Documente Academic
Documente Profesional
Documente Cultură
ABSTRACT
INTRODUCTION
Mark of Schlumberger
W e 1 ( E1 ) x1 e 2 ( E2 ) x2 e 3 ( E3 ) x3 (2)
n +1
(E )x
n i =1 i i i
W e
dW
W
n +1
d
= (n i ( E i ) x i )
i =1
(4)
(3)
(1)
BS: W = a1e
a 2 a 3 Pe
+ a4
(5)
LS: W = c1e
b2 b3 Pe
c 2 c3 Pe
+ b4
(6)
+ c4
(7)
1.70
1.2
( )
Non-barite mudcakes
Barite mudcakes
3
3
1.25<m < 2.35 g/cm 1.48 <m < 1.98 g/cm
1.57<Pem< 1.91
10 <Pem< 100
0 <hmc< 1.5 in
0 <hmc< 0.5 in
)
(8)
1+ f + g
( )
Formations
3
< < 3.03g/cm
<Pe< 6.0
(Pe)
e 6 mc
and
e 7 mc
express
the
difference in attenuation between the mudcake
and the formation, with = m and
Pe = Pe m Pem .
MODEL PARAMETER DETERMINATION
INVERSION ALGORITHM
AT
(9)
where
(10)
C2 = ( pap ) T P 1 ( pap )
5
( p previous ) 2
j =1
p 2
COST FUNCTION
(13)
2
with Wi2 = WMi
+ WSi2
where:
State equation:
It is also assumed that we have a current (a priori)
estimate of the physical parameter vector, denoted
by pap.
pap = p + b
(12)
Wi
where it is assumed that
W12
0
R=
...
0
Wn2
i =1
(Wi Fi ( )) 2
(11)
C3 = ( p ) T Q 1 ( p )
5
( p )2
j =1
C3 =
Hi = F (i )
Ki = PHiT ( Hi PHiT + R) 1
CALIBRATION
MINIMIZATION
The minimization of the cost function is
performed using a quasi Gauss-Newton method
and consists of finding a root of the gradient
function. At a given depth n, the minimization
algorithm equations are the following:
0 = p
i +1 = pap + Ki (W F (i ) Hi ( pap i )) (15)
where
p$ + (n) =
(
)1
= (I K H )P
P+ = H t RH + P 1
(16)
A minimum of two points is required to determine
accurately the calibration gains and offsets which
have to be applied to the count rates. These points
are provided by measurements in aluminum and
magnesium blocks, with and without an iron
sleeve inserted between the tool and block.
Measurements with the sleeve are needed to
calibrate low energy window count rates.
VERTICAL RESOLUTION
With three detectors, two different sets of
estimates can be computed. One is a highresolution estimate which uses the backscatter and
short-spacing measurements, whereas the other
has a standard resolution and uses all three
detectors. The reconstruction errors obtained at
low resolution are used as an estimation of low
frequency varying offset errors on the count rates,
and are subtracted from high-resolution depthmatched measurements. This is described in detail
in Chapellat et al, (1996). Such an offset
recalibration procedure ensures that the highresolution estimates follow, on average, those at
low-resolution. However, the necessary trade-off
between robustness and resolution must be
appreciated as robust measurements in bad
borehole conditions will only be obtained by use
of all three detector measurements
+
1 n Wj Fj ( p )
=
j =1
W 2j
=1
(17)
C1
W j
(18)
Reconstruction errors
Reconstruction errors, defined as the relative
difference between reconstructed and measured
count rates are computed for each energy window.
Wmeas Wmodel
W =
Wmeas
EXAMPLES
All the experimental results discussed in this
paper have been recorded with the PLATFORM
EXPRESS density tool.
Figure 6 shows a comparison run with the LithoDensity* tool in an anhydrite-shale sequence.
inversion
CONCLUSION
Logging
speed
(ft/hr)
900
1800
3600
()
-2
3
(10 g/cm )
(Pe )
(Pe unit)
0.55
0.94
1.51
0.04
0.06
0.09
ACKNOWLEDGEMENTS
The authors wish to thank the oil companies
involved for permission to present examples of
their data.
REFERENCES
Belougne V., Faivre O., Jammes L. and Whittaker
S., Real time speed correction of logging data,
Ollivier
Faivre
graduated
from
Ecole
Polytechnique in 1974. He joined Schlumberger in
1976 and worked as a field engineer in Asia. In
1982, he entered the Reservoir Description group,
in Singapore. After various field positions in
interpretation, he joined Engineering in 1991 to
work on the interpretation of new tools. Since
1994, he has been developing real-time
interpretation answer products for PLATFORM
EXPRESS. He is presently Sensor Physics &
Interpretation metier manager in the Schlumberger
Product Development Center in Clamart, France.
APPENDIX 1
MAXIMUM LIKELIHOOD ESTIMATION
Theory,
W~
pap ~
( F ( p ), R )
( p, P )
L ( ) =
e
e
1
(2 ) m R 1
1
( 2 ) n P 1
1
(W F ( ) ) T R 1 (W F ( ))
2
1
( pap ) T P 1 ( pap )
2
Or equivalently
problem:
p$ +
p$ + = arg min C( )
where
1
C( ) = [(W F ( )) T R 1 (W F ( ))
2
+ ( pap ) P 1 ( pap )]
LS
Formation
SS
Detector
x 3 , 3 (E3 )
BS
Source
x 2 , 2 (E 2 )
Formation
density
Mudcake
density
Mud density
hmc
(Mudcake thickness or standoff)
Source
x1 , 1(E1 )
De ns ity
model-theoretical
(g/cc)
model-theoritical
(g/cc)
0.06
0.04
Series1
0.02
3 sensors
Series2
BS
Series3
SS
-0.02
Series4
LS
-0.04
-0.06
1.6
1.8
2.2
2.4
2.6
th e o ri ti ca l de n s i ty (g /cc)
2.8
Figure 3a: Density estimation errors on the whole database measurements without mudcake. The
accuracy of the density estimated by each sensor separately depends on its sensitivity. The accuracy of
density computed with 3 sensors is less than or equal to 0.01g/cc.
Pe
0.5
model-theoretical
0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
1
Figure 3b. Pe estimation errors on the whole database measurements without mudcake. Accuracy of Pe
computed with 3-sensors is less than or equal to 0.1.
Mg/mc1
Mg/mc2
2.7
B-lime/mc1
B-lime/mc2
2.5
Dol/mc1
Dol/mc2
2.3
2.1
1.9
1.7
1.5
0
0.2
0.4
0.6
0.8
Figure 4. 3 detector inversion results for three formations as a function of the mudcake thickness. The
table below lists formation and mudcake properties:
Formation
Density
3
(g/cm )
Mg=Magnesium
1.75
B_lime= Bedford lime 2.44
Dol=Dolimite
3.03
Pe
Mudcake
2.55
4.85
3.14
mc1- No barite
mc2- barite
Density
3
(g/cm )
1.51
1.48
Pe
1.67
48
Raw data
(count rates)
Depth-matching
Res.-matching
Dead time
Borehole
correction
Calibration
Background
Hole diameter
Mud density
Measurement
inversion
Physical properties
(Density, Pe, Standoff)
10
11
Figure 7. Log simulation at three different logging speeds together with statistical uncertainties (one
standard deviation error).
12
Figure 9a. Density and Pe estimations obtained with the correct calibration.
13
14
15