Electricity: measuring and testing – Particle precession resonance – Using well logging device
Reexamination Certificate
1998-10-29
2001-05-01
Patidar, Jay (Department: 2857)
Electricity: measuring and testing
Particle precession resonance
Using well logging device
C324S300000, C324S322000
Reexamination Certificate
active
06225803
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates generally to the field of well logging apparatus and methods, and in particular, to processing of nuclear magnetic resonance (NMR) signals to estimate physical properties of an oil or gas reservoir.
2. Description of the Related Art
Estimating physical parameters such as effective and total porosity, pore-size distribution, and the determining hydrocarbon types, are principal purposes for NMR log interpretation. The underlying rationale that NMR logging may provide such information is based on evidence that NMR relaxation times in porous media depend on texture (e.g., pore and grain-size distributions, in single wetting fluid phase saturated systems) and additionally on fluid types (oil/water/gas) in multiphase fluid-saturated porous media. Observed NMR log data (e.g., Carr, Purcell, Meiboom and Gill [CPMG] echo trains) represent the contributions from multiple fluid phases as well as fluids in different sized pores and thus, typically, exhibit a multiexponential behavior, with transverse relaxation times, T
2
components spanning from approximately one millisecond (ms) to over one second. The practical challenge for NMR log interpretation is to discriminate between contributions due to texture and fluid saturations and to quantify fluid saturations. Moreover, the NMR log signal typically is weak, and the instrumentation systems and the logging environment contribute significant noise that may be comparable to the signal. The resulting poor signal-to-noise ratio (S/N) gives rise to significant uncertainty in the estimated petrophysical parameters.
Prior art methods of NMR log interpretation generally use an inversion technique to estimate a relaxation distribution, i.e., a T
2
spectrum, from the acquired CPMG echo train data fit to a multiexponential decay relaxation distribution model. Different fluid phases may have different relaxation times, depending on the fluid molecular interaction, the rock surface properties, the reservoir environment, the fluid wetting characteristics of the formation and other physical properties known in the art. Distinctive features on the T
2
spectra, often reveal fluid saturations and pore structures—information on which petrophysical interpretation is based. For example, in the case of a water-wet reservoir with multiphase saturation, the non-wetting, and light, oil signal distributes into long relaxation bins. Water, on the other hand, in a water-wet reservoir, interacts strongly with pore surfaces, and thus has a short relaxation time. On a T
2
spectrum, water is identified from the short T
2
region, that is, initial bins. Gas, which is also non-wetting but diffuses faster than oil and water, may be identified in the intermediate region on a T
2
spectrum, since faster diffusion of gas reduces the apparent T
2
relaxation. From the estimated T
2
spectrum, partial porosities associated with different parts of the T
2
spectrum are identified for estimating the fluid saturations in a multiphase zone. In a single wetting fluid phase zone, for example, a water zone, with relative homogenous rock mineralogy, a T
2
spectrum approximately represents the porosity distribution in terms of pore sizes. Therefore, reliable interpretation depends heavily on accurate T
2
spectrum estimation.
It is well known that inverting echo train data to the T
2
domain distribution is an ill-conditioned problem, particularly when noise is present. Although regularization methods may help to stabilize the solutions, they also smooth the T
2
distribution estimate considerably, causing most of the distinguishing features of the T
2
distribution to be lost. The possible distortion of the resulting T
2
distribution estimate makes it difficult to separate the saturating fluid types. Furthermore, when a distribution involves short and long T
2
components, the standard procedure of using the method of minimization of least squares residuals in the inversion process often fails to weight all of the T
2
components equally. The short T
2
components are effectively represented by fewer echoes than the long T
2
components. When a T
2
distribution is dominated by a very short T
2
component and a second, long T
2
component, the technique can fail to fit the short component faithfully.
FIG. 1A
shows data from a synthesized noisy echo train fitted to a multiexponential model using a singular value decomposition (SVD) inversion algorithm, as is common in prior art methods. For an example of an application of SVD to NMR echo trains, see U.S. Pat. No. 5,517,115 issued to Prammer. The solid circles in
FIG. 1
are the samples of the noisy echo train at the echo interval of 1.2 milliseconds (ms). The noisy signal is generated from a multiexponential model NMR signal in accordance with Equation (1) below, and added zero-mean Gaussian noise, as in Equation (2) below. The standard deviation of the noise is 1.2.
The solid curves plot the underlying time-dependent noise-free multiexponential signal (thin line), and the fit to the noisy signal obtained using the SVD inversion method of the prior art (bold line). When the standard deviation of the random noise is high, the estimate noticeably misrepresents the actual spectrum. The short components of the input data, t 10 ms, suffer most, underestimating the effective porosity. This is also seen in the T
2
spectrum.
FIG. 1B
shows an underlying bimodal distribution (dual peak) (∘) and the estimated T
2
spectrum (×). The underlying distribution is bimodal with peaks near a T
2
of 3 ms and 150 ms. The multiexponential model includes seventeen terms, of which five have zero amplitude. The resulting fit to the spectrum from the SVD inversion has a single, broad, peak near T
2
equal to 100 msec. The T
2
spectrum below approximately
11
ms significantly underestimates the actual spectrum, and in the range of approximately 20-90 ms overestimates the actual spectrum.
Thus, there is a need in the art for improved methods of NMR signal processing for the recovery of T
2
spectra and thereby subterranean petrophysical characteristics in a oil or gas reservoir.
SUMMARY OF THE INVENTION
The previously described needs are addressed by the invention. Accordingly, a first form of the invention is a method of nuclear magnetic resonance (NMR) well log processing. The method includes the steps of forming a wavelet decomposition of NMR data signal, thereby obtaining a set of first coefficient values having a preselected first maximum scale and preselected first minimum scale; and windowing a preselected subset of the set of first coefficient values, thereby forming a windowed set of first coefficient values. A first reconstruction of the NMR signal is formed by generating an inverse wavelet transform of the windowed set of first coefficient values.
There is also provided, in a second form of the invention a computer software product for NMR well log processing including programming for forming a wavelet decomposition of NMR data, thereby obtaining a set of first coefficient values having a preselected first maximum scale and preselected first minimum scale, and programming for windowing a preselected subset of the set of first coefficient values, thereby forming a windowed set of first coefficient values. The computer software product also includes programming for generating an inverse wavelet transform of the windowed set of first coefficient values to form a first reconstruction of the NMR signal.
The method of the invention particularly addresses the need for resolving bimodal distributions involving short and long T
2
components, and for narrow monomodal distributions (often related to gas or light oil in a formation) that are broadened by noise and regularization. Improved bimodal distributions are useful for hydrocarbon typing involving either oil and gas, water and gas, or oil and water saturations. Sharpening monomodal distributions is useful in determining the T
2
value of the fluid phase thereby improving viscosity estimation.
The f
Baker Hughes Incorporated
Fagin Richard A.
Patidar Jay
Shrivastan Brij B.
Springs Darryl M.
LandOfFree
NMR log processing using wavelet filter and iterative inversion does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with NMR log processing using wavelet filter and iterative inversion, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and NMR log processing using wavelet filter and iterative inversion will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2480215