Professional Documents
Culture Documents
ABSTRACT
The introduction of formate-based drilling muds has successfully addressed drilling challenges related to barite weighted
muds. The formate-based muds, however, exhibited peculiar
petrophysical properties that adversely affected log interpretation. First, the mud present inside the borehole and surrounding the tool required different environmental corrections;
second, invading mud filtrate present inside the formation was
difficult to account for. Because of the higher density, lower
hydrogen index and high gamma ray readings associated with
sodium/potassium (Na/K) formate-based drilling fluids, the
petrophysical analysis typically resulted in inaccurate mineralogy and pessimistic porosity and permeability estimates. Such
estimates also became strongly dependent on the extent of invasion by the mud.
This article addresses these challenges to the use of Na/K
formate-based drilling fluids in gas-bearing siliciclastic and carbonate sequences via two new approaches. Logging-while-drilling
(LWD) time-lapse data acquisition makes it possible to track
changes in log measurements between a first (drill) pass and a
second (wipe) pass as mud filtrate invades the formation.
These changes reflect the contrast in petrophysical properties
between the mud filtrate and the displaced native formation
fluids. The first approach uses them to estimate the unknown
petrophysical properties of such mud filtrate. In the second
geometrical approach, the different measurements are viewed
as representing different axes in measurement space, and then
the axes are rotated to reduce the number of rotated measurements affected by invasion to just one. This measurement is
then discarded, and the remaining rotated measurements are
used.
In all, up to six different petrophysical models of two different wells were compared with this new modeling approach.
The results indicate a step change in petrophysical analysis in
the case of Na/K formate-based drilling fluids. They demonstrate how it is possible to build a robust but remarkably simple
petrophysical model using only rotated nuclear measurements,
with all of the following characteristics. The model is extremely
stable as compared to more complex models. It requires neither
knowledge of the Na/K formate mud filtrate characteristics
nor knowledge of its volume. It does not require resistivity
SPRING 2015
INTRODUCTION
Petrophysical analysis of well log data involves defining the
petrophysical properties of the various minerals and fluids
present underground, including the drilling fluids. Typically,
some of these petrophysical properties are measured, whereas
others are computed using established models or calibrated according to core, mud logging, formation testing and formation
fluid sampling data.
In some situations, the petrophysical properties of some of
the fluids involved remain difficult to estimate. This is typically
the case with hydrocarbon gases, the petrophysical properties
of which depend on detailed chemical composition, pressure
and temperature, and may change during reservoir production.
This is also the case with relatively new, or constantly improved, drilling fluid formulations such as synthetic oil-based
muds, or formate muds, which are typically recycled to offset
their high cost1-7. These recycling mud treatment and optimization operations naturally introduce uncertainties in the
mud properties.
Prior to the advent of logging-while-drilling (LWD) techniques, petrophysicists used wireline log data for petrophysical
analysis. Because wireline data is normally acquired a few days
after the well has been drilled, petrophysicists typically make
the assumption that mud filtrate invasion has already progressed past the depth-of-investigation (DOI) of the various
wireline measurements except for true resistivity (RT),
which is derived from an array of resistivity measurements
used to characterize and correct for the mud invasion. Petrophysical analysis then proceeds on the basis that the wireline
log data are representative of the invaded zone only.
The introduction of LWD techniques, which involve acquiring log data only a few minutes to a few hours behind the drill
bit, represented both a challenge and an opportunity. The challenge was that the mud filtrate depth-of-invasion (dinv) vs. the
DOI of the different measurements became an important consideration, Fig. 1. The opportunity was that petrophysicists
INTERPRETATION PRINCIPLES
PCA is the simplest of the true eigenvector-based multivariate
analyses. It is widely used in data processing and dimensionality reduction to extract a smaller set of linearly uncorrelated
variables (variates) called principal components from a
larger set of possibly correlated variables. It involves an orthogonal rotation to account for as much of the variability of
the data as possible. The first principal component is defined
as the one with the largest possible variance, and so on, with
the constraint that each following component is orthogonal to
all of the prior ones. In computational terms, the principal
components are found by calculating the eigenvectors and
eigenvalues of the data covariance-matrix or correlation-matrix, and ranking the eigenvectors by eigenvalue (highest to
lowest). This process is equivalent to finding the axis system in
which the covariance-matrix or correlation-matrix is diagonal.
PCA is closely related to factor analysis and is also related
to canonical correlation analysis; however, the following sections only cover PCA applications to LWD time-lapse data acquisition and analysis (TLA&A). The methods described here
have a lot in common with PCA, but we are primarily interested in the principal components
not maxi
with minimum,
mum, variance. The LWD data from the drill pass and the
wipe pass are first subtracted from each other, resulting in a
time-lapse dataset as follows:
Fig. 1. The mud filtrate invasion front progresses with time through the
measurement volume of investigation (enclosed in red), initiating changes in log
responses.
(1)
where Mdrill is a vector whose entries represent actual log measurements from the drill pass.
Mwipe is a vector whose entries represent actual log measurements from the wipe pass.
M is a vector whose entries represent the difference between
the drill and wipe pass measurements.
m1drill, m2drill and m3drill represent the individual measurements
from the drill pass, assuming three log measurements only.
m1wipe, m2wipe and m3wipe represent the same three individual
measurements from the wipe pass.
m1, m2 and m3 represent the difference between the individual measurements from the drill pass and the wipe pass.
Z is an index that represents measured depth (MD).
The time-lapse data set can be visualized as the data points
coordinates in a multidimensional data space, using one axis
per measurement, Fig. 2. Although the methods considered can
work with any number of log measurements, we limit ourselves here to three measurements only, m(1, m2 and m3, for
ease of visualization. Figure 3 shows the situation where only
one measurement is affected by mud filtrate invasion. In this
figure, data points close to the origin correspond to little mud
filtrate invasion,( and the further away the data points are from
the origin, the larger the mud filtrate invasion effect. In this case,
it is enough to exclude the affected measurement from petrophysical analysis and to conclude that the other measurements
SPRING 2015
Fig. 2. Considering that each log measurement represents a different coordinate,
this figure shows three axes representing these different log measurements. The
points Mdrill, Mwipe and M represent drill pass and wipe pass data points and
their differences, respectively.
Fig. 4. Situation where all three measurements, m1, m2 and m3, are affected by
mud filtrate invasion, displayed in red.
Fig. 3. Situation
where
only one
measurement,
m3, is affected
by mud invasion.
The axis corresponding to the affected measurement is displayed in red. The other
two axes, corresponding to the measurements m2 and m3 that are not affected by
mud invasion, are displayed in green.
(2)
where
is a vector whose entries represent the timelapse data set, mi, measurement values for all z = 1,2 , n. R
is a reference to the original non-rotated measurements refer(
ence frame.
Third, this symmetric covariance-matrix
is then expressed in
(
diagonal form in the
corresponding
rotated
eigenvector axis
(
system as:
(
(
(3)
SPRING 2015
Fig. 5. This figure shows how PCA can be applied to identify the principal
components PC1, PC2 and PC3 corresponding to the same time-lapse data set
shown in Fig. 4. After PCA, only one axis, PC1, is affected by mud filtrate
invasion (shown in red), and the other two axes, PC2 and PC3, are effectively
immune to invasion.
where L1, L2 and L3 are the corresponding eigenvalues in general and reflect the magnitude of mud invasion effects in our
case. L is the only non-vanishing eigenvalue (L1) in the case of
mud filtrates invading of gas-bearing formations. (R is a(reference to the rotated measurements reference frame.(
If we refer to the coordinates of the eigenvectors E1, E2 and
E3 in reference frame R as:
(2)
( (
(4)
(2)
( (2)
then the following relationship between the different covariance-matrices holds correct:
(
(
(5)
and
(
(
(6)
(
(
(
is a matrix in which the columns
where
are formed by the coordinates of the eigenvectors.
From a mathematical standpoint, it is strictly identical to
use either the measurements m 1, m2 and m3 for petrophysical
analysis or the corresponding
rotated
variates,
V1, V2 and V
3:
(7)
index for
individual minerals and fluids
where X represents an
present.
(
MX represents
the
endpoints
of
minerals
and fluids
individual
(
present.
(
Mfilt represents the endpoint of mud filtrate.
Mffld represents
the endpoint of the native formation fluid
(
(
(
flushed
by mud filtrate.
(
Volxdrill represents
the volumetric
percentages of minerals and
fluids present at the drill pass time.
(
( minerals and
the volumetric
percentages
of
Volxwipe
represents
(
fluids present at the wipe pass time.
filt
Volffld
represents the volumetric (percentage of mud filtrate that
( (
native formation fluid ffld between
has flushed
the drill pass
and the wipe pass.
(
Therefore,
the covariance matrix can also be re-expressed in
(
formations as:
gas-bearing
(13) (
(over the
sion at all. This represents
a step-change improvement
(
situation where all the measurements, m1, m2 and m3, used for
(14)
petrophysical analysis were simultaneously affected by mud fil
(15)
ate V1 can be used to get a better handle on one or the other of
(
the unknown petrophysical properties of the invading mud filwhere k in this expression represents an unknown factor, to be
trate and the flushed native formation fluid. This is possible
(
E is actually
approximated rather than determined accurately. The idea here
because the eigenvector
parallel
to the vector
1
(
( the mud filtrate petrophysical properties from
is to infer all
joining the endpoints* (m1)filt, (m2)filt and
(m3)filt of the invad
just one. For example, if the density of the mud filtrate is
ing mud filtrate and (m1)ffld, (m2)ffld and (m3)ffld of the flushed
fluid:
(
known,
native formation
( or if we use sodium/potassium (Na/K) mud density to
approximate
the density of the mud filtrate, then the k factor
(
( (
can be determined from Eqn. 15 and applied to the other
(8)
petrophysical properties.
(
The methods discussed in this section are relevant to any
Assuming measurements m1, m2 and m3 with linear mixing
Eqn. 8 simply stems from the fact
( that M can be
number
of measurements including nuclear magnetic resolawsonly,
(
(
re-expressed
as per the following
sums: (
nance (NMR) measurements and fluids.
M = Mwipe Mdrill
(9)
(
(
(
( (10)
(
(
(
(11)
(
(12)
(
(
*The word endpoint
and the associated expression endpoints of minerals and
fluids are used to describe what different measurements would read if only one
mineral or only one fluid was present.
SAUDI ARAMCO
( JOURNAL OF TECHNOLOGY
SPRING 2015
and
SPRING 2015
(
(
Fig. 6. Summary of the LWD drill pass and wipe pass log data sets for the
sandstone example. The wipe pass data are displayed in orange in all the tracks.
The tracks, from left to right, display gamma ray (green), deep propagation
resistivity (black), bulk density (red), thermal neutron porosity (blue), and
photoelectric factor (purple) logs, respectively.
Fig. 7. Comparative summary of the drill pass vs. wipe pass PCA variates
computed (V1, V2 and V3) for the sandstone example. V1 to V3 reflect the
maximum to the minimum variance present in the time-lapse data set, respectively,
as confirmed by the good overlay of the V2 and V3 curves from the drill pass and
the wipe pass. It is important to validate this overlay to qualify the V2 and V3
curves for invasion independent petrophysical analysis.
limestone, dolomite and anhydrite. The anhydrite is present uniformly throughout the section as a combination of nodules,
mimetic replacement and cement, which is indicative of evaporitic conditions. Deposition occurred in a shallow water environment on an open carbonate slope with a very low angle of
repose and with minimal siliciclastic input.
Three environments of deposition have been suggested. The
first zone occurs offshore in a region where the sea bottom lies
below the zone of wave action and mud is deposited at the relatively undisturbed bottom. The sedimentation rate is low as
temperatures are too low for carbonate mud to precipitate;
therefore, mud is sourced from outside the area. The second
zone is one in which waves interact with the bottom, but very
little mud is deposited as a result of the high water energy. The
material deposited is mainly skeletal debris formed by local
biota. Subsequently, sediments are mainly grainstones and
packstones. Currents cause winnowing and cross-bedding.
Reefs and oolites might also develop in this region. The region
closest to the land is marked by a sharply lower water energy.
Lime mud is deposited in this area and represents shoreward
transport of fines from the skeletal disintegration of grains as
well as the transport of the products of physicochemical precipitation (whitings). Comparable facies to these might form in
environments similar to those found in the Persian Gulf and
the Bahamas. The depositional and diagenetic history of this
reservoir is divided into multiple zones, based on gamma ray
markers, which are assumed to be time lines. Other potential
chronostratigraphic lines, such as paleosols, are apparently
absent. Consequently, these zones have been interpreted as reflecting eustatic changes in the sea level. Studies performed on
the Northern Basin platform and the Central Basin platform in
the Texas Permian Basin have demonstrated the use of extensive, thin, wind-blown silt layers as intraformational seals and
chronostratigraphic markers.
The pay zones lie around 12,000 ft TVD and tested
SAUDI ARAMCO JOURNAL OF TECHNOLOGY
SPRING 2015
(16)
(17)
(
SPRING 2015
Fig. 9. Summary of the LWD drill pass and wipe pass log data sets for the
carbonate example. The wipe pass data are displayed in orange in all tracks. The
tracks, from left to right, display gamma ray (green), deep propagation resistivity
(black), bulk density (red), thermal neutron porosity (blue) and photoelectric
factor (purple) logs, respectively. The intervals labeled Zone 1 to 5 represent the
porous reservoir intervals with mud filtrate invasion effects, which were selected
for PCA.
(18)(
Covariance-matrix
Eigenvalues
Eigenvectors
Zone 1
Zone 2
Zone 3
Zone 4
Zone 5
Zone 1 - 5
Table 1. PCA results in the (PHIN, PHID) reference frame. The results are shown per zone, then for all the zones combined (bottommost line)
As V1 is discarded, Fig. 12 shows conceptually how the retained variate, V2, is used for petrophysical analysis, together
with resistivity and photoelectric measurements. Figure 12 also
displays a third axis labeled VW, which represents the volume
of water computed using resistivity assuming Archies
cementation factor and saturation exponent m = n = 2. The
figure has been oriented such that the vertical (PC2, VW) plane
coincides with the plane of the page. In this plane, the
dolomite (DOL) mineral endpoint is located at the origin,
whereas the limestone (LIM) and anhydrite (ANH) endpoints
are located on opposite sides of the PC2 line. The gas (GAS)
and water (H2O) fluid endpoints are also shown. In the case of
the dolomite formation, actual log data points will fall within
the boundaries of the triangle shown in blue, and the position
of such data points relative to the DOL, GAS and H2O endpoints defines the volumetric fraction of dolomite, gas and
water inside the formation.
In practice, lithology is driven by the photoelectric factor,
and the Archies parameters are different from 2, so a nonlinear
solver is routinely used for petrophysical analysis. Also, due to
SAUDI ARAMCO JOURNAL OF TECHNOLOGY
SPRING 2015
Fig. 10. PHID vs. PHIN crossplot. The drill pass and wipe pass data points are
displayed in purple and in orange, respectively. The rotated (PC1, PC2) frame is
shown, together with an example showing the Mdrill and Mwipe coordinates in this
new frame.
Fig. 12. This conceptual schematic demonstrates how V2, together with resistivity
and photoelectric factor measurements, is used for petrophysical analysis. The
(PHID, PHIN) axes (cross plot) are complemented by a third Vw (water volume)
axis, as may be derived from the resistivity measurement. The various axes shown
here are not to scale, and the ANH, DOL, LIM, H2O and GAS endpoints were
included for illustration purposes only. For a DOL, H2O and GAS mixture, for
example, data points will fall within the indicated blue triangle, and vice versa, a
data point within the triangle is interpreted as a unique DOL, H2O and GAS
mixture. The photoelectric factor measurement role is to steer the matrix endpoint
for different mineralogy.
PCA Model A L D W G V R P
Fig. 11. PCA input and output curves. The input apparent porosity curves for
PHID and PHIN are shown in Tracks 1 and 2, respectively. The output variates,
V1 and V2, are shown in Tracks 3 and 4. The apparent porosity PHIA, i.e., V2
normalized to feel like an apparent porosity, is shown in Track 5. As can be seen,
V2 mixes PHID and PHIN in a way that cancels the mud invasion effects,
whereas V1 mixes them in a way that boosts the invasion effects and directly
reflects the mud filtrate volume.
STD Model A L D W G F / N D R P
where A L D W G M represents the model that solves for anhydrite, limestone, dolomite, water, gas or mud filtrate, respectively, and V N D R P represents the model that uses PCA
variates, neutron, density, resistivity or photoelectric logs,
respectively.
The STD model represents a conventional model that does
not use the PCA variates and so necessitates solving for mud
filtrate. Similar to the sandstone example, the PCA first principal component and Eqn. 11 were also used in this carbonate
example to estimate the mud filtrate petrophysical properties,
assuming a mud filtrate density of 84 pcf. Figure 13 shows the
petrophysical analysis results from both the PCA and the STD
models, for both the drill pass and the wipe pass. Again, these
results are consistent with each other, except for water volumes and thereby saturations when using resistivity
measurements from the wipe pass that are affected by deep
mud invasion. In particular, the total porosity and the mineralogy remained consistent, independent of the Na/K mud filtrates
displacing of gas between passes.
Fig. 13. Comparative volumetric interpretation results from different models and
passes for the carbonate example. Computed porosity and water volume from all
four passes are displayed in the rightmost track. All four porosity curves are in
excellent agreement, whereas the water volume curves from the wipe pass show
an increase over the drill pass, due to mud invasion effects on the resistivity
measurement.
CONCLUSIONS
Two examples were presented that describe PCA applications
to LWD TLA&A. The sandstone example used 3D PCA, and
the carbonate example used 2D PCA. In these examples, PCA
appears to be the mathematical tool of choice to automatically
reveal the structure and the dimensionality of the LWD time-lapse
data set. The identified principal components represent a rotated
coordinate system whose axes are oriented along the directions
of maximum and minimum change between the drill pass and
the wipe pass. The variates corresponding to the directions of
minimum change represent measurements that are effectively
immune to mud invasion. Therefore, they can be considered to
represent the formation prior to any invasion and can be used
for petrophysical analysis without including mud filtrate in the
corresponding formation model. This effectively eliminates the
requirement for the mud filtrate petrophysical properties to be
known in advance.
The sandstone example included results from six different
ACKNOWLEDGMENTS
The authors would like to thank the management of Saudi
Aramco for their support and permission to publish this article.
This article was presented at the Abu Dhabi International
Petroleum Exhibition and Conference, Abu Dhabi, UAE,
November 10-13, 2014.
SPRING 2015
REFERENCES
1. Phillips, J.: Report on the Prediction of the Impact of KFormate Fluid on Logging Responses, Technical Report,
Schlumberger DCS, Bergen, Norway, 2001 (unpublished).
2. Al-Harbi, A.A., Ersoz, H. and AbdRabAlreda, S.H.:
Influence of Sodium/Potassium Formate-Based Drilling
Fluids on Nuclear Logs, SPE paper 94693, presented at
the SPE Latin American and Caribbean Petroleum
Engineering Conference, Rio de Janeiro, Brazil, June 2023, 2005.
3. Galford, J., Truax, J. and Moake, G.L.: Borehole and
Formation Invasion Effects of Formate-based Mud
Systems, SPWLA paper, presented at the SPWLA 46th
Annual Logging Symposium, New Orleans, Louisiana,
June 26-29, 2005.
4. Pedersen, B.K., Pedersen, E.S., Morriss, S., Constable,
M.V., Vissapragada, B., Sibbit, A., et al.: Understanding
the Effects of Cesium/Potassium Formate Fluid on Well
Log Response A Case Study of the Kristin and
Kvitebjrn Fields, Offshore Norway, SPE paper 103067,
presented at the SPE Annual Technical Conference and
Exhibition, San Antonio, Texas, September 24-27, 2006.
5. Bert, P.C., Pedersen, E.S., Lauritsen, A., Behjat, N.,
Hagerup-Jenssen, S., Howard, S.K., Olsvik, G., et al.:
Drilling, Completion, and Open Hole Formation
Evaluation of High Angle Wells in High Density Cesium
Formate Brine: The Kvitebjrn Experience, 2004-2006,
SPE paper 105733, presented at the SPE/IADC Drilling
Conference, Amsterdam, The Netherlands, February 20-22,
2007.
6. Simpson, M.A., Al-Reda, S.A., Forman, D., Guzman, J.,
Al-Fawzy, M. and Vice, P.: Application and Recycling of
Sodium and Potassium Formate Brine Drilling Fluids for
Ghawar Field HPHT Gas Wells, OTC paper 19801,
presented at the Offshore Technology Conference,
Houston, Texas, May 4-7, 2009.
7. Kho, D., Grau, J., Khan, B., Ammar, H., Al-Anzi, E.H. and
Acharya, M.: Solving Formation Evaluation Challenges in
a Potassium Formate Mud Environment A Kuwait Case
Study of Using Neutron Induced Gamma Ray Spectroscopy
Data in a Complex Carbonate Reservoir, SPWLA paper,
presented at the SPWLA 51st Annual Logging Symposium,
Perth, Australia, June 19-23, 2010.
8. Kraishan, G.M., Al-Belowi, A.R., Ab Rahim, Z.B., Gzara,
K., Minh, C.C., Jain, V., et al.: Fluid Typing in Variable
Water Salinity Environments: Application of Geodesic
Transformations of LWD Time-Lapse Measurements,
SPWLA paper, presented at the SPWLA 55th Annual
Logging Symposium, Abu Dhabi, UAE, May 18-22, 2014.
SPRING 2015
BIOGRAPHIES
Kehinde Kenny M. Fakolujo joined
Saudi Aramco in 2009 as a Senior
Petrophysicist and is currently the Lead
Engineer for the satellite gas fields in
the Gas Development Petrophysics Unit
of the Reservoir Description and
Simulation Department. He has more
than 18 years of industry experience, serving in international assignments to the Netherlands and the U.S. prior
to joining Saudi Aramco. Kenny started his career with
Shell E&P Company, where he worked at various times as
a Petrophysicist in operations, assets and development. He
concentrated on integrated reservoir modeling, resistivity
inversion, electrofacies modeling, core log data integration,
uncertainty and value of information analysis, and thin bed
and shaly sand evaluation using industry approved
standard evaluation and interpretation software packages.
His current areas of interest and specialization include
development of a formation evaluation while drilling tool
for use in slim hole underbalanced wells and the impact of
a formate mud system on wireline log response in both
clastic and carbonate reservoirs.
Kenny is a member of the Society of Petroleum
Engineers (SPE) and is the Vice President of Membership
for the Saudi Arabia Chapter of the Society of
Petrophysicists and Well Log Analysts (SPWLA).
He received his B.S. degree (with honors) in Physics
with a specialization in Solid State Physics from Obafemi
Awolowo University, Ile-Ife, Nigeria.
Ali R. Al-Belowi is the Supervisor for
the Southern Area Petrophysics Unit in
the Reservoir Description and
Simulation Department at Saudi
Aramco. He is responsible for ensuring
that open hole and cased hole logging
programs are optimized, quality log
data are acquired, and logs are analyzed on time utilizing
fit-for-purpose petrophysics methods. Ali has done
extensive work on petrophysical analysis in both
exploration and development fields.
He joined Saudi Aramco in August 1989 as a Petroleum
Engineer after receiving his B.S. degree in Petroleum
Engineering from King Saud University, Riyadh, Saudi
Arabia.
Kais Gzara is a Schlumberger LoggingWhile-Drilling (LWD) Domain
Champion for the Kingdom of Saudi
Arabia GeoMarket region. He joined
the company in 1990 as a Wireline
Field Engineer in Italy and Libya. Kais
later worked for Schlumberger as a
District Petrophysicist in West Africa; as a Drilling
Planning and Interpretation Engineer at the Sugar Land
Product Center; as the Sales and Customer Support
Manager in Port Harcourt, Nigeria; as the Geology and
SPRING 2015