+441518081136Digital Petrography: Mineralogy and porosity identification using machine learning models
2nd International Conference and Expo on Oil & Gas
December 02-03, 2019 | Dubai, UAE
Rafael Andrello Rubo
Universidade de Sao Paulo, Brazil
Keynote : J Ind Environ Chem
DOI: 10.35841/2591-7331-C3-013
Abstract:
Microscopic petrographic analysis allows evaluating
depositional environments and diagenetic processes
from sedimentary basins. The results of these analyses,
therefore, contribute to an enhanced reservoir characterization
and guide the oil and gas exploration.
Images acquired from thin sections in the visible spectrum
are used as input to create three sorts of machine learning
models: 1. Image segmentation models with representative
classes of the rock mineralogy and porosity; 2. Object
detection models to automatically identify features of interest,
such as phosphatic fragments; and 3. Classification models for
labeling images with different porosity types. Systematical
application of these models in new images standardizes
descriptions and reduces subjectivity and human errors
during thin section analysis. Convolutional filters were applied
in all the models, followed by machine learning classification
algorithms, such as artificial neural networks and random
forest. Datasets used for training are from thin sections of
carbonate rocks, which are prepared from sidewall core
samples of oil wells, specifically from the pre-salt reservoirs
of Santos Basin, on the southeast coast of Brazil. Evaluation
of the models’ abilities to generalize is done through the use
of 10-fold cross-validation tests and by correlation with other
sources of data, such as chemical microanalysis. Results show
high percentages of correctly classified instances during crossvalidation.
Correlations indicate low root mean square errors
and elevated coefficients of determination.
Biography:
Rafael Andrello Rubo is a geologist working at Petrobras, a Brazilian multinational energy corporation, where he conducts stratigraphic studies applying data science. He was graduated from Universidade Estadual de Campinas and participated in an exchange program at University of Missouri. He is pursuing a master’s degree in Petroleum Engineering at Universidade de São Paulo. He has also taken an MBA course at Fundação Getúlio Vargas in Finances and Investment Management, and he is also post-graduated in Mineral Metallurgical Systems at Universidade Federal de Ouro Preto.
E-mail: rafaelrubo@gmail.com
PDF HTML