Hybrid objective function applied to optimize infill sampling location

• Ano de defesa: 2022
• Autor(a) principal: Ramos, Gustavo Zanco
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Infill; Kriging; Modeling.; não disponível; Objective function; Optimization; Simulation; Uncertainty
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/44/44137/tde-24082022-070621/

Different moments of the exploration of mineralized bodies demand that sampling infill be made, those new samples have the objective of furthering knowledge about mineralized rock grade distribution. Usually, drillholes collars are located by geologists with experience and knowledge about the domain under analysis. Other methodologies can be applied to help the decision of where to locate the drillholes, for example, optimization of the infill drillhole location. Optimization is a method to assess the best parametrization to solve a problem, in the case of the infill location the problem depends on what the new samples are made for. Some research utilizes the kriging variance to guide the location of the new samples but has a limitation in assessing the sample distribution uncertainty. Another method that can be applied to locate the infill samples is simulation variance, which is dependent on the sample value. The application of a compost objective function to optimize the infill location is tested. This compost function considers both models kriged and simulated to search for the optimal infill drillhole configuration, therefore, considering both the sample spatial distribution and uncertainty. This method is compared with the objective function that uses either the kriged or simulated data directly to assess the competence of the compost one. Another test considers the influence of the values associated with the samples while searching for the optimum location of drillholes. Those tests have proven that the use of the simulation alone fared better in locating the infill samples in synthetic data than the compost or the kriging-dependent objective function. Both objective functions that utilize direct models, either kriged or simulated, fared better in different distributions. Considering the values associated with the samples, the median fares better than the other 3 values, mean, P10, and P90 of the simulated block distribution. Regarding the methodology of the search is important to notice that optimizing the direction of the drillhole tends to have a better response regarding the objective function but more tests should be made. The optimized infill location tends to further the representativity of the original sampling after the drillholes are done, therefore it can help assess portions of the domain with higher uncertainty that should be considered when the infill location decision is being made.