Sunteți pe pagina 1din 16

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

Error Properties of Magnetic


Directional Surveying Data
Erik Nyrnes, NTNUNorwegian University of Science and Technology, Torgeir Torkildsen,
Statoil ASA Norway, Hossein Nahavandchi, NTNU
Copyright 2005, held jointly by the Society of Petrophysicists and Well Log
Analysts (SPWLA) and the submitting authors.

in the multi-station estimation. This is a significant


potential source of error as such outliers cannot be
detected when applying conventional methods. It is
demonstrated how much these extended quality control
procedures can improve the final wellbore positions.

This paper was prepared for presentation at the SPWLA 46th Annual Logging
Symposium held in New Orleans, Louisiana, United States, June 26-29, 2005.

ABSTRACT

Key words: Wellbore positioning, Directional


surveying, Quality control, Multi-station estimation
techniques, Systematic errors, Gross errors, Random
errors.

The primary tools for directional surveying while


drilling (MWD) are based on the determination of the
magnetic azimuth. This paper presents the results from
a study that highlights several aspects of the quality of
directional surveying with magnetic tools. Survey data
from 66 North Sea well sections have been investigated
in order to evaluate the overall quality of the sensor
readings and the applicability of different estimation
and error detection techniques. The surveys cover
different wellbore geometries and are performed by
several service companies.

INTRODUCTION
Measurements collected from a number of survey
stations can through a least squares adjustment be used
to detect and account for different error terms on all
measurements, see [Brooks et al. 1998] for details.
Such techniques are in the following denoted MultiStation Estimation (MSE). In this paper the GaussNewton method is used to estimate the parameters, see
the Appendix B. The different error terms are
commonly divided into systematic, random and gross
errors. Systematic and gross errors are in this study
treated as unknown fixed parameters, the only
difference being the assumptions about whether they
affect a number of measurements (systematic) or single
measurements (gross). They will be estimated together
with the directional parameters for each survey station;
the magnetic azimuth Am, inclination I and high side
toolface . The estimations are based on the linear
regression model:

The Industry Steering Committee on Wellbore Survey


Accuracy (ISCWSA) has developed error models for
magnetic directional surveys, which now have become
an industry standard. These error models are derived for
two standard single-station processing techniques. The
error term values are settled with the basis in input from
several service companies. The results from the analysis
of the survey data presented in this paper confirm the
systematic error term values in general. However, some
of the error term accuracies are normally better than the
modelled ones.
The applicability of multi-station estimation techniques
is demonstrated. There is always a chance of
misinterpretation due to several reasons. These are:
poor geometry, low redundancy, high random noise
level, presence of gross errors, and errors in the Earth's
gravity and magnetic field references. Since the
estimated results are very dependent on the selected
parameter model, the major issues related to the
selection procedure are discussed. Procedures are also
derived for performing multi-station estimation of
parameters in a reliable and robust manner.


E (y ) = y e = [X Z ] ; Cov(y ) = ee = 2 Q ee

(1)

where y is a vector of n observations, is a vector of u


unknown fixed parameters Am, I and , X is a nu
known coefficient matrix, Z is a nr known coefficient
matrix corresponding to the r additional model
parameters , Qee is a known observation error cofactor
matrix and 2 is the variance of unit weight, usually
unknown. The inverse of Qee is the weight
1
matrix, P = Q ee
.

Finally, there is a practical demonstration of the


application of a new error detection procedure. This
method is used to detect outliers that might cause errors

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

The additional parameters may represent the


systematic error terms; linear scale factor , bias and
sensor misalignment , or gross errors (biases) in single
sensor readings (denoted i). may also represent the
Earth's reference components; gravity G, field strength
B and dip angle (defined in Appendix A). The errors
e are in this context assumed to be normally distributed,
e N(0, ee) and uncorrelated (ee diagonal). This
quantity will often be referred to as noise. The scale
factor , bias and the errors e may also be attributed
to environmental effects, and not necessarily sensor
imperfections only.

METHODOLOGY
Test of hypothesis. The general test to decide whether
the subset of systematic error terms is zero is based
on the following statistical hypothesis:

H 0 : E (y ) = X vs. H A : E( y ) = [X Z ] , 0 (2)

where H0 is the null hypothesis ( is zero) and HA is


the alternative hypothesis ( is different from zero). H0
is rejected if the 100(1) % confidence region for
does not contain 0, i.e. if:

The aim of this study is to evaluate the overall quality


of MWD magnetic directional surveys with respect to
all the three error categories described above. 66
survey-sections have been analysed in detail. The
surveys are performed by several survey companies in
the period between 1998 and 2004 and cover the
northern parts of the North Sea and the southern parts
of the Norwegian Ocean. They represent various
geometries and borehole dimensions with average
lengths of roughly 30 stations. The spread of the high
side toolface is considered good for all wellsections.

[ 0]Q [ 0]
T=

r r
2

where Q

rr

> Fr , n q r ()

(3)

is the joint cofactor matrix of the esti-

mates in the subset, 2 is the 2 estimate under HA and


Fr,nqr() is the upper (100)th percentile of the
Fisher distribution with r and nqr degrees of
freedom. In this study the significance level = 0.05 is
used for the test of the null hypothesis H0.

First, the properties of the systematic error terms are


investigated. The estimated error terms will be
compared to the error term values of the ISCWSA
MWD-error model, which consider the lumped effects
of the bias and linear scales, and two types of sensor
misalignments, see [Williamson 2000] for details.
Therefore, only the bias and scale error terms are
considered in this study. The corresponding error values
are listed in Table 8. (There is one exception, see the
Example 2, page 4). Some major issues related to the
selection of model parameters and misinterpretation of
the estimation results are discussed, in order to provide
a better understanding of multi-station estimation
techniques.

Selection of specific model parameters. Since the goal


is to distinguish specific model parameters of
importance, the simultaneous confidence property in
Equation (2) is ignored. Instead, important parameters
will be detected according to a stepwise regression
approach similar to the so-called Backward Elimination
explained in [Miller 2002]. In a complete model one
starts by deleting the parameter corresponding to the
smallest residual sum of squares after deletion (which is
the one with the smallest Student's t-test statistic). Then
a new adjustment is performed and the process is
repeated until only significant model parameters are
left. It should be considered that even the resulting
subset fits the measurements y fairly well, it may just
be one among several model combinations that have
similar effects on the residual sum of squares .

Second, the frequency of gross errors in the single


sensor readings is investigated. As undetected gross
errors may affect the results considerably, the process
of removing them is important and will therefore be
discussed in detail. See also [Nyrnes et al. 2005].

Derivation of measurement-weights. The Earths


magnetic field is commonly defined by the field
strength B, magnetic dip angle and the declination ,
see Appendix A for details. The irregular variations in
the strength and direction of the Earths magnetic field
vector will give rise to irregular variations in the
measured magnetic field strength along all three
instrument axes. The assumptions of similar
uncertainties for the errors e in all three directions are
therefore considered reasonable. See also [Torkildsen et
al. 1997]. The errors in the accelerometer
measurements are also assumed to have the same

Finally, the noise levels in the measurements are


evaluated. The ability to detect systematic and gross
errors in the measurements is dependent on this
quantity. For the magnetometer readings the estimated
noise levels are compared with the expected time and
location dependent standard deviations of the magnetic
field intensity variation to see if there are any
agreements.

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

estimated. Suppose also that the true field strength B is


constant for the whole section. If the true axial scale
actually is positive and of significant order of
magnitude, it will cause the cross-axial scale estimates
to decrease equally and cause the field strength estimate
to increase. Other estimates are left unaffected. The a
priori field intensity B may then be suspected to be of
poor quality, since it is significantly larger than the
estimated value for the entire well, and the drilling fluid
may be suspected to contain magnetically susceptible
materials since the transverse scale estimates are
negative (indicating attenuation of the field intensity)
and of similar order of magnitude.

uncertainty for all three axes. The coordinate systems


are defined in Appendix A.
The following approach is used to find an approximate
weight relation between the accelerometer and
magnetometer readings. The variance of unit weight 2
is first estimated from the accelerometer measurements
only. Then the magnetometers are introduced with
corresponding weights such that the original variance of
unit weight estimate 2 is left unchanged.
The Earth's magnetic field components are introduced
as measurements with a priori uncertainties depending
on whether standard referencing (IGRF or BGGM) is
applied solely or together with corrections for the local
crustal field (enhanced referencing). See Table 1 for
details.
Reference quality

Let us consider another example. Suppose that all three


scales are to be estimated and that the a priori field
strength B is introduced as a measured quantity. In
addition, suppose that the magnetic field intensity
observed by the sensors actually is damped. If B is
introduced with a lower value than the true value,
which by coincidence is sufficiently low to force the
cross-axial scale estimates to zero (or close enough to
zero to be declared insignificant) this may lead to the
conclusion that a positive axial scale error is the
probable reason for the model misspecification.

Standard Referencing 130 nT 0.20 deg


Enhanced Referencing 65 nT 0.12 deg
Table 1: Standard deviations for the standard referencing
according to [Williamson 2000] and Statoil values for the
enhanced referencing.

For the accelerometers the true error terms must be


even larger to be detected than for the magnetometers,
although the accelerometer measurements in general are
less noisy. The main reason is that the redundancy is
lower than for the magnetometers, maximum one on
average at each station. The introduction of
magnetometer readings will be helpful, but not always
of considerable importance. In the same way as for the
magnetometers, the estimation of the accelerometer
scale errors is sensitive to errors in the local gravity.

The prediction of the local gravity is based on the


international gravity formula and corrections for local
anomalies. See Appendix D for details. The gravity
values provided by the survey companies are used if the
measurements were given in other units than ms-2 (e.g.
counts).
The measurements of the Earths reference components
are introduced as average values for the entire section.

For straight wellbores, in addition to the cross-axial


biases it is only possible to estimate two of the four
remaining accelerometer error terms. This gives high
odds that undetected effects may remain and the chance
for misinterpretation is therefore high.

Interpretation of the results. The interpretation of the


estimation results is not straightforward because we
often have to choose between several parameter models
which have similar effects on the measurements y. As
an example, consider the subset consisting of the x, y,
and z-magnetometer scale errors and the Earth's
magnetic field strength B. The columns in the design
matrix Z representing these parameters are linear
dependent irrespective of the geometry and hence they
cannot be estimated simultaneously. From standard
theory this singular problem is often referred to as exact
collinearity. An error in the a priori Earth's magnetic
field strength B will induce equal errors in all three
magnetometer scale estimates. Accurate a priori
information about B is therefore needed in order to
estimate all scales properly.

As a general rule, the estimation of the four transverse


accelerometer and magnetometer bias error terms (gx,
gy, bx, by) can always be considered highly reliable.
Among all the additional parameters, the cross-axial
bias estimates are the ones which are least sensitive to
errors in the sensor readings, reference components and
unmodelled effects.

EXAMPLES
Four datasets with special error characteristics will now
be analysed and discussed in detail.

Let us suppose that B is estimated (i.e. no a priori


information is used) and that the axial scale error is not

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

degrees in parts with constant inclination and azimuth.


Model 3, which is a realistic one, produces considerably
more unstable azimuth estimates than Model 2 in
directions close to horizontal east-west. This is due to
the introduction of the axial scale error in the model.

Example 1. As will be demonstrated for the horizontal


wellbore shown in Figure 1, the results of the MSE are
very dependent on the selected parameter model. Note
that the azimuth varies a lot in the first half of the
wellbore, and remains almost constant in the second
half. It can be seen that the parameter models defined in
Table 2 lead to considerably different azimuth
estimates, especially in attitudes close to the horizontaleast direction. Model 1 is considered to be the most
trusted solution and is therefore used as reference.
Model

All biases

est

est

est

est

est

est

est

est

est

est

est

est

est

x & y-scales fixed


Axial scale

fixed fixed est fixed fixed fixed est

meas meas meas est meas est fixed

meas meas meas meas fixed fixed fixed

Example 2. This section turns smoothly from the east to


north direction and the inclination is roughly 40 degrees
for the whole section, see Figure 2.
Large scale errors were detected on the transverse axes,
probably caused by magnetic particles in the drilling
fluid. It can be seen that the gyro survey verifies the
correctness of the MSE results. The left hand plot in
Figure 2 shows that the azimuth differences between
the SSE and MSE are largest for the east direction, and
almost zero for the south direction. This is in
accordance with the propagation of transverse scale
errors of equal magnitude, since the joint effect is zero
for the south direction.

Table 2: The parameter models applied for the Example 1. Est


= estimated, meas = measured, fixed = not estimated
(reference fixed to initial value and systematic errors fixed to
zero )

The right hand plot in Figure 2 compares the predicted


magnetic field strength with the field strength
calculated from the measurements. The differences are
largest for the south direction and increase drastically as
the inclination becomes greater in the final parts of the
well. Even if the scale errors are very large, they cannot
be detected using conventional methods (i.e. comparing
the measured field strength with values predicted from
independent sources) in attitudes where the effect on
the calculated field intensity is small. The worst case is
when drilling closely parallel to the Earth's magnetic
field vector.

Azimuth differences. Multi-station estimation.


Reference: Model 1

2.0

Wellbore geometry

Parameter model

100

1
2
3
4
5
6
7

90
Degrees

1.5

70
60
50

Inc
Az

40

1.0

30
0

10

20

30

40

Station no.
5

0.5

52000

0.0
10

20

30

40

Station no.
(Interval: 30 m.)

3
2
1
0
-1
-2

51800
51600
51400
51200

10

20

30

40

B (nominal value)
B (calculated)

51000
50800

10

Station no.

Figure 1: Example 1. Magnetic azimuth estimates for the


parameter models defined in Table 2. Some of the models
produce azimuth estimates differing almost two degrees in the
east direction.

Wellbore Geometry
MSE (reference)
SSE
Gyro

Degrees

Magnetic field strength (nT.)

Azimuth differences (deg.)

Azimuth differences (deg.)

80

210
180
150
120
90
60
30
0

30

40

Station spacing: circa 30 m.


Inclination
Azimuth
0

It can be seen that Models 6 and 7 provide the largest


azimuth differences. The estimates are very sensitive to
errors in the reference components and sensor readings
in directions near the horizontal east-west. They
produce similar results as noted earlier. Further down
the well the azimuth differences form two main trends
depending on whether the magnetic dip angle is
measured or fixed to the initial value (i.e. treated as an
error-free quantity). The difference is approximately 0.5

20

Station no.

10 20 30 40 50

Station no.

Figure 2: Example 2. Left hand plot: The simultaneous effects


of the transverse scale errors on the azimuth estimates are
almost zero for the ordinary SSE (Single-Station Estimation
without axial correction) technique in the east direction. Right
hand plot: Comparison between the calculated and predicted
magnetic field intensity.

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

differences are larger than 5 degrees compared with the


MSE and SSE (Single-station estimation without axial
correction), see Figure 4. The application of the axial
magnetic correction technique leads to a 30 metres
displacement in position for this section, compared with
SSE without axial correction.

Example 3. MSE was first applied with a model


consisting of biases and scale error terms. A large axial
magnetometer bias was detected. After a comparison
with a gyro-survey, the azimuth deviations were still up
to four degrees, see Figure 3. After a more detailed
analysis, the most likely reason for the azimuth
differences was found to be an axial misalignment
between the accelerometer and magnetometer sensor
packages in the survey tool. The estimated
misalignment was 1.2 degrees and strongly significant.
The improved model is used as the reference in Figure
3. It can be seen that the azimuth differences are largest
in the most inclined parts of the wellbore, and
somewhat smaller in the least inclined parts. This is in
accordance with the propagation of a toolface error
[Ekseth 1998]:

Wellbore Geometry

Azimuth (deg.)

Degrees

90

80

70

60

40

Azimuth (standard)
Azimuth (MSE)
Azimuth (SSE ax. mag. corr.)
2

10

12

14

Survey station no.


(Interval: circa 30 m.)

(4)

Figure 4: Example 4. The application of the axial magnetic


correction technique leads to erroneous azimuth estimates
near the east-horizontal directions.

SYSTEMATIC ACCELEROMETER ERRORS

Cross-axial bias errors. Totally 41 x-axial and 44 yaxial biases were declared significantly different from
zero by testing the hypothesis defined in the Equations
(2) and (3). The corresponding cross-axial bias
estimates are shown in Figure 5. The largest estimates
are about 1 Gal in magnitude.

5
4
3
2
1
0

Significant x-axial
accelerometer biases (Gal)

-1
20
Well bore geometry
210
180
150
120
90
60
30
0

40
Station no.

Inclination
Azimuth

60

80

Station spacing: circa 30 m.

Optimal model (Reference)


Misalignment parameters
not modelled
Gyroscopic azimuth

0 10 20 30 40 50 60 70 80 90
Station no.

Figure 3: Example 3. Comparison between the gyroscopic


azimuths and the MSE-azimuths with and without sensor
misalignment parameters in the model. The azimuths of the
improved model (including misalignment parameters) agree
most with the gyro azimuths.

Significant y-axial
accelerometer biases (Gal)

20
18
16
14
12
10
8
6
4
2
0
-1.5

Frequency

Frequency

Azimuth differences (deg.)

10 20 30
Station no.

50

At an inclination of 80 degrees, Equation (4) shows that


a toolface error of 1 degree induces a negative error in
the azimuth estimate of about 3.5 degrees. The MSE
(with misalignment parameters) and SSE with axial
magnetic correction lead to a wellbore position
difference of approximately 170 metres.

Degrees

Inclination
Azimuth
0

sin sin I cos A - cos cos I


dA =
d
cos

120
100
80
60
40
20
0

-1.0

-0.5

0.0

0.5

1.0

1.5

20
18
16
14
12
10
8
6
4
2
0
-1.2 -0.8 -0.4

Estimates (Gal)

0.0

0.4

0.8

1.2

1.6

Estimates (Gal)

Significant cross-axial
accelerometer biases (Gal)
30
25

Frequency

Example 4. The first part of this horizontal wellbore lies


near the east-west direction. The results produced by
three different estimation techniques (MSE, SSE with
and without axial correction) are compared in Figure 4.
Because the axial magnetometer bias is very small and
insignificant, the MSE and SSE without magnetic
correction produce similar azimuth estimates.

20
15
10
5
0
-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

Estimates (Gal)

Figure 5: Estimated values of significant accelerometer bias


error terms.

The SSE with axial correction produces very unstable


azimuth estimates in the horizontal east direction. The

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

masks the effect of the gravity error. Like the gravity


error, an undetected axial scale (i.e. one that is not
estimated) also affects the cross-axial scales equally.

Nearly 50 % (19 out of 41) of the x-axial biases have


their values outside the interval 0.4 Gal, which are
almost 30 % out of the 66 examined sections. As a
comparison, the corresponding value of the MWD-error
model is approximately 0.39 Gal (1 standard deviation).
For the y-axis bias estimates, about 40 % (18 out of 44)
of the absolute values exceed this level, equivalent to
about 30 % out of all the examined survey sections.

The ability to detect the cross-axial scale errors is


reasonable good, provided that the gravity reference is
of good quality. Usually, errors even smaller than the
corresponding MWD-error model value (0.05 %) can
be detected with high probability.

The x and y-axial bias estimates show some different


properties. While the x-axial bias estimates are
symmetrically distributed around zero, we see that a
large number of the y-axial biases are negative.

Sometimes the a priori gravity reference was suspected


to be of poor quality, since the cross-axial estimates
became similar and strongly significant (also in the
presence of the axial scale in the model.) However, it
was considered reasonable to estimate the cross-axial
scales for several bit-runs. RMS values for the
estimates of significant error terms turned out to be
almost 0.1 % for both axes (see Table 8). However,
when considering the potential disagreement between
the predicted and true gravity, there is no reason to
claim that the true errors are larger than specified by the
MWD-error model. Since the differences between the x
and y-axial biases for each pair of estimates also tended
to be small (not shown), this assertion is further
confirmed.

Although the parameters according to the tests are


classified as gross errors, most of them are hardly large
enough to have any practical influence on the final
wellbore positions compared with other sources of
systematic errors. Due to the normality assumption of
the measurement errors and the MWD-error model
terms, maximum one third of the estimated error values
should lie outside the interval 0.39 Gal to be in
accordance with the MWD-error model. We see that
this criterion is fulfilled for both axes. The surveys are
considered to be of acceptable quality with respect to
the cross-axial bias error terms.

SYSTEMATIC MAGNETOMETER ERRORS

For about 50 % of the wellsections (34 out of 66), none


of the cross-axial bias estimates turned out to be larger
than 0.39 Gal in absolute value, regardless of whether
the parameters differed significantly from zero or not.

Cross-axial bias errors. The two upper histograms in


Figure 6 show that the cross-axial bias estimates are
similarly distributed and almost of the same order of
magnitude. All estimates of significant error terms have
their values within the interval 200 nT except of a few
x-axial biases of almost 250 nT in magnitude (see the
histogram to the left). The majority of estimates are
smaller than 100 nT in absolute value for both axes.

The ability to detect the cross-axial biases is usually


very good, even if they are considerably smaller than
the corresponding MWD-error model values.

Axial bias errors. The variance of the estimated axial


bias usually becomes high, especially in the presence of
the axial scale in the model. The effect of an undetected
axial scale propagates mainly into the axial bias
estimate. The axial bias estimate is sensitive to errors in
the gravity reference G. Due to these uncertain factors,
the axial bias estimate can usually not be trusted.

The cross-axial estimates are represented together by


the histogram in the middle of Figure 6. We see that the
largest number of estimates, approximately 70 % out of
the 58 significant, have their values inside the interval
of tolerance: 70 nT. This means that only about 15 %
out of the 132 (266) potential estimates exceed the
limits 70 nT.

Scale factor errors. A proper estimation of these


parameters requires precise knowledge about the local
gravity. However, the estimation involves several other
uncertain factors, as will be explained below.

The results indicate that the true errors generally are


lower than that modelled by the MWD-error model.
Estimates exceeding the tolerance limits only
corresponded to much less than one third of the
surveys.

The variances of the cross-axial scale estimates usually


become similar and much lower than for the axial one.
If the estimates themselves also turn out to be similar,
the cause may be a gross error in the a priori gravity,
and not necessarily sensor imperfections. Because the
variance of the axial scale estimate usually becomes
high, the effect of the sensor reading errors usually

For about 80 % of the examined sections, none of the


cross-axial bias estimates turned out to exceed the
limits 70 nT simultaneously for the same section,
regardless of whether they differed significantly from
zero or not.

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

Even if the cross-axial bias errors are far below 70 nT,


they can still be detected with high probability. The
overall ability to detect these error terms is considered
good.
Estimated y-axial magnetometer
biases (significant).

10
9
8
7
6
5
4
3
2
1
0

Frequency

Frequency

Estimated x-axial magnetometer


biases (significant).

-200

-100

applied if the conditions permit, i.e. if the deviation


from horizontal east-west is acceptable. By doing so it
does not matter theoretically how large the true bias is.
However, an axial error will have the largest influence
on the azimuth estimation in such positions if standard
SSE is applied.
About 35 % of the sections were surveyed without
correcting for axial magnetic interference. Significant
biases were detected for the majority of these surveys.
Most of them were some hundreds of nanoteslas larger
than the tolerance (165 nT) and some almost ten times
larger. Some basic statistical parameters are given in
Table 3.

10
9
8
7
6
5
4
3
2
1
0
-200

100

-100

100

200

Estimates (nT)

Estimates (nT)

The results show the potential magnitude of the axial


bias errors, which mainly reflect the magnetic
interference from drillstring/BHA.

Estimated cross-axial magnetometer


biases (significant).
14
12

Frequency

10

Mean Std.dev.

8
6

Max.

Min.

470 nT 370 nT 1548 nT 202 nT

4
2

Table 3: Basic statistics of the absolute values of the axial


bias estimates for surveys which had not been corrected for
axial magnetic interference.

0
-200

-100

100

Estimates (nT)

Estimated z-axial magnetometer biases,


larger than 1000 nT and smaller
than -1000 nT (significant).

10

10

Frequency

Frequency

Estimated z-axial magnetometer biases,


less than 1000 nT and larger
than -1000 nT (significant).

6
4
2

Cross-axial scale errors. If either the axial scale or the


Earth's magnetic field strength B deviate significantly
from its initial values, the values of the cross-axial scale
estimates are dependent on whether we choose to
estimate the axial scale or B. This is shown in Figure 7
and will be discussed more in detail in the next section.

6
4
2

0
-750 -500 -250

250 500 750

0
-4000 -2000

Estimates (nT)

We see from the upper plots in Figure 7 that some of


the estimates are more than 2 % in magnitude. One is
almost 4 % (geomagnetic references measured). The
smallest error is approximately 0.25 %, which is almost
0.1 % larger than modelled (0.16 %). In general, errors
of such small magnitudes can hardly ever be detected
since the magnetometer measurements usually are too
noisy. This is mainly due to the high level of irregular
variations in the Earth's magnetic field strength. As a
result, the transverse scale errors turned out to be
insignificant for about 60 % of the examined surveys
sections.

2000 4000 6000 8000 10000

Estimates (nT)

Figure 6: Estimated magnetometer bias error terms.

Axial bias errors. We now consider the axial bias


estimates. Estimates of smaller magnitude (within
1000 nT) are shown on the lower left histogram in
Figure 6, while the larger ones (outside 1000 nT) are
shown on the lower right. The corresponding ISCWSA
tolerance is set equal to the lumped uncertainties of the
systematic sensor reading error (70 nT) and the axial
magnetic interference (150 nT); derived from the
AMID parameter in [Williamson 2000] for the North
Sea conditions:
(70 nT ) 2 + (150 nT) 2 165 nT

The cross-axial scale estimates are plotted against each


other in Figure 7. We see that all estimates (except from
the one in the right plot) are negative and of similar
order of magnitude. It is well known that the
magnetically susceptible properties of drilling fluid
(due to steel particles from wear, weight material etc.)
may lead to a cross-axial damping of the Earth's
magnetic field intensity surrounding the sensors and
give rise to cross-axial scale errors of similar magnitude
[Wilson, Brooks 2001], [Torkildsen et al. 2004].

(5)

If the axial magnetic interference is unacceptably high,


an axial magnetic correction technique should be

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

Figure 7) for each pair of estimates therefore do not


indicate any overall disagreement with the MWD-error
model. (This was also the case for insignificant scale
error terms.)

Axial scale errors not estimated.

Axial scale errors estimated,


if significant.

Estimated x-axial scale


factor errors (significant).

Estimated x-axial scale


factor errors (significant).

12

12

10

10

Frequency

Frequency

In this study the errors are defined in such a way that


they will become negative in the case of damping. The
results of the analyses presented in this paper show that
the damping is larger when ilmenite is used compared
with barite. The attenuation also tends to increase for
greater dimensions and is often larger for the 12.25 inch
sections than for the 17.5 inch sections. This was also
pointed out in [Torkildsen et al. 2004].

6
4

6
4

Since the estimated scales also mirror errors in the


geomagnetic reference, it is not possible to conclude
whether the model-misspecifications are due to sensor
imperfections or errors in the geomagnetic components.
The mismatch may also be due other underlying effects
which cannot be modelled properly by the bias and
scale factor errors only.

0
-4

-3

-2

-1

-4

-3

1
0
-1
-2
-3
-4
-5
-4

-3

-2

-1

Estimated x-axial scale errors (%)

Estimated y-axial scale errors (%)

Estimated cross-axial magnetometer


scale factor errors (significant).

-5

-2

-1

Estimates (%)

Estimates (%)

Estimated y-axial scale errors (%)

Axial scale errors. The axial scale was declared


significantly different from zero for about 15 survey
sections when the reference components were treated as
measurements. These estimates were mostly positive
and could be several percent in magnitude. Note that in
the presence of the axial bias (which is usually more
significant and therefore chosen in the expense of the
axial scale) the true axial scale must be extraordinary
large to be detected. This condition is strongly
dependent on the inclination and azimuth variation.

Estimated cross-axial magnetometer


scale factor errors (significant).
1
0

Errors in the Earth's geomagnetic components. For the


above mentioned surveys (where the axial scale became
significant), the axial scale was excluded and the
magnetic field strength B and dip angle were
estimated. The result of this was that the magnetic field
strength B differed significantly from zero for all these
survey sections (often strongly significant), while the
dip angle was significant only for a few. Some basic
statistical parameters are given in Table 4.

-1
-2
-3
-4
-5
-5

-4

-3

-2

-1

Estimated x-axial scale errors (%)

Figure 7: Estimates of cross-axial magnetometer scale error


terms deemed significant according to test. Left hand plots:
The axial scale error is not estimated and the geomagnetic
reference is estimated, if it is significantly different from
initial value. Right hand plots: The axial scale error is
estimated if it is significant, and the geomagnetic reference is
measured.

Since B and are closely related parameters, one


should expect that if one of them is of poor quality, the
other one will also be of poor quality. When
considering this argumentation, the cause of the above
disagreement is not the geomagnetic reference.

It should be noted that large scale errors of similar


magnitude can also be attributed to sensor
imperfections, if by mistake the input magnetic field
intensity B is given a wrong value in the calibration
process. If the a priori magnetic field intensity deviates
significantly from the true average value of the entire
section, it will give rise to similar scale errors on all
three axes. For the magnetic dip angle, an equivalent
error will affect the cross-axial scales equally, but the
axial scale differently. Since the error characteristics
were similar for about 40 % of the sections, there is a
fair reason to believe that magnetic environmental
effects are the main cause for all of these surveys.

Parameter

Mean

Std.dev.

655 nT

390 nT 1616 nT 215 nT

Max.

Min.

0.13 deg. 0.13 deg. 0.43 deg 0.01 deg.

Table 4: Basic statistics of the absolute values of the


difference between the a priori and estimated magnetic field
strength and dip angle.

When considering the sensor specific errors only, it is


unlikely that the true scales are equal for the transverse
axes simultaneously. The small differences between the
transverse scale error estimates (see the lower plots in

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

computed wellbore position. However, removing them


is important to achieve an optimal estimation.

THE FREQUENCY OF GROSS ERRORS


Because undetected outliers might bias the estimated
parameters considerably, this part of the quality control
process is of crucial importance for the Multi-station
analysis.

The main trend was that the measurements in the


beginning of the sections corresponded to the largest
test-statistics. This trend applied irrespective of the bore
hole dimension and inclination. However, they were
several times insignificant and thus they were not
classified as outliers. Since this usually involved a
maximum of three stations, the errors are divided into
groups depending on whether they were detected at the
first three stations, or at other stations. A summary is
given in Table 5. Under similar conditions as in this
study, one must generally expect to reject about 2 % of
the axial magnetometer readings in parts below the first
few stations, and about 1 % of the cross-axial readings.

Methodology. The test for the detection of outliers in


individual sensor readings yi is based on the one
dimensional version of the hypothesis in Equation (2).
This test is considered useful for magnetic surveying
quality control purposes, see [Nyrnes et al. 2005]. To
specify the measurement which is most likely
corrupted, the following datasnooping procedure is
applied; by expanding the design matrix in Equation (2)
with the unit vector ci = [00 1 00]T, and letting i
vary from 1 to n (n is the number of measurements), the
test-statistic in Equation (3) corresponding to every
is evaluated. This procedure
single error estimate
i
can be carried out in a more straightforward way using
residual analysis, see [Koch 1999], [Teunissen 2000]
and [Nyrnes et al. 2005]

Frequency

Outliers detected in the axial


magnetometer measurements.

The lower limit for the significance level i of the


individual tests is adjusted according to i = /n for
survey sections with less than 35 stations, in order to
obtain a type I error probability of approximately
0.05 for the test of H0 (i.e. no outliers in the
measurements). For longer wellsections, a lower limit
of i = 0.0001 is used for the two-tailed test. It is
important to notice that the relation i = /n assumes
independent tests, which is hardly the case. Despite of
these simplifications and assumptions, it must be
considered that the goal of applying such tests is to
remove errors which may cause major harm to the
estimation. It would not have considerable effects on
the overall result if the lower limit for i were set to
0.001 instead of 0.0001, for example.

16
14
12
10
8
6
4
2
0
-4000

-3000

-2000

-1000

1000

2000

3000

4000

Estimates (nT)

Outliers detected in the y-axial


magnetometer measurements (nT).

10

10

8
Frequency

Frequency

Outliers detected in the x-axial


magnetometer measurements (nT).

6
4

750

1500

Estimates (nT)

Outliers in the magnetometer measurements. The


results are presented in the Figure 8 and the Tables 5
and 6.

4
2

2
0
-750

2250

0
-2000 -1000

1000

2000

Estimates (nT)

Figure 8: Magnetometer measurements classified as outliers.


Upper plot: Outliers detected in the axial measurements. The
majority is smaller than 500 nT. Lower plots: Outliers
detected in the cross-axial magnetometer readings.

The estimates of the axial magnetometer errors are


presented by the histogram to the upper in Figure 8. It
can be seen that the largest amount of estimates are
smaller than 500 nT. Some errors were estimated to be
even larger than 4000 nT in absolute value. The crossaxial errors are shown in the lower plots in Figure 8.
The majority of estimated biases are smaller than 1000
nT.

If more than one measurement at the same station is


rejected, the ability to pinpoint the corrupted one out of
the remaining two is almost impossible. It may also be
that both the remaining measurements are corrupted.
Basic information about the excluded stations is given
in Table 6. It can be seen that almost all of the excluded
stations belong to the five categories "St. no. 1" to "St.
no. 5".

Compared with other sources of errors such as borehole


misalignments and declination errors, many outliers
will in practice have no significant influence on the

The values in Tables 5 and 6 clearly indicate that the


magnetic surveys performed at the first few stations in
general are of the poorest quality. As an example,

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

terms of higher redundancy, thus increasing the power


of the statistical tests. This is considered helpful when
the actual wellbore provides low redundancy.

almost 40 % of all stations number 1 are corrupted by


outliers, either in one or at least two measurements.
Magnetic interference from the existing casing can
explain the high frequency of large errors in the
beginning of the wellsections. The errors may also be
related to sensor failure, magnetic disturbances from the
Earth's crust, human errors and sudden fluctuations in
the Earth's magnetic field intensity.

For a few wellsections, several stations were rejected


(totally 15), because more than one accelerometer
measurement were most likely corrupted. A rather large
number of outliers was also detected in single
accelerometer measurements. The estimated outliers are
shown in Figure 9. The majority of biases range
between approximately 5 Gal for all three axes. Table
7 shows that the z-axial errors are about two to three
times more frequent than the transverse ones. One
possible explanation is the movements of floating rigs,
causing the z-axis sensor to become more restless than
the transverse ones.

Magnetometers
St. 1

St. 2 St. 3 Elsewhere

bx

17 (0.9%)

by

25 (1.3%)

bz

13 (20%)

34 (1.8%)

Frequency

Magnetometers

St. 1

St. 2

12 (18%) 9 (14%)

St. 5

Elsewhere

5 (8%)

14

12

12

10

10

8
6
4

8
6
4
2

0
-15 -12 -9 -6 -3

8 (12%)

St. 4

14

St. 3

5 (8%)

Outliers detected in the z-axial


acc. measurements

Outliers detected in the transverse


acc. measurements

Table 5: Total number of outliers in single sensor readings


detected at the 3 first stations and at stations elsewhere in the
well. The total number of measurements is about 1900.

Frequency

Sensor

0
-12 -9

-6

-3

Estimated biases (Gal)

Estimated biases (Gal)

Outliers detected in the transverse


and z-axial acc. measurements
16

Table 6: The number of excluded stations. If at least two


magnetometer measurements turned out to be corrupted, the
whole station was excluded.

14

Frequency

12
10
8
6
4

Accelerometers

2
0
-15

Sensor Number of biases


gx

15 (0.8%)

gy

8 (0.4%)

gz

34 (1.7%)

-10

-5

10

Estimated biases (Gal)

Figure 9: Errors (estimated biases) classified as outliers. Unit:


Gal. Upper plots: Outliers detected in the cross-axial and axial
accelerometer measurements. Lower plot: All outliers lumped
together. Most of the errors range between 5 Gal and 5 Gal
for all three axis.

Table 7: Outliers in single accelerometer readings out of the


total number of measurements (approximately 2000) and
percentage share. The axial accelerometer is more exposed to
outliers than the cross-axial ones.

Comments on the results. Despite of the low


significance level used (about 0.1 ) the large number
of detected outliers clearly indicates that the tail of the
real distribution of the accelerometer and magnetometer
reading errors deviates significantly from the assumed
normal distribution. Removing them is therefore
important in order to avoid unwanted biases in the
estimates.

Outliers in the accelerometer measurements. The


ability to detect outliers in the accelerometer readings is
poorer than for the magnetometer readings. In general,
the errors have to be even larger to be detected. In
addition, it may sometimes be more difficult to pinpoint
the measurement which actually is corrupted. In
directions close to the east-west the corrupted
measurement cannot be pinpointed, since the residuals
become totally correlated. In this case the contribution
from the magnetometer measurements is given only in

10

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

included in the histogram) and 0.027 Gal, both


representing relatively steep 17.5 inch sections.

RANDOM NOISE
In this context, noise is expressed as the estimated
uncertainty of the errors e of the accelerometer and
magnetometer readings. This quantity is of crucial
importance for the analyses. In general, a high noise
level results in low power for the statistical tests of
significance.
The estimated in situ uncertainty yi of a particular

sensor reading yi is given by:

y i = p i1

(6)

where pi is the weight of the measurement and 2 the


estimate of the variance 2 of unit weight. See
Appendix C for details.
The dominant error source in magnetometer
measurements is usually irregular variations in the
Earth's magnetic field. Such time and location
dependent variations are due to the complicated
interactions between the magnetic field and Solar winds
[Torkildsen et al. 1997]. However, irregular errors
related to the performance of the survey tool may also
have significant effects when solar activity is low.
The gravity field can be considered as a stable reference
over time. However, the accelerometer measurements
may be affected by vibrations. This is expected to be
the dominant error source and will usually mask the
effects of random sensor errors completely.

22
20
18
16
14
12
10
8
6
4
2
0
-0.1

Frequency

Frequency

Estimated uncertainties of accelerometer


reading errors (Gal)

0.0

0.1

0.2

0.3

0.4

Standard deviation (Gal)

0.5

0.6

Figure 11: Upper plot: Standard deviations (linear


interpolated between 69.66, 62.07 and 60.13 degrees latitude)
are extrapolated assuming an 11 year periodic solar activity.
The standard deviations are based on the Geomagnetic
Reference Report [Torkildsen et al. 1997]. Lower plot:
Estimated noise in magnetic survey data vs. latitude and
years.

Estimated uncertainties of magnetometer


reading errors (nT)
22
20
18
16
14
12
10
8
6
4
2
0

The histogram to the right in Figure 10 shows that the


estimated standard deviations of the magnetometer
measurements are lower than 100 nT for almost 80 %
of the survey sections, while about 60 % are lower than
60 nT. One is even lower than 20 nT. The estimates are
plotted against latitude and years, see the lower plot in
Figure 11 and also Figure 12. The noise levels are
highest for northern areas and lowest for southern areas.
The same trend can also be seen from the contour plot
in the upper part of Figure 11, where the noise levels
for the years 1991-1996 are based on the time and
location dependent variations (noise) of the magnetic
field observed at monitor stations in the North Sea and
Norwegian Ocean area [Torkildsen et al. 1997]. The
plot also shows the future noise predicted under the
assumption of a sun activity cycle of 11 years. A
comparison between the upper and lower plots in
Figure 11 show some similar trends, especially for the

0 20 40 60 80 100120140160180200

Standard deviation (nT)

Figure 10: The estimated noise standard deviations in the


accelerometer and magnetometer measurements.

Comments on the results. The estimated standard


deviations of the errors in the accelerometer readings
are given in the left plot in Figure 10. Nearly 55 % of
the estimates have their values within the interval: 0.15
Gal - 0.25 Gal. Moreover, about 20 % of the estimated
standard deviations are larger than 0.25 Gal, and almost
15 % are even larger than 0.4 Gal. The two most
extreme noise levels are approximately 17 Gal (not

11

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

latitude dependent noise. The results indicate that the


estimated weight relations between the accelerometer
and magnetometer readings are reasonable. The results
also confirm that the major effects of outliers and
systematic errors are removed from the measurements.

when taken into consideration the relatively high


chance for the a priori local gravity to be of poor
quality, the estimated error values do not show an
overall disagreement with the MWD-error model.
For the surveys which had been corrected for axial
magnetic interference, the estimated magnetometer bias
errors can be considered to be fairly in accordance with
the MWD-error model.

It is important to notice that the level of irregular


variations of the magnetic field intensity in a shorter
period of time may differ considerably from the average
over a longer period of time. If the estimated noise level
turns out to be unexpectedly high, it does not
necessarily mean that the functional model (Equation
(1)) is misspecified. The reason might be that the
surveys are performed in a period with higher solar
activity than normal, for example in the case of
magnetic storms.
Noise standard deviation (nT)
versus latitude.

180

180

160

160

Estimated standard deviation (nT)

Estimated standard deviation (nT)

Noise standard deviation (nT)


versus years.

140
120
100
80
60
40
20
0
1998 1999 2000 2001 2002 2003 2004 2005 2006

140
120
100
80
60

Sometimes the axial scale errors turned out to be clearly


significant. It was shown that this also could be
attributed to errors in the Earth's magnetic reference
components, and it could therefore not be proved
whether sensor imperfections were the cause.

40
20
0
56

58

60

Year
Estimated std.dev. of magnetometer
measurements (nT).

The estimates of the cross-axial magnetometer scale


errors were up to a few percent in magnitude for several
surveys. The errors were insignificantly different from
zero for about 60 % of the surveys. It was pointed out
that due to the high noise level in the measurements, it
will usually not be possible to detect errors smaller than
the corresponding MWD-error model value (0.16 %).
Therefore, any agreement or disagreement with the
MWD-error model could not be proved. Since the
transverse error estimates were similar and negative
(indication of damping), the magnetic properties of the
drilling fluid was most likely the reason.

62

64

66

68

Latitude
Estimated std.dev. of magnetometer
measurements (nT).

Gross errors. It was demonstrated how multi-station


estimation techniques can be used to detect gross errors
in single sensor readings. Removing these errors is
necessary to obtain an optimal estimation. The first few
stations were most critical for the magnetometers,
probably due to magnetic interference from existing
casing. The worst case was for the first station of the
survey section. About 40 % of them were affected by
outliers. However, it was often sufficient to reject the zaxis measurement only. In parts below the first few
stations, almost 2 % of the z-axis magnetometer
measurements were rejected, and about 1 % of the
cross-axial readings. For the accelerometers, up to 2 %
of the z-axial measurements were rejected, and about 1
percent of the cross-axial readings.

Figure 12: Left hand plot: The estimated survey uncertainty


(noise) increases for greater latitudes. Right hand plot: The
noise is lowest for the most recent surveys.

SUMMARY AND CONCLUSIONS

The error properties of magnetic directional surveys


from more than 60 wellbore sections have been
investigated using multi-station estimation techniques.
The most important results are summarized in Table 8.
The ISCWSA error term values. First, the properties of
systematic error terms were investigated. The
International Steering Committee on Wellbore Survey
Accuracy (ISCWSA) has developed error models for
magnetic directional surveys (MWD). The results
showed that the cross-axial accelerometer bias errors
were in accordance with the error model, while the
magnetometer bias errors tended to be somewhat lower
than modelled.

The large number of detected outliers despite of the low


significance level used (about 0.1 ) indicated that the
normality assumption of the measurement errors is not
reasonable.
Random errors. Finally, the properties of the random
errors ("noise") in the surveys were evaluated. The
estimated noise standard deviations in the
accelerometer readings ranged between 0.15 Gal and
0.25 Gal for about 50 % of the examined sections. For
an equivalent number of magnetometer readings the
standard deviations ranged between 40 nT and 60 nT.

Because the estimation of the axial accelerometer bias


and scale error terms involved several uncertain factors,
they were not considered. For the cross-axial
accelerometer scale factor errors the estimates showed a
tendency to larger values than modelled. However,

12

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

The surveys performed in northern areas could be


several times noisier than surveys performed 5 degrees
farther south. The estimated noise levels were in
accordance with the expected noise levels of the
magnetic field. These agreements indicate that the
weight relations between the accelerometers and
magnetometers are reasonable, and that the major
effects of outliers and systematic errors are removed
from the measurements.

The quality of the Earths reference components. The


results of the analyses showed that the gravity values
often provide insufficiency quality for multi station
estimation purposes.

The results showed that the quality of the input Earth's


magnetic field intensity sometimes may be worse than
expected. Except from this, nothing indicated that the
overall quality of the magnetic field intensity and dip
angle were unacceptable.

Accelerometers

Sensor Error term ISCWSA


rand.
gx

gy

bias

ACKNOWLEDGEMENTS

RMS
0.34 Gal

0.05 %

0.084 %

rand.

0.34 Gal

scale

The authors acknowledge Statoil ASA for permission to


publish this paper and for providing the survey data.

0.39 Gal 0.47 / 0.37 Gal 0.8 %

scale

bias

Outliers

APPENDIX A: COORDINATE SYSTEMS

0.39 Gal 0.46 / 0.38 Gal 0.4 %


0.05 %

gx

gy

High side
direction

bx

0.096 %

Horizontal plane

by
gz

gz

rand.

0.34 Gal

bias

0.39 Gal

scale

0.05 %

bx

73 nT

bias

70 nT

95 / 64 nT

scale

by

bz

High side
toolface
Inclination

(Plumb line)

bz

1.7 %

Figure 13: Figurative definition of the instrument xyzcoordinate system, high side toolface and inclination. The xand y-axes are often referred to as the cross-axial direction,
transverse to the along-hole (z-axial/axial) direction.

Magnetometers

rand.

High side
direction

0.9 %

0.16 % - 1.06 / 0.65 %

rand.

73 nT

bias

70 nT

72 / 47 nT

scale

0.16 %

1.01 / 0.62 %

rand.

73 nT

bias

165 nT

594 nT

scale

0.16 %

Horizontal plane
Magnetic
north

1.3 %

A
Am

Horizontal
wellbore
direction

(Earth's magnetic
field vector)

(Plumb line)

1.8 %
Figure 14: Figurative definitions of the azimuth A, magnetic
azimuth Am, magnetic declination and the magnetic dip
angle . At a particular point, is defined as the angular
difference between the horizontal component of the Earth's
magnetic field vector and the true north. It is by definition
positive when magnetic north lies east of true north, and
negative when magnetic north lies west of true north.

Table 8: Third column: Error values according to the MWDerror model developed by ISCWSA. Fourth column: RMSvalues based on the estimates corresponding to significant
parameters. RMS-values behind virgule include zero values
for insignificant parameters. The RMS for the axial
magnetometer bias is representative for surveys which are
corrected for axial magnetic interference. Fifth column:
Number of single sensor readings rejected (percent) in parts
below the first three stations.

13

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

where k is the number of iterations needed to make the


elements in d smaller than some preset amount close

APPENDIX B: THE GAUSS-NEWTON METHOD

Again, we consider the linear regression model


introduced in Equation (1):

to zero. See [Koch 1999] for more details.

d
E(y ) = y e = [X Z]
d

APPENDIX C: THE PROPERTIES OF LEAST


SQUARES

(7)

The properties of the least squares estimators are in the


following given in accordance with the theory of linear
algebra, see [Koch 1999]. Theoretically, these
properties apply only approximately for nonlinear least
squares. Suppose in the following that represents all
fixed parameters (angular components and systematic
errors). It can be shown that if e ~ N (0, 2 P 1 ) , the
least squares estimators are normally distributed:

with E(e) = 0, Cov(e) = Cov(y) = ee = 2Qee. These


equations are often referred to as observation equations
or just error equations. In our case the functional
relationship between E(y) and [ ]T is nonlinear, so
instead of Equation (8) we get the inconsistent system
of nonlinear equations:
y e = h( [ ]T )

(8)

~ N (, 2 ( X T PX) 1 )
E(y ) ~ N( X, 2 X( X T PX) 1 X T )

An approach to solve this nonlinear least squares


problem is that of Gauss-Newton. Using the first two
terms of a Taylor expansion of Equation (9) to form a
linear approximation about initial values
gives:

e ~ N(0, ( P

[ 0 0 ]T

2 =

of the unknown

corrections [d d ]T is obtained by solving the


Equations (10) iteratively subject to the constraint
1
e T Q ee
e = minimum. The unknown corrections are
estimated by:
d = ( X T PX) 1 X T Pf

e T Pe

=
nu nu

(n u ) 2
~ 2 (n u )
2

(10)

(16)

(17)

(18)

The residuals e are for nonlinear least squares given by


e = y E (y ) , where the elements of E( y ) are given by
the last update of the approximate measurements h 0
(defined in Equation (11), Appendix B). In the case

when 2 is unknown, the covariance matrix C


of

(11)

is estimated by:

The estimator is approximated by the last update of:


i +1 = i + d i , i { 0,..., k}

(15)

and distributed as:

The i'th element in f can be considered as the difference


between the actual measurement and the approximate
measurement:
f i = y i h i ([1 , 2 ,..., u 1 , 2 ,... r ]T ) 0

X( X PX) X )

(14)

this it follows that the unbiased estimator 2 of the


variance 2 of unit weight can be computed from:

respect to the elements in [ ] .

which is chi-squared distributed with n-u degrees of


freedom, ~ 2 (2n u ) with mean E() = n u. From

where h is a vector of nonlinear functions of the


unknowns, [X Z] is a matrix of partial derivatives with

= e T Pe

(9)

The least squares estimator d d

The weighted residual sum of squares is given by:

y e h( 0 + d, 0 + d) =
d
= h( 0 , 0 ) + [X Z]
d

(13)

= 2 Q = 2 ( X T PX) 1
C

(12)

14

(19)

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

where Q is the matrix of cofactors. See [Koch 1999]


for further details. The application of the least squares
criterion requires no prior distribution about the
measurement errors to be specified. However, in the
case of normally distributed observation errors, the
method of least squares and the maximum likelihood
method lead to identical unbiased estimators for the
unknown parameters. However, the least squares
criterion applied to normally distributed errors also
provide unbiased and minimum variance properties for
the estimate 2 of the variance 2 of unit weight.

Offshore formula:
G = 978030 + 5186 sin

comparing G derived from 3-component


accelerometer readings with the nominal
value.
using
2-accelerometer
systems
for
determination of inclination.

(21)

G
(22)

REFERENCES
Brooks AG, Gurden PA, Noy KA (1998) Practical
Application of a Multiple-Survey Magnetic
Correction Algorithm. Paper SPE 49060, 1998 SPE
Annual Technical Conference and Exhibition, New
Orleans, 27 September 30 October 1998
Ekseth R (1998) Uncertainties in connection with the
determination of wellbore positions. ISBN 82-4710218-8, Doctoral thesis 1998, Norwegian University
of Science and Technology, 1998:24 IPT-rapport
Heiskanen WA, Moritz H (1993) Physical geodesy.
Technical University Graz, 1993
Koch KR (1999) Parameter Estimation and Hypothesis
Testing in Linear Models. ISBN 3-540-65257-4,
Springer-Verlag 1999
Miller AJ (2002) Subset Selection in Regression. 2nd
edition, ISBN 1584881712, Chapman & Hall/CRC,
2002
Moritz H (2000) Geodetic Reference System 1980
http://www.
gfy.ku.dk/~iag/HB2000/part4/grs80_corr.htm
Nyrnes E, Torkildsen T, Nahavandchi H (2005)
Detection of Gross Errors in Wellbore Directional
Surveying with Emphasis on Reliability Analyses.
Accepted for publication in Kart og Plan vol. 2 2005
Teunissen PJG (2000) Testing theory: an introduction.
ISBN 90-407-1975-6, Delft University Press 2000
Torkildsen T, Sveen RH, Bang J (1997) Time
Dependent Variations of Declination. Geomagnetic
Reference report No. 1, IKU Petroleum Research
1997
Torkildsen T, Edwardsen I, Fjogstad A, Saasen A,
Amundsen PA, Omland TH (2004) Drilling Fluid
affects MWD Magnetic Azimuth and Wellbore
Position. Paper SPE 87169, 2004, IADC/SPE Drilling
Conference Dallas, Texas, 2-4 March 2004

This appendix presents a simplified and generalized


approach to calculate G. The formula precision is better
than 0.0005 m/s2 (50 mGal), and is assumed to be
precise enough for the quality control of directional
surveying services. However, the service companies
should derive G-reference values in a more reliable way
for calibration purposes.

Formulas. The main trend is according to the Geodetic


Reference System 1980. The original formula is
truncated and given a best fit globally (within 3 mgal
precision).
Additionally the Free-air, Bouguer and regional
Isostatic corrections are applied.
General formula:
G = 978030 + 5186 sin 2 () +
0.309 (TVD H 0 ) 0.084 TVD

( ) + 0 . 14 TVD

See [Moritz 2000]. See also different textbooks, for


example [Heiskanen, Moritz 1993].

Using precise gravity values is important when

Onshore formula:
G = 978030 + 5186 sin 2 () +
0.10 TVD 0.31 H 0

APPENDIX D: GRAVITY REFERENCE FOR


DIRECTIONAL SURVEYING

offshore sites by using a bulk density of 2.0


g/cm3, and approximating installation height to
0 m.
onshore sites by using a bulk density of 2.5
g/cm3 .

(20)

where G is the predicted gravity in mgal (1mgal =


10-5m/s2), H0 is the height above MSL at installation in
metres, TVD is the vertical depth in metres measured
from installation, is the latitude and is the bulk
density in g/cm3 from installation to the actual TVD.
The missing Anomaly, Topographic and local Isostatic
corrections may cause significant errors (>50 mgals) in
mountainous areas.
The above formula can be simplified further for:

15

SPWLA 46th Annual Logging Symposium, June 26-29, 2005

Williamson HS (2000) Accuracy Prediction for


Directional Measurement While Drilling. Paper SPE
67616, SPE Drilling and Completion 15 (4),
December 2000
Wilson H, Brooks AG (2001) Wellbore Position Errors
Caused by Drilling Fluid Contamination. Paper SPE
71400, 2001 SPE Annual Technical Conference and
Exhibition, New Orleans, 30 September 3 October
2001

ABOUT THE AUTHORS


Erik Nyrnes is a research fellow at the Norwegian
University of Science and Technology (NTNU). He
holds a masters degree in surveying from the same
university.
Torgeir Torkildsen is a specialist in directional
surveying at Statoil ASA Norway. He holds a doctoral
degree in geodesy from NTNU.
Hossein Nahavandchi is an associate professor in
geodesy at NTNU. He holds a doctoral degree in
geodesy from the Royal Institute of Technology,
Sweden.

16

S-ar putea să vă placă și