Use of Artificial Neural Network to Identify Turbidite Deposit Gradation Generated in Laboratory

Manica, Rafael¹, Ana Luiza de Oliveira Borges Borges¹ (1) Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil

Artificial Neural Networks (ANN) are intelligent computational systems inspired by the human biological system capable of reproduce the human brain ability in solving problem tasks. Based this, the present study aims to introduce this complementary non-subjective tool in order to identify turbidity depositional patterns inferred from physical simulation. In this case, three different types of bed gradation patterns which are usually found in natural turbidity systems (normal grading, inverse grading and massive grading). The procedure employed digital images of the deposits generated from physical simulation. These images were also used to training (pattern learning), validating (efficiency checking) and simulating the ANN. Thus, after 18 turbidity currents experiments have been performed with all the three proposed types of bed grading patterns, it was observed that the ANN's results demonstrated accuracy factors ranged between 53-92% (mean of 76%) for normal grading, 52-99% (mean of 73%) for massive grading and 38-76% (mean of 53%) for inverse grading. The methodology introduced in this study, although quite simple, presented very concise results, which closely match up the results obtained through conventional analyzing tools (grain size analysis and qualitative observations) being an alternative tool to reduce uncertainty and subjectivity in physical simulation analysis. Furthermore, this non-intrusive methodology (based in digital images only) can be extrapolated to field observations allowing the classification of several turbidity facies and/or sedimentological process in natural sites