You are on page 1of 12

SPE/IADC 125677

Minimum Requirements for Multi-Station Analysis of MWD Magnetic


Directional Surveys
Erik Nyrnes, SPE, StatoilHydro ASA, Torgeir Torkildsen, SPE, SINTEF Petroleum Research, Harry Wilson, SPE,
Baker Hughes

Copyright 2009, SPE/IADC Middle East Drilling Technology Conference & Exhibition

This paper was prepared for presentation at the SPE/IADC Middle East Drilling Technology Conference & Exhibition held in Manama, Bahrain, 26-28 October 2009.

This paper was selected for presentation by an SPE/IADC program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have
not been reviewed by the Society of Petroleum Engineers or the International Association of Drilling Contractors and are subject to correction by the author(s). The material does not
necessarily reflect any position of the Society of Petroleum Engineers or the International Association of Drilling Contractors, its officers, or members. Electronic reproduction, distribution, or
storage of any part of this paper without the written consent of the Society of Petroleum Engineers or the International Association of Drilling Contractors is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE/IADC copyright.

Abstract
Multi-Station Analysis (MSA) is a technique widely used in MWD directional surveying to provide additional quality control
and to correct for systematic errors. Although the method’s potential to enhance survey quality has been demonstrated
through several publications, experience has shown that MSA can produce unstable solutions and poorly interpretable results.
In such cases, it is likely that the uncertainty estimate assigned to the data is invalid. Unfortunately, there is no standard in the
oil industry defining the correct use of MSA. Different companies have developed different requirements and acceptance
criteria, making it difficult to judge and compare survey quality between companies. The industry should therefore seek to
establish a common standard for the application of MSA.
This paper presents a set of fundamental requirements that have to be fulfilled in order to ensure the correct application of
MSA in drilling operations. The requirements consist of a set of mathematical rules and corresponding acceptance limits.
Survey quality can be verified against any MWD error model by using the same mathematical framework, but changing the
acceptance limits. Requirements and acceptance criteria are presented for two specific error models: Basic MWD without
axial interference correction (Williamson 2000) and the same model assuming enhanced geomagnetic referencing.
The results presented in this paper show which systematic errors can be estimated for any given data set. A mechanism
for determining tolerable survey noise is also included. This prevents misapplication of MSA under poor surveying
conditions such as magnetic storms and drill-string vibration. The objective has been to develop requirements that are easy to
apply in operations and which can form the basis for standardization across the industry.

Introduction
Multi-Station analysis (MSA) is a more powerful survey quality evaluation method than conventional single station
calculations (Ekseth et al. 2006). It makes it possible to identify and quantify different types of systematic errors, providing
greater proof of whether or not surveys meet specification (see e.g. Brooks et al. 1998). Where systematic errors of
significant magnitude are identified they can be corrected for. However, the ability of MSA to correct failed surveys and
provide increased confidence that surveys meet their stated specification is limited by several factors. For example, only a
limited number of systematic error terms can be estimated accurately for a given set of survey data, this being dependent on
factors such as the geometry of the wellbore, the number of survey stations and the number of estimated error terms, see e.g.
Nyrnes et al. (2005), Nyrnes and Torkildsen (2005); Torkildsen et al. (2004). In current common practice, it is not always
clear how the application of MSA to a given survey log affects its status with respect to its own accuracy specification.
Improved accuracy may be claimed, but the link between MSA data manipulation and the error model that quantifies
accuracy is usually not explicit.
In this paper, a basic MSA methodology suitable for adoption across the industry is proposed. The methodology consists
of a set of requirements and acceptance limits that have to be fulfilled when applying MSA. The requirements can be applied
to any survey log to determine which error terms can be validly estimated. In addition, mechanisms are included to ensure
that the random noise level in the survey data is of tolerable magnitude. The requirements are easy to apply and
straightforward to implement, and should therefore be easily communicated throughout the survey industry. An earlier
version of the MSA requirements was presented and discussed at the 26th SPE WPTS (former ISCWSA) meeting in
November 2007 (Nyrnes et al. 2007).
2 SPE/IADC 125677

It is a common practice to describe the error characteristics of a magnetic MWD directional surveying sensor by two
systematic error terms; a bias error and a scale error, see e.g. Williamson (2000). In addition, random errors will be present.
MSA acceptance criteria should be derived from the error model applicable to the MWD instrument and survey method used
to acquire the data. The example acceptance criteria presented in this paper are for two specific MWD error models: Basic
MWD without axial interference correction (Williamson 2000) and the same model with the addition of enhanced
geomagnetic referencing. By fulfilling the requirements, compliance with the actual error model can be claimed. If
corrections are applied, they merely bring failed surveys back within the relevant error model’s specification.
The MSA requirements consist of a set of simplified mathematical measures and corresponding acceptance limits. The
mathematical measures describe the actual well geometry for the given data set. An example of such a measure is the
arithmetic mean of the axial magnetometer readings, which indirectly describes the average direction of the wellbore with
respect to the vector of the Earth’s magnetic field. These so-called geometrical measures have been assigned suitable
acceptance limits. When applying MSA, the decision to estimate a specific error term is based on there being an acceptable
match between the calculated value of the geometrical measures of relevance and the associated acceptance limits. The
geometry measures and acceptance limits for accelerometer and magnetometer error terms are given in Tables 1 and 2
respectively, along with expressions for test variables and acceptance limits for the noise tests. By fulfilling the requirements
presented in this paper, the resulting uncertainty of the estimated inclination, azimuth and systematic errors should be
consistent with the relevant MWD-error model.
The derivation of the noise test for accelerometer measurements are shown in Appendix A. An example of use of the
MSA requirements is given in Appendix B. Theory of basic statistics and least squares estimation are described in Appendix
C, while Appendix D presents a flow chart illustrating the process of applying the MSA-requirements. It should be noted that
the requirements are valid for an orthogonal instrument coordinate system of which the z-axis of the surveying instrument
coincides with the direction of borehole, see e.g. Williamson (2000). The z-axis of the survey instrument will also be referred
to as the axial direction.

Derivation of the MSA-requirements


Derivation of the requirements was based solely on applying MSA to simulated survey data. The simulations were performed
under the assumption of random noise standard deviations of 0.003 ms‫־‬² and 70 nT for accelerometer and magnetometer
measurements respectively. These numbers represent realistic estimates of the noise levels in MWD magnetic directional
surveys, see Nyrnes et al. (2005). Correspondingly, the input standard deviations of the Earth’s magnetic field intensity and
dip angle have respectively been set to 130 nT and 0.2° for the standard referencing and 65 nT and 0.12° for the enhanced
referencing. A wide variety of wellbore geometries was considered in the simulations, and the number of survey stations was
varied for each case to see how this would affect the results. In this way, potential critical situations were identified for each
particular combination of systematic error terms. The identification was done by means of assessing the standard deviations
of the estimated inclination, azimuth, and systematic error terms obtained by MSA for each particular case. The results, in
terms of estimated standard deviations, were evaluated against the actual MWD error model, and classified as unacceptable
when significant deviations from the error models were detected.

Description of the MSA-requirements


This section presents a detailed description of the MSA requirements to assist users in implementation and understanding of
the requirements.

Common requirements for accelerometer and magnetometer error terms


Some of the proposed requirements apply equally for accelerometer and magnetometer measurements. The common
requirements are sufficient variation in high side toolface, and sufficient number of survey stations.

Variation of high side toolface


A fundamental requirement for applying MSA is minimum spread of the high side toolface angles among all surveys in the
data set. A variation in toolface is necessary to obtain accurate estimation of most error terms. The requirement is that the
difference between any two neighbouring toolface angles is not greater than 100º. This will prevent the measurements from
being clustered in one sector.

Number of survey stations


This requirement is introduced to ensure sufficient redundancy in the estimation. The requirement is that the number of
survey stations is at least three times the number of estimated error terms and always greater or equal to four.

Requirements for accelerometer measurements


The requirements for estimation of systematic accelerometer error terms are shown in Table 1. It is recommended that
accelerometer measurements be processed before magnetometer measurements, since magnetometer measurements will
normally have a minor influence over the estimation of accelerometer specific parameters. The corrected accelerometer
measurements are then used as input for the magnetometer processing.
SPE/IADC 125677 3

Systematic errors
A maximum of five systematic accelerometer error terms can be estimated. These are the two cross-axial bias errors (bias gx,
bias gy), the two cross-axial scale factor errors (scale gx, scale gy), and the axial bias error (bias gz) or the axial scale factor
error (scale gz). Note that an estimation of any accelerometer error term always requires adequate toolface variation.
An estimation of the cross-axial accelerometer biases (bias gx, bias gy) requires that the average of the absolute values of
the axial accelerometer measurements is less than 9.4 ms‫־‬². This means that the average inclination has to be greater than
approximately 15º. However, as is indicated by the footnote (1), this requirement does not have to be fulfilled if the survey is
a rotational shot of which the inclination for each survey is constrained to be equal. For the cross-axial accelerometer scale
errors the average of the z-accelerometer measurements has to be less than 8.5 ms‫־‬², entailing that the average inclination has
to be greater than approximately 30º. The reason for using the size of the axial measurements as the criterion for estimation
of cross-axial error terms is because they directly reflect the size of the cross-axial measurements. The larger the cross-axial
measurements are, the more accurate the cross-axial bias and cross-axial scale estimates will be. Note that the gravity field G
can be substituted for scale gx or scale gy as indicated by footnote (4).
In order to estimate the axial accelerometer bias error, one requirement is that the average of the z-axis measurements is
greater than 2.5 ms‫־‬². This means that the average inclination has to be less than approximately 75º. As indicated by footnote
(3), an estimation of the axial bias together with the cross-axial scale factor errors requires a minimum spread of z-axis
measurements of 2.5 ms‫־‬². It is indicated by footnote (2) that the axial bias error can be replaced with the axial scale error
(scale gz), but they are not allowed to be estimated simultaneously.

Gravity field vector


As mentioned above, the gravity field G can be estimated in some situations. It can be replaced with one of the cross-axial
scales or the axial bias, as indicated by footnotes (2) and (4).

Random noise
Two different test approaches have been developed to quality control the level of random noise in the survey data; the Multi-
Station overall accelerometer noise test which is used to verify the overall noise level in the data set, and the Multi-Station
accelerometer gross error test which provides a useful verification of the noise level in each individual survey station. The
tests are developed to have a similar sensitivity as the conventional single-station G-total test introduced by Torkildsen and
Lotsberg (2001). The noise tests are based on a chi-squared distributed test variable and a significance level of 0.27 %, which
is the same significance level that was used for the various survey quality tests presented by Ekseth et al. (2006, 2007).
When applying the Multi-Station overall noise test, the noise level is considered acceptable if the estimated survey noise
is less than or equal to a certain test limit, that is, if the following condition is satisfied:

1 n
0.006 ms −2 ⎛ 2 ⎞ Eq. (1)

( n − e ) i =1
ΔGi2 ≤
⎛ e⎞
⎜⎜1 +

⎟⎟
n−e ⎠
⎜1 + ⎟
⎝ 5⎠

where ∆Gi is the difference between the nominal gravity and the gravity derived from the accelerometer readings which have
been corrected for the estimated systematic errors, e is the number of error terms estimated and n is the number of survey
stations. The nominal gravity can be predicted from independent sources, see e.g. Nyrnes et al. (2005). The expression on
the left-hand side of Eq. (1) is the unbiased estimator of the measurement noise standard deviation and the expression on the
right-hand side is its acceptance limit. The constant 0.006 ms‫־‬² denotes an approximation of the basic standard deviation of
an accelerometer measurement, that is, the a priori noise level when the contributions from all sensor specific error
components are included, such as white noise and systematic errors. The term (1 + e/5) compensates for the varying number
of estimated systematic errors, while the term (1 + 2/ n - e ) compensates for variation in redundancy.
The Multi-Station gross error test is similar to the previously mentioned test, except that it is applied to every survey
station in the data set. The actual survey station of interest is not a gross error if:

0.018 ms −2 , Eq. (2)


ΔGi ≤
⎛ e⎞
⎜1 + ⎟
⎝ 5⎠

where the noise term 0.018 ms‫־‬² is the approximate basic standard deviation of 0.006 ms‫־‬² multiplied by three to achieve a
significance level of 0.27 %. Equation (2) shows that if the number of surveys is one and no systematic error terms are
estimated, the test limit is 0.018 ms‫־‬², which is equal to the test limit of the conventional single station G-total test
(Torkildsen and Lotsberg 2001).
4 SPE/IADC 125677

Parameter Requirement
max ( Δα i ) < 100 °
general n ≥ 3e
n≥4
1 n (1)
bias gx, bias gy ∑ gzi < 9.4 ms -2
n i =1
1 n
scale gx, scale gy (4) ∑ gzi < 8.5 ms -2
n i =1
n
1
bias gz (2) n
∑ gz i > 2.5 ms - 2
i =1

{max ( gzi ) − min ( gzi )} > 2.5 ms -2 (3)


1 n
0.006 ms −2 ⎛ 2 ⎞
max. noise (n−e)
∑ ΔGi2 ≤
⎛ e⎞
⎜⎜1 +

⎟⎟
n−e ⎠
i =1
⎜1 + ⎟
⎝ 5⎠
0.018 ms −2
ΔGi ≤
max. single survey error ⎛
⎜1 + ⎟
e⎞
⎝ 5⎠

Table 1: Requirements for the estimation of accelerometer error terms.

(1): Not necessary in case of rotational shots (i.e. the inclinations are equal).
(2): Can be substituted with scale gz or G.
(3): Only necessary if the cross-axial scales are estimated together with bias gz.
(4): Scale gx or scale gy may be substituted with G.

Requirements for magnetometer measurements


The requirements for estimation of systematic magnetometer error terms are given in Table 2. Whether the accelerometer
measurements are corrected for systematic error sources or not, they should be input with realistic uncertainties in order to
have proper influence when MSA is applied to magnetometer measurements. Realistic accelerometer uncertainties can be
obtained from suitable MWD error models (e.g. Williamson 2000). Realistic random noise components can be found in
Nyrnes et al. (2005). The test limits of the MSA-requirements presented in this paper are only valid if the dip angle for the
actual location is within the interval ±80°. For other locations the test limits of the requirements need to be adjusted.

Systematic errors
The total number of systematic magnetometer error terms that can be estimated is six. These are three bias errors (bias bx,
bias by, bias bz) and three scale factor errors (scale bx, scale by, scale bz).
An estimation of the cross-axial magnetometer biases (bias bx, bias by) only requires sufficient toolface variation, as long
as the number of survey stations is sufficient and the dip angle criterion is satisfied. In order to estimate the axial
magnetometer bias error (bias bz), the wellbore attitude has to be sufficiently offset from the horizontal magnetic east-west
attitude. The average deviation from horizontal east-west has to be at least 35º when applying standard magnetic referencing.
More specifically, the average wellbore direction has to be outside a circular window with radius 35º centred exactly in
horizontal east-west. Mathematically, this criterion is fulfilled if the average of the absolute values of the product of sin(I)
and sin(A) is less than 0.82. For the enhanced magnetic referencing the requirement is less stringent. In this case the deviation
from east-west has to be at least 25º, which means that the average of the absolute values of sin(I)sin(A) is less than 0.91.
A reliable estimation of the cross-axial magnetometer scale factor errors (scale bx, scale by) requires that the average of
the absolute values of the axial magnetometer measurements is less than 40 % of the total magnetic field intensity for the
standard referencing, and less than 70 % of the total field intensity when enhanced referencing is applied. Note that when
estimating the cross-axial scale errors together with the z-axis bias (bias bz), the requirement is somewhat tighter for the
enhanced referencing for wells with limited variation in inclination and azimuth. As indicated by a footnote, this applies if
the spread of the axial magnetometer or the axial accelerometer measurements is less than a certain amount. This is
considered necessary to avoid certain orientations where a collective estimation of these three error terms provides uncertain
results, see e.g. Nyrnes and Torkildsen (2005). The east-west requirement is tightened 5º by reducing the sin(I)sin(A)
acceptance limit from 0.91 to 0.87. Additionally, the average of the axial magnetometer readings has to be smaller than 50 %
of the total magnetic field intensity, instead of 70 % when the cross-axial scales are estimated without the axial bias.
The requirements for estimation of the axial scale error term (scale bz) are that the spread of the z-axis measurements is
greater than 60 % of Earth’s total magnetic field intensity, and that the sinIsinA–test value is less than 0.91. Note that the
estimation of the axial scale error term is not allowed for the standard referencing.
SPE/IADC 125677 5

Magnetic field reference components


The magnetic field intensity B and the magnetic dip angle Ө can be substituted with one of the magnetometer scale factor
errors as indicated in the last row of Table 2. For example, if standard referencing is applied, both of B and Ө can be
estimated concurrently at the expense of one of the cross-axial scale error terms (scale bx, scale by). However, if enhanced
referencing is applied, also the axial scale error term (scale bz) can be substituted by B and Ө in addition to scale bx and scale
by.

Random noise
Two tests have been developed to quality control the noise level in the magnetometer measurements; the Multi-Station
overall magnetometer noise test and the Multi-Station magnetometer gross error test. The derivation of these tests can be
done in a similar way as for the accelerometer noise tests described in Appendix A. The tests are designed to have a similar
sensitivity to the conventional single-station combined B-total/Dip test introduced by Torkildsen and Lotsberg (2001).
The standard referencing version of the Multi-Station overall noise test is given by:

∑( )
1 n
85 nT ⎛ 2 ⎞ Eq. (3)
ΔBi2 + (Bi Δθ i )2 ≤ ⎜1 + ⎟⎟
( 2n − e ) i =1 ⎛ e ⎞ ⎜⎝ 2 n − e ⎠
⎜ 0.4 + ⎟
⎝ 8⎠

where the differences ∆Bi and ∆Өi are calculated from the magnetometer measurements obtained at the i’th survey station,
and which have been corrected for estimated systematic errors. The expression on the left-hand side of Eq. (3) is the unbiased
estimator of the survey noise in magnetometer measurements, while the expression on the right-hand side represents the
acceptance limit of the estimated survey noise. The term 85 nT/(0.4 + e/8) predicts the a priori noise standard deviation of
magnetometer measurements according to the number of estimated systematic errors, while the term (1 + 2/ 2n - e )
compensates for the change in redundancy. The only difference between the standard referencing version and the enhanced
referencing version of the Multi-Station overall noise test is that the numerator on the right-hand side of Eq. (3) is reduced
from 85 nT to 60 nT. If the estimated survey noise exceeds the acceptance limit, the conclusion is that the noise in the actual
survey log is unacceptably high and therefore the survey log does not qualify for MSA.
The objective of the Multi-Station gross error test is to detect gross errors in magnetometer measurements at individual
survey stations. This test is based on a modification of the previous mentioned test. For standard referencing the test is
expressed by:

255 nT Eq. (4)


ΔBi2 + (Bi Δθ i )2 ≤
⎛ e⎞
⎜ 0.4 + ⎟
⎝ 8⎠

The expression on the left-hand side of Eq. (4) is simply the B-total/Dip vector at the i’th survey station (Torkildsen and
Lotsberg 2001), while the expression on the right-hand side is the acceptance limit of the test, which is simply an
approximation of the expected standard deviation of the measurement noise multiplied by three to obtain a significance level
of approximately 0.27 %. If the term on the left-hand side of Eq. (4) exceeds the acceptance limit given on the right-hand
side, the conclusion will be that the actual survey station has a gross error and shall therefore be rejected from the dataset.
The enhanced referencing version of the Multi-Station gross error test is obtained by reducing the numerator of Eq. (4) from
255 nT to 180 nT.

Implementation
The intention is that the MSA requirements presented in this paper shall be applied more or less automatically. It is therefore
important that the different parts of the requirements are applied in the correct sequence. The following sequence is
recommended:

Step 1: Check that Earth’s magnetic dip angle for the actual location is within ±80°. If the dip angle falls outside this interval,
the MSA-requirements are not valid. Use the mid-point of the survey log as reference.
Step 2: Calculate the variation in high side toolface. If the criterion of minimum toolface variation is not satisfied, the survey
data are not qualified for use of MSA.
Step 3: Count the number of survey stations in order to specify the total number of error terms that can be estimated.
Step 4: Specify which types of systematic error terms that can be estimated by comparing relevant geometry measures with
corresponding acceptance limits. If these error terms are greater in number than the number specified in Step 3, one or more
error terms have to be excluded such that the total number of systematic error terms is less than or equal to that of Step 3.
Step 5: Apply MSA and correct the sensor readings for the effects of systematic errors.
Step 6: Apply the Multi-Station noise tests to the corrected survey data. Remove survey stations that do not pass the Multi-
Station gross error tests and return to Step 2. When no gross errors are left, apply the Multi-Station overall noise test. Reject
6 SPE/IADC 125677

the entire dataset if the total survey noise is greater than accepted, i.e. if the Multi-Station overall noise test fails.

The overall process of applying the MSA requirements is shown by the flowchart in Appendix D.

STANDARD REF. ENHANCED REF.

Parameter Requirement Requirement


max (Δα i ) < 100 ° max ( Δα i ) < 100 °
-80º < Ө < 80º -80º < Ө < 80º
general n ≥ 3e n ≥ 3e
n≥4 n≥4
bias bx, bias by Only general requirement necessary Only general requirement necessary

1 n 1 n
bias bz ∑ sin I i sin Αi < 0.82
n i =1
∑ sin I i sin Αi < 0.91
n i =1

1 n 1 n
scale bx, scale by ∑ bzi < 0.4B
n i =1
∑ bzi < 0.7B
n i =1
n n
1 1
n
∑ sin I i sin Αi < 0.82 n
∑ sin I i sin Αi < 0.87 ( 0.91 )*
scale bx, scale by, i =1 i =1
bias bz 1 n
1 n

n
∑ bzi < 0.4 B n
∑ bzi < 0.5 B ( 0.7 B) *
i =1 i =1

{max (bzi ) − min (bzi )} > 0.6 B


scale bz Not allowed 1 n
∑ sin I i sin Αi < 0.91
n i =1

∑( ) ∑( )
1 n
85nT ⎛ 2 ⎞ 1 n
60 nT ⎛ 2 ⎞
ΔBi2 + (Bi Δθ i )2 ≤ ⎜1 + ⎟⎟ ΔBi2 + (Bi Δθ i )2 ≤ ⎜1 + ⎟⎟
max. noise ( 2n − e ) i =1 ⎛ e ⎞ ⎜⎝ 2 n − e ⎠ ( 2 n − e ) i =1 ⎛ e ⎞ ⎜⎝ 2n − e ⎠
⎜ 0.4 + ⎟ ⎜ 0.4 + ⎟
⎝ 8⎠ ⎝ 8⎠
255 nT 180 nT
max. single survey ΔBi2 + (Bi Δθ i )2 ≤ ΔBi2 + (Bi Δθ i )2 ≤
⎛ e⎞ ⎛ e⎞
error ⎜ 0.4 + ⎟ ⎜ 0. 4 + ⎟
⎝ 8 ⎠ ⎝ 8 ⎠

B and Ө Can be substituted with scale bx or scale by Can be substituted with scale bx or scale by or scale bz

Table 2: Requirements for the estimation of magnetometer error terms, the Earth’s magnetic field intensity and dip angle.

*Acceptance values in parenthesis are to be used if {max (bzi ) − min (bzi )} > 0.15B or if {max ( gzi ) − min ( gzi )} > 0.7 ms −2 .

Summary
A set of fundamental requirements have been developed that make MSA largely user independent, providing consistent and
valid outcomes regardless of the estimation technique being used. The requirements consist of basic mathematical measures
that describe the geometry of the well and the noise level in the surveys. Misapplication of MSA is avoided if the values of
calculated measures do not exceed the respective acceptance limits. The MSA requirements take several factors into
consideration, such as:
• the accuracy of the accelerometer and magnetometer sensor measurements
• the accuracy of the Earth’s gravity and the geomagnetic reference field values
• the number of survey stations
• the wellbore orientation relative to the gravity and geomagnetic field vectors
• the variation of the wellbore azimuth and inclination
• the variation of the high side toolface direction
Highly skilled users may prefer methods which are more optimized for evaluation of MSA results, for instance direct
interpretation of least squares standard deviations of estimated inclination, azimuth and systematic error terms. However,
SPE/IADC 125677 7

although the method proposed in this paper is relatively uncomplicated and the criteria have been simplified, both method
and criteria are based on correct mathematical principles.
Although the example test limits are designed to harmonize with the standard MWD error model, the mathematical
framework can be use to verify survey quality against any MWD error model; only the test limits need to be changed. The
requirements show that, when validating data against the standard MWD error model, MSA cannot provide simultaneous
estimation of all types of systematic errors.

Acknowledgement
The authors acknowledge StatoilHydro ASA, SINTEF Petroleum Research and Baker Hughes for permission to publish this
paper.

Nomenclature
{i,j} Indices: 1,…, n
I Inclination
A Magnetic Azimuth
n Number of survey stations
e Number of estimated systematic errors
gz Nominal along-hole accelerometer reading
bz Nominal along-hole magnetometer reading
α Toolface angle
nT nanoTesla
∆αi Difference between neighbouring toolface directions
∆Gi Difference between the nominal gravity and the gravity derived from the accelerometer readings which have been
corrected for the estimated systematic errors
B Intensity of Earth’s magnetic field
Ө Dip angle of Earth’s magnetic field vector
∆Bi Difference between the nominal magnetic field intensity and the field intensity derived from the magnetometer
readings which have been corrected for estimated systematic errors
Δθ i Difference between the nominal dip angle and the dip angle calculated from the magnetometer readings which have
been corrected for estimated systematic errors

References
Brooks A. G., Gurden P. A., Noy K. A. (1998): “Practical Application of a Multiple-Survey Magnetic Correction Algorithm”. Paper SPE
49060 presented at the SPE Annual Technical Conference and Exhibition, New Orleans, Louisiana, USA, 27 September - 1 October
1998.
Ekseth R., Torkildsen T., Brooks A., Weston J., Nyrnes E., Wilson H., Kowalenko K. (2006): “The Reliability Problem Related to
Directional Survey Data”. Paper SPE 103734 presented at the IADC/SPE Asia Pacific Drilling Technology Conference and
Exhibition, Bangkok, Thailand, 13–15 November 2006.
Ekseth R., Torkildsen T., Brooks A., Weston J., Nyrnes E., Wilson H., Kowalenko K. (2007): “High Integrity Wellbore Surveys: Methods
for Eliminating Gross Errors”. Paper SPE/IADC 105558 presented at the SPE/IADC Drilling Conference, Amsterdam, The
Netherlands, 20 – 22 February 2007.
Koch K.R. (1999): “Parameter Estimation and Hypothesis Testing in Linear Models”, second edition. Springer-Verlag, Berlin, Heidelberg,
New York, 1999.
Nyrnes E., Torkildsen T., and Nahavandchi H. (2005): “Error Properties of Magnetic Directional Surveying Data”. Paper presented at the
SPWLA Annual Technical Conference and Exhibition, New Orleans, Louisiana, USA, 27-29 June 2005.
Nyrnes E., and Torkildsen T. (2005): “Analysis of the Accuracy and Reliability of Magnetic Directional Surveys”. Paper SPE 96211
presented at the SPE/IADC Middle East Drilling Technology Conference & Exhibition, Dubai, UAE, 12-14 September 2005.
Nyrnes E., Torkildsen T., Wilson H., and Brooks A. (2007). “Minimum Requirements for Multi-Station Analyses”. Presentation given at
the 26th meeting of the SPE WPTS, Anaheim, CA, USA, 15 November 2007.
Torkildsen T., Edwardsen I., Fjogstad A., Saasen A., Amundsen P. A., and Omland T. H. (2004): ”Drilling Fluid affects MWD Magnetic
Azimuth and Wellbore Position”. Paper SPE 87169 presented at the IADC/SPE Drilling Conference, Dallas, Texas, 2-4 March 2004.
Torkildsen T., Lotsberg O. (2001): “Acceptance Values for MWD Magnetic Wellbore Surveying”. Presentation given at the 13th and the
14th meeting of the Industry Steering Committee on Wellbore Survey Accuracy (now SPE-WPTS), Harlem, Netherlands, 2 Mars
2001, and Denver, Colorado, USA, 7 June 2001.
Williamson H. S. (2000): “Accuracy Prediction for Directional Measurement While Drilling”. SPE Drill & Compl (December 2000) 221.
8 SPE/IADC 125677

Appendix A

Derivation of the Multi-Station accelerometer noise tests


The Multi-Station accelerometer noise test is given by:

1 n
0 .006 ms −2 ⎛ 2 ⎞ Eq. (A-1)
(n − e)
∑ ΔGi2 ≤
⎛ e⎞
⎜⎜ 1 +

⎟⎟
n−e ⎠
i =1
⎜1 + ⎟
⎝ 5⎠

In the following a brief explanation of the derivation of this test will be given. See previous sections for definition of the
different terms of Eq. (A-1).
An unbiased estimator of the variance of an accelerometer measurement can be calculated by:
n
1 Eq. (A-2)
σˆ 2 =
( n − e ) i =1

ΔGi2

For the test given by Eq. (A-1), the basic a priori standard deviation of an accelerometer measurement is set to 0.006 ms-2.
This approximation is representative for an average instrument attitude where no bias and scale error terms are estimated and
corrected for, and is derived with basis on the systematic error uncertainties of the standard MWD error model (Williamson
2000) and a random error uncertainty of 0.003 ms-2 (Nyrnes et al. 2005). To account for the reduction in variance when
systematic error terms are estimated, the a priori basic standard deviation of 0.006 ms-2 is tuned according to:

0.006 ms −2 Eq. (A-3)


σ=
⎛ e⎞
⎜1 + ⎟
⎝ 5⎠

For example, if two systematic error terms are estimated by the MSA, in the same or two different sensors, the standard
deviation of the accelerometer measurements reduces to 0.006 ms-2/(1+2/5) ≈ 0.0043 ms-2. This tuning is necessary in order
to balance with the unbiased noise estimate on the left-hand side of Eq. (A-1).
The test given by Eq. (A-1) is based on the following chi-squared distributed test variable:

σˆ 2 ( n − e ) σˆ χ(2n − e ) Eq. (A-4)


~ χ(2n − e ) → ~
σ 2
σ (n − e)

where χ( n − e ) denotes a chi-squared distribution with n – e degrees of freedom, and σ̂ 2 and σ 2 denote the estimated and the a
2

priori accelerometer variance respectively. What the Multi-Station noise test actually does is to test whether the estimated
survey noise σ̂ is significantly greater than its a priori value σ . The significance level of the test is 0.27 %, which
corresponds to the limit 3σ when the number of degrees of freedom is one. For this significance level the following
approximation applies:

χ(21−0.0027 , n−e ) ⎛ 2 ⎞, Eq. (A-5)


≈ ⎜⎜1 + ⎟⎟
(n−e) ⎝ n−e ⎠

thus, it follows from Eq. (A-4) that a test limit T for the Multi-Station noise test can be approximated by:

⎛ 2 ⎞ Eq. (A-6)
T ≈ σ ⎜1 + ⎟
⎝ n−e ⎠

which, through Eq. (A-3), equals the expression on the right-hand side of Eq. (A-1). For the Multi-Station accelerometer
gross error test, the acceptance limit T in Eq. (A-6) becomes 3σ .
A similar procedure applies for the derivation of the Multi-Station magnetometer noise tests (Multi-Station B-total/Dip
tests). However, the number of degrees of freedom is expressed by 2n - e for this case.
SPE/IADC 125677 9

Appendix B

Example of application of the MSA requirements


This appendix presents a demonstration case were values for the MSA requirements in Tables 1 and 2 are calculated for an
example survey. It will be explained how to apply the requirements for this particular well. The inclination, toolface and
azimuth values for the example well are given in Table B-1, along with respective accelerometers and magnetometer
measurements. The Earth’s magnetic field intensity and dip angle for the demonstration well are set to 51000 nT and 75°
respectively. It should be noted that the Multi-Station noise tests are not considered in this demonstration.

Accelerometers
The calculated requirements for accelerometer measurements are shown in Table B-2. The maximum toolface interval for the
actual dataset is 52.0º, and that is acceptable. The dip angle criterion is also satisfied. This dataset does therefore have a
potential for MSA, and since the number of survey stations is 15, a maximum of five systematic error terms have the
potential to be estimated, depending on whether the remainder of the requirements are fulfilled.
Let us first consider the cross-axial error terms. Since the arithmetic mean of the axial accelerometer measurements gz is
2.2 m/s2, an estimation of all four cross-axial error terms is approved. However, this is not the case for the axial bias. As
shown in Table B-2 the spread of the axial accelerometer measurements is 3.62 ms-2, and that is sufficient, but since the
arithmetic mean is less than the limit of 2.5 ms-2, this data set does not qualify for an estimation of the axial bias.

Station Inclination Azimuth Toolface gx gy gz bx by bz


1 64.0 30.0 207.0 4.006 7.864 4.3048 11940 37980 31870
2 66.0 32.0 112.0 -8.317 3.360 3.9941 -40120 8660 30263
3 68.0 35.0 287.0 8.707 -2.662 3.6786 42010 -4920 28479
4 70.0 38.0 142.0 -5.681 7.271 3.3586 -32710 28670 26623
5 72.0 41.0 176.0 -0.651 9.316 3.0345 -11690 43060 24697
6 74.0 44.0 256.0 9.159 2.283 2.7067 41180 19710 22706
7 76.0 47.0 36.0 -5.600 -7.708 2.3756 -19000 -42580 20652
8 78.0 50.0 303.0 8.055 -5.231 2.0416 44430 -16800 18541
9 80.0 53.0 54.0 -7.823 -5.684 1.7052 -31930 -36230 16377
10 82.0 55.0 92.0 -9.718 0.339 1.3666 -48070 -9140 14353
11 84.0 58.0 153.0 -4.433 8.701 1.0264 -31880 37910 12106
12 86.0 62.0 283.0 9.545 -2.203 0.6850 50080 390 9618
13 84.0 66.0 355.0 0.851 -9.729 1.0264 16230 -47190 10489
14 85.0 79.0 257.0 9.531 2.200 0.8558 44680 23610 6803
15 85.0 80.0 217.0 5.887 7.812 0.8558 19030 46850 6577

Table B-1: Input data for demonstration case. Inclination, azimuth and high side toolface are in degrees, while the units for accelerometer and
2
magnetometer measurements are respectively m/s and nanoTesla. Sensor readings are approximate values.

Accelerometers
Parameter Requirement (calculated values) OK?
Max (Δα i ) = 52 °
General n = 15 Yes
3e = 12
15
1
bias gx, bias gy
15
∑ gzi = 2.2 ms -2 Yes
i =1
15
1
scale gx, scale gy
15
∑ gzi = 2.2 ms -2 Yes
i =1
15
1
bias gz 15
∑ gzi = 2.2 ms -2
No
i =1

{max ( gzi ) − min ( gzi )} = 3.62 ms -2


Table B-2: Calculated values for the accelerometer specific requirements given in Table 1. The axial bias is not allowed to estimate since the
2
average value of the gz-measurements is less than 2.5 m/s .
10 SPE/IADC 125677

In a real case the correct procedure would be to correct the sensor readings for the effects of estimated error terms and carry
out the noise tests shown in the two last rows of Table 1. If for example the results of this noise test turned out to be that four
survey stations were rejected, the process for selection of systematic error terms would have to be repeated. Since in this case
only 11 stations were available for further analysis, one of the four cross-axial error terms could not have been estimated.

Magnetometers
The requirements for magnetometer measurements are given in Table B-3. Since the toolface and dip angle criterion is
satisfied, the cross-axial bias errors (bias bx, bias by) can be estimated. The number of survey stations are 15, which means
that totally five systematic error terms can be estimated since 3e = 15 ≤ 15. However, if only 14 stations were available, only
four systematic error terms could have been estimated.
Since the arithmetic mean of the axial magnetometer measurements is 18767.9 nT, which is less than 0.7B = 35700 nT,
the cross-axial scale error terms (scale bx, scale by) can be estimated. However, this is not the case for the axial bias. As is
shown in Table B-3 the spread of the axial magnetometer measurements is 25293 nT, and that is sufficient, but since the
sinIsinA criterion is not satisfied, an estimation of this error term is not allowed.

Magnetometers
Parameter Requirement (calculated values) OK?
max (Δα i ) = 52 °
General (toolface variation ,
dip angle, number of survey Ө=75° Yes
stations) n = 15
3e = 12
bias bx, bias by Yes
15
1
bias bz
15
∑ sin I i sin Αi = 0.73 Yes
i =1
15
1
scale bx, scale by
15
∑ bzi = 18676.9 nT Yes
i =1
15
1
15
∑ sin I i sin Αi = 0.73
i =1
scale bx, scale by, bias bz 15
Yes
1
15
∑ bzi = 18676.9 nT
i =1

{max (bz i ) − min (bz i )} = 25293 nT


15
scale bz 1 No
15
∑ sin I i sin Αi = 0.73
i =1

Table B-3: Calculated values for the magnetometer requirements presented in Table 2. The demonstration is performed only for the enhanced
magnetic referencing.

Appendix C

The method of least squares


Multi-Station Analysis (MSA) techniques are based on the method of least squares. In this appendix, a brief explanation of
the method will be given. Matrices and vectors will denoted by bold fonts.
The set of equations that form the basis for estimation of directional parameters and systematic error terms can be
represented by the following regression model:

⎡β⎤
E( y ) = y − e = [X Z ] ⎢ ⎥ Eq. (C-1)
⎣∇ ⎦

where E(y) is the expectation values of the observations y, β is a vector of unknown directional parameters; magnetic
azimuth, inclination and high side toolface, ∇ is a vector of unknown systematic error terms, X is a known coefficient matrix,
Z is a known coefficient matrix corresponding to the systematic error terms∇. The errors e are assumed to have a zero mean
multivariate normal distribution; e ~ N(0, Cee). Equation (C-1) is often called observation equations. The matrices X and Z
define the mathematical relationship between sensor measurements and unknown parameters.
In connection with MSA the functional relationship between E(y) and [β ∇]T is nonlinear, so instead of Eq. (C-1) we get
the inconsistent system of nonlinear equations:

y − e = h( [β ∇ ]T ) Eq. (C-2)
SPE/IADC 125677 11

where h is a vector of nonlinear functions of the unknowns [β ∇]T.


Least squares estimation problems based on systems of non-linear equations can by solved using Gauss-Newton’s
method. A brief explanation of Gauss-Newton’s method will be given in the following. A more detailed explanation of the
method can be found in Koch (1999).
Using the first two terms of a Taylor expansion of Eq. (C-2) to form a linear approximation about the initial values
[β0 ∇0 ]T of the unknown parameters [β ∇]T gives:
⎡ dβ ⎤
y − e ≈ h(β0 + dβ , ∇ 0 + d∇ ) = h( β0 ,∇ 0 ) + [X Z ] ⎢ ⎥ Eq. (C-3)
⎣ d∇ ⎦

where [ X Z ] now denotes a matrix of partial derivatives with respect to the elements in [β ∇]T . The least squares estimator
[dβˆ d∇ˆ ] of the unknown corrections [dβ d∇] is obtained by solving Eq. (C-3) subject to the constraint that the weighted
T T

T −1
residual sum of squares e Cee e is minimum. This will lead to the following equations for the unknown corrections:

⎡dβˆ ⎤
⎥ = ( [X Z ] C ee [X Z ] ) [X Z ] C ee f
T −1 −1 T −1
⎢ Eq. (C-4)
ˆ ⎥
⎢⎣d∇ ⎦

where the i'th element in the vector f can be considered as the difference between the actual sensor measurement and the
approximate measurement, expressed by:

f i = yi − hi ( [β 1 , β 2 ,..., β u ∇1 ,∇ 2 ,...∇ r ]T )0 Eq. (C-5)

where u and r are the number of unknown directional parameters and systematic error terms respectively.
The estimator βˆ ∇ [ ]
ˆ T is obtained iteratively by:

[βˆ ∇ˆ ] T
n +1 [ ] + [dβˆ d∇ˆ ] , n ∈{ 0,..., k}
= βˆ ∇
ˆ T
n
T
n
Eq. (C-6)

where k is the number of iterations needed to make the elements in dβˆ d∇


ˆ [ ] T
n
smaller than some preset amount close to zero.

The covariance matrix C β̂∇ˆ of βˆ ∇


ˆ [ ] T
is estimated by:

C βˆ ∇ˆ = ( [X Z ] C ee−1 [X Z ] )−1
T
Eq. (C-7)

The diagonal elements of C β̂∇ˆ are the standard deviations of the least squares estimates βˆ ∇
ˆ [ ] T
. Note that the covariance
matrix obtained from Eq. (C-7) does not take the estimated survey noise into account. In order to do so, the covariance
matrix C β̂∇ˆ has to be scaled by the estimated survey noise variance (variance of unit weight), see e.g. Koch (1999) for further
details.
12 SPE/IADC 125677

Appendix D

User: selects error


Raw MWD Error
model that data will be
data1 models
evaluated against

NO
Program: does data meet min.
station number, min/max dip angle
and toolface spread tests?

YES

Program: evaluates data against


minimum requirements specific to
the selected model

List of
parameters that Raw data
can be solved minus failed
for is output stations

User: selects the


parameters to be
evaluated
User: Data not qualified for
MSA. Revert to single-station
analysis Program: evaluates selected
parameters and applies single
station noise test

YES
Program: have any
stations been rejected?

NO

Program: applies overall noise


NO
test

NO
Program: is the overall
noise level acceptable?

YES

Corrected inclinations
Error estimates for
Program: applies all valid and azimuths
the selected
corrections Station QC Pass/Fail
parameters
status

User: NO User: does log of surveys


determines what
passing QC meet
corrective action
user’s max
is necessary, if
course length policy?
any2

YES

User: rejects data or re-runs User: assigns selected model


User: takes corrective action
MSA against less stringent model to survey log

Notes:
1. Independent of whether or not data pass single station QC tests
2. May relate to the current log, e.g. re-run MSA to solve for reference field, or may relate to the acquisition of
subsequent surveys, e.g. trip BHA to replace MWD tool or adjust non magnetic spacing.

Figure D-1: The various steps involved in the application of the MSA-requirements.

You might also like