of the fracture surface and match them with the qualified models.The block regions successfully match with most of qualified models are extracted and defined as macroscopic geometric feature indexes.(3)The engineering applications of the basic algorithm and optimization theory are shown in different situations and their validation are checked.Based on the information of local fracture surface in engineering field,several complete fracture surface models with different sizes and similar roughness characteristics with local fracture surface are established using the basic algorithm.Based on the permeability measured in the engineering field,a large number of complete fracture surface models are generated using the optimization algorithm.The seepage numerical simulation results show that a large number of models can meet the requirement of the permeability after the optimization based on feature indexes.Based on the local fracture surface information and permeability,a large number of complete fracture surface models are generated using the quadratic optimization algorithm.The results show that most of generated models can not only inherit the roughness characteristics of the local fracture surface,but also meets the permeability requirements of fracture surfaces. |
