Automatic Selection of Fracture Sets Using Clustering Techniques

Fracture set classification is an important task in the areas of mining, hydrocarbon industry, coal-bed methane production, carbon capture and storage (CCS), geothermal energy, nuclear waste storage. Traditional classification methods are supported by visual inspection of stereonets, which is subjective and sometimes error-prone. Recently several clustering techniques have been developed using data analysis methods; however, one important parameter that must be defined in order to perform clustering algorithms is the selection of the number of clusters present in the analyzed data. This parameter is defined subjectively by an expert or by forcing a search for a limited number of clusters (e.g. from 2 to n). For both, the strategy consists in performing a finite number of clustering runs (each with different number of clusters) and choosing the one with the best performance based on an index. In this paper, we propose a preprocessing method in order to choose the number of clusters and also to initialize clusters centers for the K-means and Fuzzy K-means algorithms. The method is based on calculating the poles density in the stereonet, and defining a threshold value in order to select the clusters prototypes. Additionally, a post process to estimate area of influence for candidates is performed in order to deal with large areas of high-density poles.