Figures at Thesis_RAMIS/Figs_PI related with PhD Thesis:
AN INFORMATION-THEORETIC SAMPLING STRATEGY FOR THE RECOVERY OF GEOLOGICAL IMAGES: MODELING, ANALYSIS, AND IMPLEMENTATION
Data for the LaTeX version of the document
In this thesis the role of preferential sampling has been systematically addressed for the task of geological facies recovery using multiple-point simulation (\emph{MPS}) and for the problem of short-term planning in mining. In the context of facies recovery using simulations, the task of optimal sampling is formalized and addressed using a maximum information extraction criterion. This sampling principle has the ability to locate samples adaptively on the positions that extract maximum information for the objective of resolving the underlying field. A formal justification is provided in this thesis for adopting this information-driven sampling criterion as well as concrete ways of implementing this principle in practice. In addition, the practical benefits for \emph{MPS} in the context of simulating channelized facies models is demonstrated using synthetic data and real geological facies. Importantly, this strategy locates samples adaptively on the transition between facies which improves the performance of conventional \emph{MPS} algorithms. In conclusion, this work shows that preferential sampling can contribute in \emph{MPS} even at very small sampling regimes and, as a corollary, demonstrates that prior models (obtained form a training image) can be used effectively not only to simulate non-sensed variables of the field, but to decide where to measure next.
Furthermore, the proposed sampling strategy has been adapted to the problem of short-term planning for the task of classifying blocks to be processed as waste or ore in the production stage of a mining project. The problem has been formalized using the principle of maximum information extraction criterion and the obtained solutions was validated using three data sets of real mining projects. Importantly, the proposed methodology takes advantage of the information available from the previously sampled locations, allowing to improve the performance as compared with some of the classical non-adaptive sampling schemes used for advanced drilling tasks. From the results obtained across these three real scenarios explored in this thesis, it is possible to see that the proposed methodology achieves better performances than ...