AUTOMATING OREBODY MODELING USING COST-EFFECTIVE ADAPTIVE ALPHA-SHAPE CONSTRUCTION

Modeling mineral deposit plays a critical role for exploration and mine planning, for which a number of methods have been developed for efficiently and reliably estimating ore body shapes. However, existing methods often rely heavily on the human curation through manually adjusting the key parameters of the estimation models. Although computer-aided modelling is widely used in industry, the results can be severely affected by the subjectivity due to human interventions and finding the optimal model parameters is often time-consuming.

To tackle these challenges, we propose a novel adaptive alpha-shape method to estimate ore body shapes from input point clouds using Poisson surface constructions. To improve the efficiency for computing high-resolution shapes, we first create an octree mesh and populate its nodes distributed on a nonuniform grid, because this can decrease the grid size at the regions with the high point density. We then estimate the inward normal vectors of possible surfaces within the input points such that as many points as possible are placed on one side of the surface locally. From the estimated normals, we solve the Poisson problem to find the scalar field whose gradient can best approximate the normals of the input points. The scalar field is used to construct the isosurface that populates the nodes of the octree. This new set of points on the isosurface can interpolate the grids between the input data points. The high-resolution ore body shape is then constructed over the interpolated points using the alpha-shape method that creates meshes surrounding the input point cloud with details decided by the parameter of alpha-radius. Instead of using the fixed alpha-radius in traditional alpha-shape method, we also develop a method that adapts the alpha-radius to the local point densities. Our method requires minimal manual intervention and can better describe shape details. We test the proposed method on the large drill-hole dataset from a volcanogenic massive sulfide deposit to estimate the ore body shape of precious metals. We observe that the proposed method can generate the high-resolution shape without the costly and manual parameter searches.