Nuclear magnetic resonance evaluation using independent component analysis (ica)-based blind source separation

Abstract

Disclosed is a non-lineal statistical independent component (ICA) analysis methodology for calculating T2 or T1 distributions of nuclear magnetic resonance logs. In one aspect, the invention employs a classical blind source separation (BSS) approach with the input data (T2 or T1 distributions) being considered not only horizontally (in relaxation time units), but also vertically (in depth). The statistical variations are used for separating the principal independent components and their corresponding weighting matrix. The result of such ICA based BSS is an efficient separation of T2 components correlative to the presence of particular conditions (e.g., clay bound water, heavy oil, capillary bound water, free water, mud filtrate (water and oil), and noise). Individual saturation of estimated fluids can be calculated from the weighting matrix generated in accordance with the invention. In accordance with a further feature of the invention, it is contemplated that independent component analysis techniques may be applied to the underlying time domain data prior to its transformation to a T2 distribution. This advantageously results in 'de-noising' of the signal, leading to more precise and accurate results following analysis of the T2 distribution.