You are on page 1of 15

SPE/IADC 140192

Improving the Quality of Ellipse of Uncertainty Calculations in Gyro Surveys


to Reduce the Risk of Hazardous Events like Blowouts or Missing Potential
Production through Incorrect Wellbore Placement

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
Roger Ekseth, SPE, Gyrodata; John Weston, SPE, Gyrodata; Adrian Ledroz, SPE, Gyrodata; Barry Smart, SPE,
Gyrodata; Astrid Ekseth, SPE, Gyrodata

Copyright 2011, SPE/IADC Drilling Conference and Exhibition

This paper was prepared for presentation at the SPE/IADC Drilling Conference and Exhibition held in Amsterdam, The Netherlands, 1–3 March 2011.

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
Wellbore position uncertainty estimation (error modeling) became a reality. The whip-stock had to be mounted and
is a key element in safe and cost effective drilling. The oriented downhole, a process that required the
quality of error modeling has improved a lot over the last measurement and display of its downhole horizontal
10 – 15 years through the publication of SPE papers like orientation (azimuth); and the vertical orientation
SPE-67616, 90408, 103734 and 105558. These papers (inclination) had to be monitored to ensure correct tuning
taken together form a good framework for error modeling of weight-on-bit relative to the desired build/drop rates of
calculations with one exception. None of these papers the well trajectory. As a result of these demands, various
supplies the drilling industry with realistic publicly instruments that utilize the Earth’s gravity, magnetic and
documented uncertainty estimates for gyroscopic survey spin fields to determine the inclination and azimuth at
tools. It has been left to the gyro service providers given locations within the wellbore were developed, and
themselves to supply the rest of the industry with the the directional surveying industry became a reality.
necessary gyro model inputs, mainly without any external The main objective of the directional surveying
review/audit process. This is not an ideal situation, and industry has, over the years, changed from providing the
represents a potential safety hazard. The gyro service drillers with inclination and azimuth information to
providers do not only compete on price, but also on survey supplying both drillers and reservoir scientists with 3D
accuracy. This paper is an attempt to start a process to wellbore position data (north, east and vertical
close this safety gap by presenting the actual derivation of coordinates). This evolution has been driven by the simple
a brand new set of realistic uncertainty estimates for some fact that directional surveying instruments combined with
of the existing gyroscopic tools based on real downhole pipe tally or wireline depth constitute a downhole 3D
data and guidelines given in the above mentioned SPE- navigational system. Further, it would have been
papers. impossible to satisfy the operator’s resource and profit
driven demand for more and more complicated wells and
Introduction well clusters, without the availability of high quality
The complexity of oil and gas well drilling has increased wellbore position.
significantly over the last 100 years. Basic shallow vertical Real time 3D orientation and positional data of both
wells drilled in widely spaced grids have to a large extent the drill-bit and drilled sections of the well under
been replaced by complicated wells such as extended construction is needed to steer the bit safely towards the
reach, horizontal and multilateral wells, and so-called drilling target, and to avoid collisions with adjacent wells;
designer wells, which are often drilled out of closely access to 3D positional data for such wells is of course
placed template slots. This evolution would not have been vital for collision avoidance. Losing a target or drilling
possible without a broad range of technological into an existing dry well might lead to large financial
improvements including the introduction of totally new penalties, while penetrating the reservoir “too early” or
disciplines such as directional surveying. drilling into a producer well, might result in blowouts,
Directional surveying became a necessary drilling aid with consequent environmental pollution and loss of
as soon as the drilling of deviated wells controlled through human life. Safe and cost effective drilling is therefore not
whip-stocks and depth dependent weight-on-bit programs only reliant on access to wellbore positional data, but also
2 SPE/IADC 140192

on the quality of these data. Directional data including For gyro surveying, in which direction is determined
depth measurements proved early on to be of variable with respect to the Earth’s rotation vector, the situation is
quality. Experience showed that the success rate of hitting quite the opposite. State of the art gyro instruments are far
small size targets and avoiding wellbore collisions less accurate than the stability of the Earth’s rotation rate.
depended on the choice of survey type and the level of Tool design and operational procedures vary significantly
quality control applied. Tailoring of survey programs between the different suppliers, and survey accuracy is
based on cost, benefit and risk considerations became regarded as a competitive advantage. For these reasons,
more and more common, and the need for tools to quantify the SPE WPTS decided to provide the drilling industry
and predict surveying uncertainties associated with the with a framework for mathematical error modeling of
different service types became obvious, and led to the gyroscopic survey tools4, and not to supply numerical

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
formation of a new scientific discipline, ”Wellbore error parameter inputs. It was left to the gyro service
position uncertainty estimation”, which is frequently providers themselves to supply the rest of the industry
referred to as “Error modeling”. with the necessary gyro model inputs, mainly without any
As a result of major contributions to the subject such external industry accepted guidelines or review/audit
as Wolff and deWardt’s systematic error model concept1 processes in place. Historically, many examples have
and John Thorogood’s instrument performance models2, arisen of service providers of all kinds, compelled to
the quality of the wellbore position uncertainty estimation compete on accuracy, as well as price, coming under
has, over the years, increased significantly, and today is a pressure to skew their performance claims so as to present
key element in safe and economical drilling. State of the an overly optimistic view of the service that they can
art error modeling is to a large extent based on methods provide. Numerical error model inputs supplied to the
and theories originating from collaborative work over the industry from the different gyro survey providers may
last 10 – 15 years within the SPE WPTS (formerly the therefore be too optimistic and give an incorrect accuracy
ISCWSA), which has been made available to the rating. This is not an ideal situation, and represents a
directional surveying industry through the publication of a potential safety hazard.
series of SPE papers; SPE 67616, 90408, 103734 and Operators and directional drillers might believe they
1055583-6. The first paper by Williamson3 sets out a are operating within acceptable safety margins while, in
mathematical framework for error modeling in connection reality, they may operate with critically low safety
with magnetic surveying, and recommends numerical margins. This paper is an attempt to start a process to close
uncertainty inputs to be used for basic MWD surveying. this potential safety gap by presenting the derivation of a
The second paper by Torkildsen et al4 is the counterpart brand new set of realistic uncertainty estimates for some
for gyro surveying, but this gives only the mathematical existing gyroscopic tools based on statistical analyses of
framework for gyro error modeling. The WPTS has real downhole data.
formed an Error Models Maintenance sub-committee to
perform regular updates and maintenance of the MWD Layout and Scope of the Study
and gyro error models. It has been operational for some In order to provide a broad framework for the modeling of
years and therefore, the current recommendations differ gyro errors that can be adapted to the range of gyro survey
slightly from what is presented in the two error model systems that are available to the industry, the SPE WPTS
papers. The two last papers by Ekseth et al5,6 give gyro error model4 allows a large number of error terms to
recommendations regarding the quality control of different be specified. For any given gyro survey, only a limited
types of directional surveying that should be implemented number of the available error terms will be applicable,
to satisfy the demand for gross error free data, which is depending on factors such as the type of sensors used and
taken for granted in the MWD and gyro error model the chosen sensor configuration. Some terms may be
papers. inappropriate for the running gear used or the running
MWD makes use of the Earth’s magnetic field as a procedures adopted, while others may be insignificant in
reference in its azimuth calculations. The sensor errors the anticipated surveying environment. Furthermore, the
present in modern MWD instruments are small in error model theory is based on the assumption that all
comparison to typical variations in the Earth’s magnetic survey data are, or will be, free from gross errors. For the
field, which constitutes a rather unstable reference. The purposes of this analysis, and the related studies, a
MWD positional uncertainty is therefore, to a large extent, significance/confidence level of ±3σ has been chosen.
a result of environmental errors originating from natural Therefore, errors which exceed the ±3σ tolerance level
fluctuations in the magnetic field and distortions caused by defined by the error model (ERM) are treated as gross
adjacent magnetic material, and not variable sensor errors and the quality control (QC) system in use must be
quality. Knowledge and techniques on how to minimize capable of identifying surveys where at least one
the effects of the unstable magnetic reference have been significant error source lies outside the ± 3σ interval.
known for many years, and the practical implementation The creation of dedicated error models for the
does not vary significantly between the major MWD different services delivered by a gyro service provider
suppliers, who have focused on factors other than survey should start with a detailed examination of each of the
accuracy in their marketing. It was therefore possible to individual services offered. The main objective of this
create an MWD error model3 which has been accepted as examination is to establish which error terms are
an industry standard. significant and therefore need to be included in an error
SPE/IADC 140192 3

model, and whether the degree of significance may vary the tool. Both alignment options, centralized and
with different sensor packages, correction methods and decentralized, are used for casing surveys. All services are
factors including alignment equipment, platform types, offered on land rigs, platforms and semi-submersibles.
depth systems, running gear and operational procedures. Misalignment related to casing inside conductor, drill-pipe
The examination should also determine which QC inside casing and tubing inside conductors etc. requires
alternatives exist in the different cases. Gyrodata Inc., one special consideration. These misalignments, while
of the major gyro survey companies (referred to as “the important for planning and collision avoidance near the
company” throughout the rest of this paper), offers a range surface, have little or no contribution to the final bottom
of survey services which, among others, include the hole wellbore position and have not been analyzed in this
following: work.

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
Following the recommendations for quality control
• Stationary surveys given in SPE 1037345, the company has adopted a policy
- Multistation gyrocompassing on wireline of implementing Multistation Corrections and Continuous
- Multistation gyrocompassing with battery tools Drift Corrections as appropriate. It is also required to run
dropped into the well Roll Tests in the field (before entering and after leaving
• Continuous surveys the hole) and at the base warehouse (before leaving and
- Continuous survey on wireline in deviated wells after returning to the base). The application or absence of
- Continuous survey on wireline in vertical wells these correction methods influences the choice of many of
the error model parameter values.
The scope of this study is limited to the analysis of In summary, the error model that is applicable for a
data from surveys of wells, or sections of wells, where the particular gyro survey service, i.e. the relevant error terms
main objective is to establish uncertainties associated with and the values assigned to each one together with the
the mathematical reproduction of longer 3D wellbore propagation type, must be selected not only to be
(section) profiles, and does not incorporate single depth representative of the sensors used and the configuration in
angular uncertainties. The study of stand-alone angular which they operate, but also in accordance with the type of
uncertainties, usually in connection with kick-off and rig, running gear, well type and correction procedures
orientation surveys, requires a very special treatment, applied. The derivation of error parameter values for a
since running procedures and QC could be quite complex range of different gyro services is described in the Section
and are subject to many special conditions; these include 4 of this paper.
changing tools after a determined number of orientation It must be stressed that the error model data that is
shots and tying together surveys produced by different given here has been derived from survey data originating
types of tool. The error modeling of Gyro While Drilling from a limited number of services and tools produced and
(GWDTM) steering operations also requires special operated by a single gyro company. The physical design of
treatment since the level of criticality of the operation may gyro tools, operational practices, calibration and QC
well affect the running procedures and QC. As described procedures vary greatly between the different gyro
in SPE 1037345, QC of single shot gyrocompassing companies. For these reasons, the error models presented
surveys is not completely reliable and verification shots in this paper are only applicable to the actual services
are normally conducted as part of the standard procedures. described in this study. All other services or tool
The fact that the QC is not straightforward makes these configurations, including extra services offered by the
services special cases. It is the intention of the authors to study company, will generally be different, and this will
produce a further paper which will address the error model need to be reflected in the associated error models.
issues specific to these services.
The company operates instruments containing two or Statistical Analysis
three accelerometers and one or two dual-axis spinning The aim of the study was to determine the impact of all
wheel gyros of different types. This study will focus on a error sources on the survey data. An important
typical configuration containing two or three consideration was to ensure that sufficient data was
accelerometers and one of the higher performance gyros available for the diverse range of services offered by the
(gyro type 1) used in connection with high performance company, taking account of the different tool operating
wellbore positioning. Instruments referred to in the study configurations for each service and some regional
use an indexing technique which facilitates sensor bias variation in the requirements for different services; some
reduction. services are performed predominantly in one region of the
Drop surveys are taken inside drill-pipe with the tool world (e.g. vertical continuous surveys in the Middle
resting in a fixed location in the BHA, while continuous East).
surveys are performed in conductors, casings and drill- The examination of all survey jobs completed globally
pipes. Casing surveys are run predominantly with spring- over a set period of time (3 months) has allowed sufficient
bow decentralization, while drill-pipe surveys are run with data for all services and operating configurations to be
roller centralizers. The choice of centralization, along with collected. A total of 484 survey jobs have been analyzed
the tubular (drill-pipe, casing or conductor) in which the for the study described here. All data has been used with a
survey is to take place affects the alignment accuracy of view to reducing the occurrence of sampling biases.
4 SPE/IADC 140192

To ensure consistency in the data analysis across the survey services are derived. The error estimation processes
study, all surveys have been re-calculated using the most applied makes extensive use of the field calibration
recent field processing software. processes implemented by the company, outline
The statistical analysis described in the following descriptions of which are given below.
section of the paper, which has been used to obtain
estimates of the various sources of error, assumes the data Stationary surveys. The company has developed a
available for each survey service to be distributed in method for field calibration; Multistation Correction.
accordance with a zero mean Normal (Gaussian) Since it is impossible to determine accurately all of the
distribution. In practice, the data does not always conform calibration terms in the field, the goal of multistation
precisely to this assumption and alternative methods of correction is to focus on those terms that are more likely to

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
analysis have had to be adopted. The histogram shown change, namely gyro residual biases and gyro direct mass
below is typical of the distribution of gyro bias estimates unbalance. In addition, a multistation accelerometer test
derived for a particular survey service, showing a clear has been included, thus XYZ accelerometer measurements
departure from the theoretical Normal distribution that is are checked throughout the survey.
expected. Multistation correction is a very powerful tool that
updates the calibration values of the biases (GBX and
GBY), and the direct mass unbalance (M) for the gyro,
and also checks the performance of the accelerometer
package. Additionally, since multistation correction is
based on the least-squares adjustment technique, the
standard deviations of GBX, GBY and M are available
and can be considered as the residual values of these error
terms after the corrections have been applied. The
information generated by the multistation correction
process is used to determine values for the various error
terms as described in this section, each of which represents
a physical error process.

Continuous surveys. In contrast to the stationary error


Figure 1: Example distribution of gyro bias estimates
model, but in accordance with the SPE gyro survey error
In view of such issues, the error data generated for model4, the model adopted for continuous survey systems
each survey type has been subjected to ‘data snooping’, a is purely empirical; system imperfections are represented
data qualification test commonly used in land surveying by an initial azimuth error, a linear drift error and a
and geodesy, to ensure data sets are free from gross errors. random walk term which together represent the combined
Data snooping is an iterative process where the sample effects of the physical errors that may actually be present
furthest apart from the population based confidence in such a system.
interval is removed; the process is repeated until all The company has developed a method for the final
remaining members lie within the confidence interval. A ± calculation of continuous surveys called Continuous Drift
3σ confidence level has been used on apparent Gaussian Correction. It is logically equivalent to the averaging of in-
distributed data sets corrupted by outliers, while a ± 2σ run and out-run surveys and uses the knowledge of survey
confidence level has been used on data sets in which the tool orientation at departure and return. The drift
distribution was more uniform. The justification for this correction method is used to calculate estimates for the
action stems from the fact that such cases would be random walk effect and the linear drift correction for
detected as gross errors by the QC checks implemented by survey accuracy prediction from field data. This is
the company, and the survey rejected. achieved through least-squares adjustments of neighboring
The application of these procedures and details of the in-run/out-run azimuth differences using data from
error estimation process related to each survey service are multiple continuous surveys.
described in the following section. In addition, continuous drift correction is used to
check for tool misalignments. This is possible provided
that the in-run and out-run high side toolface angles are
Estimation of uncertainty parameters
The different error terms listed in the gyro error model not the same throughout the survey. The misalignment
paper4 vary greatly, and it is almost impossible to find a correction compensates for systematic misalignment for
way of determining all terms at once. However, it is the whole survey and provides additional quality control
possible to divide them into different groups based on of the data and the tool centralization.
physical and/or operational conditions, and then to utilize
different calculation methods for terms in the different Accelerometer errors
groups. In this section, accelerometer errors, gyro errors
and environmental effects are examined separately, and Stationary accelerometer measurements. The effect
error terms applicable for both stationary and continuous systematic accelerometer errors (bias, scale factor, sensor
SPE/IADC 140192 5

misalignment) have on the wellbore position uncertainty is The physics indicates that a fixed zero expectation
not only dependent on the precision of the chosen should be used in the calculation of the standard deviation,
accelerometers, but also of the actual accelerometer and the Z accelerometer bias uncertainty for a Z indexing
configuration, and if indexing of the sensors is used. tool type can be calculated using equation 1:
Three-axis systems, containing X, Y and Z
accelerometers, can measure the local gravity in addition ∑
to the inclination and toolface. The company has (1)
implemented a multistation based gravity error test in
accordance with the recommendation given in = mean gravity error from survey number i
SPE 1037345. This is purely a QC test, and the software = total number of used surveys

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
does not perform any corrections. It reports an average
From the viewpoint of wellbore position uncertainty,
gravity error and its standard deviation.
ABZ is the only accelerometer error term from the list
For a two-axis system, a gravity error test cannot be
given in the gyro error model paper4 that needs to be
conducted, and such systems will always have a singular
populated for Z indexed 3 axis systems. However, in
mathematical solution for the inclination when horizontal.
SPE 1037345, a one-to-one relationship between an error
This is the case even for instruments with Z axis indexing,
model and the associated QC is recommended. In this
which are designed to cancel the effect of X or Y biases.
situation, the tolerance for the multistation gravity error
The accuracy of indexing XY accelerometer systems is
test becomes zero for certain wellbore orientations and the
therefore a function of random noise and not of
QC test will always fail. This problem can be overcome if
accelerometer biases, which will be zero. SPE 904084
a random accelerometer uncertainty term (RA) is added to
gives weighting functions for the accelerometer biases, but
the model and the test. Such an error will propagate into
not for the random noise. The predicted inclination
the inclination and gravity in accordance with a weighting
uncertainty at horizontal will therefore be calculated as
function equal to the root-sum-square (RSS) of the
zero instead of infinity, which is the correct answer. This
associated ABX, ABY and ABZ weighting functions,
makes it impossible to fit Z indexing XY accelerometer
described in the SPE WPTS error model papers3, 4, but
systems directly into the current gyro error model
with propagation into the wellbore position being random
framework without adding a random component to the
rather than systematic. It should be noted that this
error budget.
approach is only valid when the same accelerometer type
A statistical analysis of the repeatability between two
is used for all three measurement axes. A random
or more consecutive calibrations conducted using data
accelerometer term derived for an indexing three
collected from sixty nine XY accelerometer tools indicates
accelerometer tool is also an excellent candidate for the
that accelerometer scale factors and accelerometer
random accelerometer uncertainty for indexing two
misalignments are relatively stable over time, and that
accelerometer tools with similar accelerometers. Since the
residual errors after calibration are usually insignificant.
gravity error test cannot be applied to these tools, any two
The standard deviation of the calibration to calibration
accelerometer error model will have to be accompanied by
scale factor stability was found to be 0.0001
an alternative quality control linked to the index based bias
(corresponding to a maximum inclination error of 0.006°)
correction.
and the XY misalignment stability to be 0.009°. Further,
A numerical input for a random accelerometer error
the bias stability was found to be 0.001 m/s2
term can be calculated as the square root of the average
(corresponding to a maximum inclination error of 0.006°).
sum square of estimated standard deviations of single
Although these figures easily can be transformed into error
accelerometer readings obtained from multiple
model inputs, it is not recommended. They are based on
surveys/wells/tools. However, the software used for the
data obtained under ideal conditions, and may be skewed
implementation of the g-total test does not report this
with respect to the more hostile downhole environment
quality control measure, but an alternative, namely the
From an error modeling point of view, it is acceptable
estimated standard deviation of the mean gravity error.
to set the accelerometer scale factor and sensor
This lumped quality control figure represents the
misalignment uncertainties to zero, and to let the biases
combined effect of the measurement noise, the
account for all discrepancies that may occur. Indexing
geometrical strength of the test and the statistical
about the Z axis, which is a company standard, cancels the
redundancy. Therefore, the contribution from the latter
X and Y biases that exceed the background noise, leaving
two effects must be removed from the lumped figure
the Z bias as the dominant contributor in any significant
before it can be used as a basis for the calculation of
mean gravity error result (absolute value greater than 3
random accelerometer uncertainty. Ideally, this should be
times its standard deviation). Therefore, the standard
done for each individual survey, but it is also acceptable to
deviation of the mean gravity error, calculated from a
make an approximation using equation 2, which focuses
population of different Z indexing tools of the same type
on the average behavior of the entire population instead of
and run in different wells with different surveyors, reflects
each individual survey.
the statistical spread in the Z accelerometer bias (ABZ) and
is an excellent candidate as an error model uncertainty
input. ∑
· (2)
6 SPE/IADC 140192

= standard deviation of the gravity error from survey number i for three accelerometer systems to only two; one multishot
= total number of surveys model, which should be used also in connection with
is a constant representing the average of the continuous initialization (discussed further in Section
combined redundancy/geometrical information contained 4.2.1), and one drop model.
in all gravity error test design matrices from the entire The same approach has been adopted for two axis
population of surveys. It can either be calculated directly tools. Since both the two and three axis tools use the same
from the design matrices involved, or with simple accelerometer type, the random noise values derived for
empirical models. For example: · / 1 as the three axis tool are applicable for both tool types. In
used in this study. This model makes use of the following accordance with the two axis accelerometer model
three important population based average proposed in the gyro error model paper4, the Z
redundancy/geometrical inputs:

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
accelerometer bias term (ABZ) is substituted with a gravity
= square root of the average number of surveyed depths error term, which should be based on the accuracy of the
for all surveys within the population gravity model that is used. A latitude error of one degree
= the average number of survey stations taken at each gives a gravity error of ca 0.0006 m/s2, while a depth error
depth for the entire population of 1000 meters (typical if the surface gravity is used
= the average population based absolute extreme throughout the survey) gives an error of 0.003 m/s2, the
correlation coefficient for the gravity tests recommended figure.
The gravity error data was divided into groups Numerical inputs for the necessary four different
dependent on survey methods (continuous, multishot or accelerometer error models are given in Table 1. It should
drop), pipe type (conductor, casing and drill-pipe), be pointed out that these models are only valid for Z
alignment method (blade centrollers, roller centrollers, indexing tools, and that they can be used regardless of
spring bow), and finally into platform types (land rigs, running gear and platform type.
jack-ups, platforms, semi-submersibles and drill-ships).
Each of these subsets, were analyzed independently to Table 1: Accelerometer errors – 2 and 3 axis
identify operational and environmental factors that have a survey tools
significant effect on the accelerometer performance, but it Survey Gravity
Survey ABZ RA
turned out to be difficult to establish any significant tool error
type (m/s2) (m/s2)
difference between the different groups. The number of type (m/s2)
available samples within a group was, in most cases, too Multishot 3 axis 0.0022 0.0012 N.A.
small, or the variation in calculated mean gravity errors
within a group too large to justify any quality rating. The Drop 3 axis 0.0019 0.0007 N.A.
large spread in the mean gravity error for the two different Multishot 2 axis N.A. 0.0012 0.003
drop survey types shown in Figure 2, is a typical example
based on the latter observation, and may indicate that the Drop 2 axis N.A. 0.0007 0.003
accelerometer performance is more dependent on non-
predictable factors such as the actions of the deck crew Continuous accelerometer measurements. A
(surveyor, wireline operator and rig crew) rather than the continuous survey is of course conducted while the survey
physical environment. tool traverses the path of the wellbore. However, such a
process uses gyrocompassing/stationary measurements to
initialize the periods of a continuous surveying and to
conduct bias updates at intervals along the well. All such
stationary updates are adjusted using the same multistation
correction procedure as adopted for conventional
gyrocompass surveys. The application of this procedure
when employed to initialize a continuous survey is
referred to as ‘advanced initialization’, and is an integral
part of the continuous surveying process. The
gyrocompassing accelerometer dependent error model
parameters used to model continuous surveys can
therefore be defined in the same way as described above
Figure 2: Gravity error tests for drop surveys conducted on for the stationary accelerometer measurements.
land rigs – mean gravity error
It is noted that continuous inclination can be obtained
using accelerometer measurements alone, or through a
Even though no significant difference was found combination of accelerometer and gyro measurements.
between the initial groups selected, the data indicated that The tool types used in this study are all of the latter type.
some correlation could be present. An iterative approach The recommended structure for the continuous error
of reuniting closely linked gravity error data sets without model given in the gyro error model paper4 is to model the
significant difference was then implemented to test this accelerometer errors with the same parameters, weighting
observation. This work resulted in a recommendation to functions and station to station propagation as would have
limit the number of stationary accelerometer error models been used in a similar stationary survey.
SPE/IADC 140192 7

Gyro errors within an east/west sector, has therefore been integrated


within the software. As a result of this procedure, the
Stationary gyro measurements (Gyrocompassing). multistation correction will not usually work in connection
For the purpose of stationary azimuth determination, with spring-bow surveys (no tool rotation), tangent section
indexed X and Y gyro measurements are the sole source of surveys and surveys of wellbores that lie predominantly in
data, regardless of system type. For the two and three axis an east or west direction. Geometrical tests are conducted
tools considered here, this is the only option since such to identify these cases, and such surveys are recalibrated
tools are equipped only with an XY gyro. with an alternative method utilizing the pre and post job
In addition to the gyro random noise, the following roll tests as described next.
important error sources for gyrocompassing with XY gyro Roll test data will be needed in addition to the
systems are specified in the gyro error model4: gyro bias,

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
downhole data to produce a result with the multistation
mass unbalance, gyro scale factor errors and gyro correction software when the geometrical tests fail. Roll
misalignments. These highly significant gyro error sources tests obtained at the drill site a short time prior to and
are measured and corrected for at least once at master immediately after a survey (field roll tests), are generally
calibration facilities for all gyro survey instruments more consistent than roll tests conducted over a larger time
operated by the company. The company in fact goes a step span, tests performed at the warehouse prior to shipment
further as part of the master calibration procedure by also and after return (base roll tests) for example. Field roll
correcting for the following less severe, but still tests are therefore preferred, but experience shows that
significant, sources of error: input axis g-sensitive error acceptable field roll tests may be difficult to achieve on
(quadrature effect), spin sensitivity and anisoelastic floating drill-rigs like semi-submersibles, and that it often
effects. is necessary to recalibrate with base roll tests when
While most of these terms remain constant, the mass operating from these types of rig.
unbalance and the X and Y gyro biases are known to Regardless of which recalibration method is used, it is
change with respect to the master calibration and may concluded from the discussion above that error models for
produce large survey errors if left uncorrected. The all tools used in this study require only the following four
company operating the tools and services analyzed here sensor dependent gyro errors from the list of options given
has developed a number of methods to control the time in the gyro error model paper4: the apparent gyro bias
dependent errors, by embedding the following processes errors (GBX and GBY), the direct mass unbalance error
into the standard operational software and practices: (M) and the random gyro noise (RG). Nevertheless, it is
necessary to populate these terms with numbers derived
• Z axis indexing independently for each of the recalibration methods. Both
• multistation correction the multistation correction and the different roll tests
provide estimates of GBX, GBY and M, their standard
• pre and post job field roll tests
deviations (σGBX, σGBY and σM), and the random noise
• pre and post job base roll tests (RG). The necessary numerical error model inputs can
therefore be derived through statistical analysis of
These correction methods can be looked upon as multistation correction results from the different surveys
simplified recalibrations of the survey tool. The Z axis and tools.
indexing is designed to remove X and Y biases on The multistation corrected/uncorrected error models
individual survey station basis. However, many years of are most simple to derive since a one to one relationship
experience has shown that part of the biases, or bias-like exists between the test output and the input values required
lumped residual errors, may not be removed through the for these model inputs. A multistation uncorrected model
indexing process, and form apparent bias errors which must not be misinterpreted as an error model that can be
tend to be systematic within a given survey and random used regardless of quality control level. As described
between surveys (independent). above, multistation corrections will not usually be applied
The company has implemented a horizontal Earth rate to surveys of east-west wells and tangent sections, or
based multistation correction algorithm to measure and where the toolface does not vary significantly without the
correct for the two apparent biases and the direct mass use of roll test data in the test. The uncorrected systematic
unbalance. The drawback with this correction method is, parameters (GBuc, Muc) are given as the zero mean
as indicated in the quality control paper SPE1037345, that standard deviation of the actual corrections given in
it is not applicable for all wellbore geometries and running equations 3 and 4, and the corrected systematic parameters
configurations. The multistation correction quality (GBco, Mco) are given as the root-sum-square of the
depends on latitude, the amount of inclination and toolface estimated standard deviation of the actual corrections
variation over the well section surveyed, and on the given by equations 5 and 6. In both cases, the random
dominant horizontal direction of the well. The dependency noise parameter (RG) is given as the root-sum-square of
is quite complicated, and it is often difficult to decide the actual random output as defined in equation 7. It must
whether a multistation result can be trusted or not without be noted that only results passing the internal correlation
some kind of numerical geometrical evaluation. An coefficient and east/west tests should be used in these
automatic test based on the actual latitude, the calculated equations.
correlation coefficients and the percentage of stations
8 SPE/IADC 140192

∑ ∑ been used in these calculations. The bias errors and the


(3) random noise are given in units of degrees/hour while the
mass unbalance terms are in degrees/hour/g.
∑ The corrected model should always be employed
(4) provided that all QC requirements are satisfied; standard
operating procedures dictate that the gyro measurements
∑ ∑
are always corrected using the multistation estimates of
(5) gyro bias and mass unbalance under such conditions. The
uncorrected values serve purely to define the corrections
applied (1σ values) following the successful

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
(6) implementation of the multistation correction QC
procedure. It is noted that gyro biases propagate randomly
for operation in a rotating survey tool, while bias

(7) propagation is systematic if the tool is not rotating. Mass
unbalance always propagates as a systematic error,
regardless of the running conditions.
= systematic X bias correction applied to survey number i
= systematic Y bias correction applied to survey number i
= systematic mass unbalance correction applied to survey Verification of downhole data. Following completion
number i of the analysis of downhole surveys presented above,
= standard deviation of systematic X bias correction applied to additional advanced initialization data became available,
survey number i
= standard deviation of systematic Y bias correction applied providing an opportunity to confirm the results given in
tosurvey number i Table 2. A further 10 sets of data for surveys run with
roller centrollers have been analysed and the results are
= standard deviation of systematic mass unbalance correction presented here in graphical form. Figure 3 shows the
applied to survey number i
= standard deviation of one gyro measurement - survey number i
corrected gyro biases and mass unbalance values obtained
= total number of surveys together with the parameter ERM, which is the RRS of the
associated angular error model values. Figure 4 shows the
After pre-examination of the data, and based on a corresponding noise values.
general knowledge on the physics of the actual gyroscopes
and the running procedures used, it has been decided to
make corrected and uncorrected parameters for each of the
following scenarios:

• Multistation correction applied to drop


gyrocompassing data.
• Advanced initialization – gyrocompass shots on
continuous surveys for tools running on roller
centrollers.
• Advanced initialization – gyrocompass shots on
continuous surveys for tools running decentralised.
• Multishot surveys run on wireline for tools on roller
centrollers.

The results obtained through the application of


equations 3 to 7 to these four data sets are given in Table Figure 3: Gyro biases and mass unbalance – verification data
2. Only results from surveys with adequate geometry have

Table 2: Gyro errors – stationary surveys with multistation correction


Survey type Running configuration GBuc Muc GBco Mco RG
Drop Rotating pipe 0.06 0.18 0.01 0.02 0.07
Advanced initialization Roller centrollers 0.07 0.13 0.02 0.02 0.06
Advanced initialization Decentralised 0.18 0.18 0.05 0.04 0.06
Multishot Roller centrollers 0.04 0.13 0.01 0.03 0.09
SPE/IADC 140192 9

roll test data from a lower quality gyroscope (type 2) to


illustrate the model degradation that can be expected when
using lower quality gyroscopes.
The ability to perform a successful field roll test is
also dependent on platform type, and a study of roll tests
conducted on a large number of different drilling rigs has
confirmed the expected variation in the results. However,
based on an appraisal of the results obtained for the
various types of drilling rig, it has been decided to limit
the roll test calculation to two groups of platforms; a

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
combined land rig/platform group and a floater group.
Base and field roll test results for both gyro types (1 and 2)
have been derived using equations 8 to 11 and are given in
Table 3.
Figure 4: Gyro noise – verification data
Table 3: Gyro errors – stationary surveys with roll
The gyro biases, mass unbalance and noise values are test correction
seen to be comparable or smaller than the 1σ error model Field roll tests
Base
values, and well within the 3σ tolerances defined by the Gyro Land Floating
roll tests
model. type rigs/platforms rigs
GBbr Mbr GBfr Mfr GBfr Mfr
Roll test analysis. The changes in estimated mass
unbalance and bias offsets between consecutive roll tests 1 0.06 0.23 0.06 0.16 0.09 0.21
are usually small, compared to the size of the estimates
themselves, indicating almost full correlation between 2 0.11 0.42 0.08 0.40 0.19 0.41
consecutive roll tests. With this in mind, equations 8 to 11
have been derived for calculation of the systematic error The quality of the data available for roll tests
model parameters for surveys that have to rely on roll test conducted on floating rigs is, as expected, generally of
inputs. lower quality and the number of such data sets available is
very limited. In practice, field roll tests would most
commonly be replaced by base roll tests for surveys
∑ , , ∑ , , conducted from floating platforms. Comparison of base
(8)
and floating rig results reveals some improvement in the
quality of the gyro bias estimates extracted from the base
roll tests.
∑ , , The gyro random noise reported by the roll test
(9)
routine is corrupted by test induced errors not present
downhole, and is therefore not suited for error modeling
purposes. However, the random noise from the different
∑ , , ∑ , , multistation corrections will also be valid as error model
(10)
inputs for the roll test based models. The standard
procedure during multi-shot gyrocompassing is to pre-scan
the gyro measurements during the indexing process, and to
∑ , , remove all measurements that exceed a standard
(11) acceptance limit obtained through historical performance.

= field pre-job roll test X bias estimate for survey number i Continuous gyro measurements. Continuous azimuth
,
, = field post-job roll test X bias estimate for survey number i readings can be obtained through the integration with
…… respect to time of the XY gyro measurements, through
= total number of field based pre-job/post-job differences integration of Z gyro measurements, or through a
, = base pre-shipment roll test X bias estimate for survey number i combination of both methods, as described in the gyro
, = base post-shipment roll test X bias estimate for survey number i error model paper4. This study covers two different
……
= total number of pre- shipment /post-shipment differences
continuous services. A standard continuous service
utilizing XY gyros for wellbore azimuth determination for
Roll test quality is generally not dependent on the type sections with sufficient inclination to avoid singularity
of survey, but on the actual type of instrument used and problems (typically 10° or more), and a vertical
the design of the indexing. The tools used in this study continuous service utilizing Z gyros for gyro toolface
have the same index design, and make use of the same measurement near the vertical where the gravity based
type of gyroscope (type 1). It was therefore decided to add high-side toolface is close to singular (typically 0 to 3°).
10 SPE/IADC 140192

The inclination range for the vertical continuous service


has been recently extended to 10° through suitable
modification of the error model.

Standard continuous. A full physical error model of a


continuous survey system based on the time integration of
XY gyro measurements, is quite complex, and the gyro
error model paper4 recommends the use of an empirical
model in which the following error contribution terms are
used: an initialization error, a linear drift error, and a

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
random walk error.
Initialization is accomplished using the Figure 5: Comparison of random walk estimates for
gyrocompassing methods for continuous systems continuous surveys in drill-pipe and casing
discussed in the previous section, and the accuracy of
initialization varies depending on the application of the While statistical analysis of in-run/out-run azimuth
advanced initialization process. Where the advanced differences from multiple continuous azimuth drift
initialization process has been implemented successfully, analyses, as discussed above, is an excellent method to
the error will be governed by the terms , and establish numerical error model inputs for basic
given in Table 2. In the absence of a successful advanced continuous surveying, it is not the case for surveys which
initialization process, the terms and , or are corrected based on the same type of individual drift
and (from Table 3) derived from pre and post job roll analysis. The statistical method is based on the assumption
tests should be used. that, on average for a population of surveys, the azimuth
As part of standard operating procedure, the company difference observed on return to the initialization point is
has implemented a drift correction algorithm for all caused solely by the empirical linear drift term in the error
continuous surveys; an enhanced version of the continuous model. However, this is not the case for a given survey,
azimuth drift test given in SPE 1055586. The algorithm where the return azimuth difference is caused by the
first reduces the effect of short period (compared to the combined effect of a number of integration errors,
length of the survey) harmonic oscillations, which from a including semi-systematic terms and errors that are not
wellbore position uncertainty point of view are without normally distributed. In contrast to the statistical error
significance. The next step is to estimate the average in- analysis that is applicable to the basic continuous survey
run/out-run azimuth drift, and to apply a correction to both error model, a drift correction involves a stand-alone
the in-run and the out-run surveys. The last step is the analysis, and can result in the over or under compensation
creation of a final survey by averaging corrected in-run of the errors that may be present.
and out-run results at the same depths. The software The company introduced a continuous drift correction
reports the standard deviation of the average amplitude of as an integral part of its standard operational practices
the short term oscillations, the average in-run/out-run drift, almost a decade ago, and it became necessary at that time
and the associated random walk. to find a way of to quantify the expected residual errors
The continuous error model parameters, after drift correction for error modeling purposes. A
(uncorrected linear drift) and (uncorrected random review of the current error model for drift corrected
walk) for an uncorrected quality controlled single direction surveys has been conducted recently. Experience has
(in-run or out-run) survey, can then be created by shown that large differences between drift corrected
statistical analysis of the reported drift and random walk continuous azimuths and independent multistation
values from multiple continuous drift corrections corrected gyrocompassing azimuths are far less frequent
according to equations 12 and 13. than predicted by the current model. The motivation for
revising the model and the method used to quantify the

error terms is outlined in the following paragraphs.
(12) A linear drift correction will not only remove the true
linear drift, but also part of the random walk drift, a fact
that was ignored in the first model derivation due to its

(13) complexity and as part of a cautious and somewhat
conservative approach. A closer theoretical examination of
= estimated linear drift from survey number i
the problem has shown that the linear drift correction will,
= estimated random walk from survey number i on average, remove as much as 75% of the random walk
= total number of surveys effect if only normal distributed integration errors are
present. This figure can be verified easily by using basic
Pre-analysis of the data has revealed that it is calculus to find the maximum error by substituting a
necessary to treat casing surveys and drill-pipe surveys bounded square-root function (same shape as a random
separately. Figure 5 shows the increased spread in the walk curve) with a straight line passing through the same
random walk estimates for casing surveys compared to end points.
drill-pipe surveys.
SPE/IADC 140192 11

Examples of typical in/out-run difference plots are The field data used in the derivation of the new
given for deviated wells in Figure 6 and for extended uncorrected model described above, indicates non-normal
reach wells in Figure 7. Survey 1 is an example of a low distributed random integration errors and/or non-linear
drift/low noise survey where most of the in-run/out-run drift like errors. Ekseth7 shows through his recursive
drift will be removed through the drift correction. Survey continuous azimuth error equations that a drift pattern
2 is a common example of a high drift survey with almost close to a 2nd order drift should also be present. A special
exponential growth in the in-out azimuth differences. Most mathematical phenomenon causes second order and linear
people will find such graphs disturbing, and probably do drifts to be lumped together, causing linear corrections to
not trust the results, not even the corrected surveys. be applied erroneously to second order drift terms. The
However, there is no reason for such a concern. The non- residual error will, according to basic calculus, be 25% or

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
linearity is caused by an inverse-sine-inclination azimuth less.
dependency in the rate integration, which will be The improvement between the uncorrected and the
recognized and accounted for by the drift correction corrected linear drift error in the original error model
process. Survey 3 is a low drift survey that is exposed to version is only 21%. This is far less than the 75%
an environment that is generating harmonic gyro indicated above. 21% and 75% are extremes, and it can
oscillations. In this case, the accuracy of the drift easily be concluded that the statistical expectation has to
correction can be severely compromised depending on the be somewhere in between. However, the selection of the
amplitude and phase of the harmonic content at the point true correction factor is not so straight forward. The
of return, and it will be necessary to expand the list of market for multiple same well continuous gyro surveys is
quality control parameters to include the average very limited, and it was impossible to find a sufficient
amplitude of any oscillatory behavior. Survey 4 is an number of cases with three or more overlapping drift
example of a survey with almost no in/out-run difference corrected surveys to calculate a reduction factor through
at the point of return, and with large differences in statistical analysis of real downhole data. Therefore, it was
between. The drift correction applied in this case will give decided to establish the factor through a simulation of the
little or no accuracy improvement. The survey is therefore process with artificially generated data.
an important reminder of the importance of having A purpose built simulator designed to generate a
tolerances on both the applied correction and the random multiple set of semi-random wells with semi-random
walk based on the uncorrected error model, even if a drift continuous surveys, was used to calculate the square root
correction has been applied. Finally, survey 5 is a of the average sum square of the linear drift and random
moderate noise/moderate drift survey which, like survey 1, walk for both the uncorrected and the corrected surveys,
removes most of the drift through the drift correction and finally to use these numbers to calculate the
process. percentage improvement to be gained by the application of
the drift correction. The results from four such
simulations, with up to 100% variation in population size,
in gross error tolerances and in random number generation
limits, are given in Table 4.

Table 4: Simulated continuous drift correction


results
Uncorrected Corrected
Improvements
std. dev. std. dev.
LD RW LD RW
LD RW
(°/hr) (°/√hr) (°/hr) (°/√hr)
Set
0.37 0.22 0.09 0.11 77% 51%
Figure 6: In/out azimuth differences for three deviated wells 1
Set
0.47 0.29 0.11 0.14 77% 51%
2
Set
0.41 0.12 0.09 0.06 77% 53%
3
Set
0.35 0.22 0.09 0.11 76% 51%
4

Table 4 shows a significant variation in both the


population based standard deviation of the linear drift
(LD) and the associated random walk (RW). This is as
expected, and proves the validity of the simulator.
However, the table also shows a more surprising fact. The
ratio between individual uncorrected and drift corrected
Figure 7: In/out azimuth differences for two extended reach continuous error model parameters seems to be constant,
wells
12 SPE/IADC 140192

and not to be affected by the magnitude and nature of the


underlying errors.

In the absence of suitable field data, it should be


acceptable to use these findings to create the new drift
corrected error model by simply down scaling the new
uncorrected model in accordance with the following
equations:

0.23 · (14)

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
0.49 · (15)
Figure 8: Vertical continuous survey 1 – northing vs. easting
The results from applying equations 12 to 15 to the
two data sets relating to continuous surveys in casing and
drill-pipe are given in Table 5.

Table 5: Error model parameter values –


continuous surveys
Survey type
Casing 0.21 0.25 0.05 0.12
Drill-pipe 0.25 0.21 0.06 0.10

It is important to notice that only results from surveys


that have passed the advanced initialization correlation Figure 9: Vertical continuous survey 2 – northing vs. easting
coefficient test have been used in the calculations leading
to Table 5. This to ensure against underestimation of the Toolface errors, for example, maybe too large for the
linear drift uncertainty by avoiding the use of data from assumption of uncorrelated inclination and azimuth errors
constant or near constant toolface surveys, that may well to remain valid, and result in a circular combined
mask some of the linear drift effects in in-run/out-run inclination/azimuth error response, in contrast to the
differences. along-hole response predicted using the empirical model.
A circular error model will of course only be valid at or
Vertical continuous. Vertical continuous, as described near to the vertical. The operational demand is for a model
previously, is a configuration in which measurements of valid up to ca 10° inclination, a demand that cannot be
tool turn rate about the Z axis, provided by the Z-axis fulfilled without additional QC to the standard error model
gyroscope, are integrated to create continuous estimates of linked tolerances used elsewhere, in the form of tolerances
the gyro toolface. This information, combined with high- on the initialization and the gyro drift.
side toolface, can be used to compute azimuth as the well Initialization is obtained using gyrocompassing
orientation deviates from true vertical, and the process is methods, the accuracy of which depends on whether
usually initialized using gyro compassing information. multistation correction methods can or cannot be applied.
As with regular continuous surveys, the SPE error Owing to the lack of geometrical variation seen in vertical
model paper4 recommends the use of an empirical azimuth and near vertical wells, multistation correction results
error model consisting of an initialization error, a linear frequently show high correlations and therefore the
drift error and an angle random walk error. However, to corrections should not be trusted. Therefore, the gyro
model well position uncertainties with these terms at or toolface initialization tolerances are selected in accordance
near to the vertical is not trivial, and it was decided to with the gyrocompassing tolerances that arise in the
develop an alternative empirical model with a circular absence of multistation correction.
horizontal position uncertainty. Examination of the available data indicates that linear
Sample plots showing north position versus east drifts and random walk, which are believed to be present,
position for two vertical continuous surveys are shown in are hard to model dependably. Figure 10 shows that there
Figures 8 and 9. In one case the well inclination increases is no clear relationship between depth (or length) of the
steadily in a south easterly direction, while the second well run and gyro toolface drift. In the particular case of gyro
remains close to vertical and follows a spiraling trajectory. toolface for vertical or near vertical wells, drift is a
It is much easier to fit surveys with such differing error function of several variables, including running speed, the
performance into a circular error model than a lateral temperature profile of the well, drill pipe characteristics
model. and the rotational speed and acceleration of the tool while
tripping in or out of the well.
SPE/IADC 140192 13

establish tool performance. Normal running procedures for


the vertical continuous tool require the collection of both
in-run and out-run data. These are usually run with the
same tool and the same running configuration. Whilst
systematic gyro errors will not be detected when
comparing in-run and out-run data, drift on the Z-axis gyro
can be checked and corrected during the run. Random
gyro errors and random errors in the inclination are
checked by comparing in-run and out-run data, and
performance values can be established.

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
Lateral errors in position (north and east errors) are
expected to increase with both the depth and the
inclination of the wellbore. An empirical model has been
adopted in which the lateral position errors grow in
Figure 10: Gyro toolface drift vs. depth. accordance with an assumed random inclination error, the
magnitude of which has been selected following a trial
The effect of the drift is the main contributor to the implementation of the CDT along with a visual inspection
positional error in this type of service. Current software of the in/out-run positional differences over a sample set
calculates and removes the overall drift; however, non- of vertical continuous surveys for varying depths of well.
linearities and their effects must be quantified. Analytical The use of random error propagation in continuous mode,
study of linear drift fused with random walk shows that or with high density data, may give rise to variable results
removal of the combined linear effect leaves a residual as a result of changes in the depth sampling frequency. To
error of up to 25% of the non-linear effect (Figure 11). overcome this concern, an apparent inclination error
Another point to take into consideration is related to the proportional to the square root of depth has been
fact that gyro toolface is an auxiliary parameter. The effect introduced. The size of the north and east errors are made
of gyro toolface errors on azimuth, and furthermore, on equal, and so give rise to a circular error probability in the
position, is not straightforward and is one of the main horizontal plane. The uncertainty value for the
reasons for choosing an alternative method to determine misalignment error propagation has been chosen to be
positional uncertainties. 0.0033°/√m, with north and east weighting functions of
cos ·√ and sin ·√
respectively. These are scaled versions of the type 3 and
type 4 misalignment weighting functions given in the gyro
error model paper4.
In general, the positional errors that arise may become
excessive if the depth interval or the inclination change
between successive gyrocompassing updates exceed 1000-
1500 meters or 3° respectively. Operational procedures for
the vertical continuous system have been selected in
accordance with these restrictions.

Survey tool misalignment


As part of the continuous drift correction process
Start depth described above, and used for the determination of drift
Figure 11: Residual gyro toolface errors after linear drift and random walk errors that may be present in continuous
correction surveys, the estimation of misalignment terms can also be
achieved by a similar least squares estimation process; the
As already mentioned, an alternative to the error in-run/out-run misalignment test described in detail in
model described in the gyro error model paper4 has been SPE 1055586. This process forms an integral part of the
established linked directly to the QC procedure that has continuous QC procedures implemented by the company.
been devised specifically for the vertical continuous The procedure relies on the availability of both in-run and
service. The data must of course be free from gross errors, out-run survey data for the entire survey section, and
as with any other error model. Gyrocompassing stations makes use of the in-run/out-run inclination differences
are checked for gross errors using single station tests; taken at the same depths as the fundamental QC
gravity test for the accelerometers and Earth rate test for measurements. It is possible to derive error model inputs
the gyros. Once the initialization data has been classified through statistical analyses of such QC results obtained
as being free from gross errors, the continuous data must from a large number of surveys originating from different
be examined. sources and locations. However, it is important to
The coordinate different test (CDT) introduced in SPE remember that the accuracy of most of these methods is
1055586 can be used to detect gross errors and also to dependent on wellbore geometry and/or operational
conditions.
14 SPE/IADC 140192

The continuous misalignment parameters, MX and Error model example: for a continuous wireline
MY, can be determined through statistical analysis of the survey of a deviated well in drill-pipe, the trajectory of
reported misalignment values reported from multiple which is not predominantly in an east-west direction, the
continuous drift corrections according to the following following sets of error terms should be chosen to form the
equation: appropriate error model: A1 + B2 + C2 + D2. It is noted
∑ ∑ that gyro error terms from both B and C should be used in
(16) this case; B2 denotes the error terms needed to obtain the
correct initialization and C2 contains the errors for
, = x and y misalignment estimates for survey number i continuous parts of the planned survey.
= total number of surveys
These error model inputs are derived from survey data

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
The results from applying equation 16 to the two data originating from a limited number of services and tools
sets relating to continuous surveys in casing and drill-pipe produced and operated by a single gyro survey company.
are given in Table 6. The in-run/out-run misalignment test As pointed out earlier in the paper, the physical design of
is not suited in connection with multishot and drop gyro tools, operational practices and QC procedures vary
surveys. The specification of misalignments based on real greatly between the different gyro companies, and the
downhole data is difficult for such surveys at the present error models given are therefore limited to the services
time, but should, as a result of newly completed and described here. All other services will generally be
on-going special projects, be possible within one to two different. For example, the simple step of removing the
years time. index mechanism, without other changes, will lead to quite
different error models.
Table 6: Misalignment errors
A. Accelerometer errors
Misalignment (X/Y) Bias Random error Gravity
Survey type Restrictions/
(°) (Z) (XYZ) error
comments
Casing survey 0.06 (m/s2) (m/s2) (m/s2)
3 axis tool
Drill-pipe survey (not drop survey) 0.04 A1 0.0022 0.0012 N.A. Wireline
Drop survey TBD services
3 axis tool
A2 0.0019 0.0007 N.A. Drop survey
Summary of error model results services
The error models described in this paper for a range of
2 axis tool
survey services are summarized in the following tables,
A3 N.A. 0.0012 0.003 Wireline
each table focusing on a particular type of error, viz.
services
A. Accelerometer errors 2 axis tool
B. Gyro errors – stationary surveys A4 N.A. 0.0007 0.003 Drop survey
C. Gyro errors – continuous surveys services
D. Misalignments
E. Vertical continuous surveys B. Gyro errors – stationary surveys
Gyro Gyro mass Gyro
A complete set of error model parameter values, Restrictions/
bias unbalance noise
applicable for a particular service type, may be formed by comments
(°/hr) (°/hr/g) (°/hr)
selecting the appropriate set of error terms from the
appropriate row in each of the tables; values from table C B1 0.01 0.02 0.07 Drop MSC-QC
should only be selected where a continuous survey error Continuous advanced
model is to be defined. The right hand column in each B2 0.02 0.02 0.06 initialization MSC-QC
table indicates where a set of error values is only Roller centralization
applicable to a particular type of service or a particular Continuous advanced
running configuration. The comments in table B further B3 0.05 0.04 0.06 initialization MSC-QC
highlight those cases in which the multistation correction Decentralised tool
QC process undertaken for stationary surveys can be
completed successfully; denoted MSC-QC. In the cases of B4 0.01 0.03 0.09 Multishot MSC-QC
east-west wells or tangent sections and where survey Non MSC
operation from a floating rig is to be undertaken, where the B5 0.06 0.16 0.09 Field roll test on land
MSC-QC process is expected to fail, it is necessary to rigs/platforms
make use of roll test QC procedures. In such cases, for Non MSC, non field
survey operations from land rigs or platforms and from B6 0.06 0.23 0.09 roll test
floating rigs, data sets B5 and B6 respectively should be Base roll tests
chosen.
SPE/IADC 140192 15

C. Gyro errors – continuous surveys The paper illustrates the procedures adopted by one
Linear gyro Random gyro survey service company for the extraction of realistic
drift walk Restrictions/comments error model data, resulting in the derivation of a new set of
(°/hr) (°/√hr) uncertainty estimates for some existing gyroscopic tools.
The effort and resource commitment needed to
C1 0.05 0.12 CDC* – wireline generate error models of the type described in this paper
should not be underestimated. Many man years of effort
C2 0.06 0.10 CDC* – drill-pipe have been expended in the generation and justification of
the error model values given here.
C3 0.25 0.21 No CDC*

Downloaded from http://onepetro.org/SPEDC/proceedings-pdf/11DC/All-11DC/SPE-140192-MS/1690139/spe-140192-ms.pdf by University of Trinidad & Tobago user on 07 May 2021
* continuous drift correction Acknowledgements
The authors wish to thank Gyrodata for permission to
D. Misalignment publish this paper.

Misalignment (X/Y) References


Restrictions/comments
(°) 1. Wolff C.J.M. and de Wardt J.P. (1981) ‘‘Borehole
Position Uncertainty–Analysis of Measuring Methods
D1 0.06 Casing survey and Derivation of Systematic Error Model,’’ JPT,
December 1981.
Drill-pipe survey
D2 0.04 2. Thorogood J.L. (1990) “Instrument Performance
(not drop survey)
Models and their Application to Directional
D3 TBD Drop survey Surveying Operations”. Paper SPE 18051, SPE
Drilling Engineering 5 (4), December 1990.
3. Williamson H.S. (2000) “Accuracy Prediction for
E. Vertical continuous surveys
Directional Measurement While Drilling”. Paper SPE
Apparent inclination error 67616, SPE Drilling and Completion 15 (4),
Restrictions/comments December 2000.
(°/√m)
4. Torkildsen T. et al. (2008) “Prediction of Wellbore
E1 0.0033 Casing survey Position Accuracy When Surveyed With Gyroscopic
Tools”. Paper SPE 90408, SPE Drilling and
Completion 23 (1), March 2008.
5. Ekseth R. et al. (2006) “The Reliability Problem
Concluding remarks Related to Directional Survey Data”. Paper SPE
This paper draws attention to the vital need for 103734, IADC/SPE Asia Pacific Drilling Technology
representative and justifiable error model inputs in the Conference and Exhibition, Bangkok, Thailand, 13-15
quest for safe and reliable surveying when using November 2006.
gyroscopic tools. 6. Ekseth R. et al. (2007) “High Integrity Wellbore
Individual service companies can and should provide Surveys: Methods for Eliminating Gross Errors”.
error model data based on the statistical analysis of real Paper SPE 105558, IADC/SPE Drilling Conference,
downhole data for each type of tool and service on offer. Amsterdam, The Netherlands, 20-22 February 2007.
Simulation and theoretical analysis should be used 7. Ekseth R. (1998) “Uncertainties in connection with
wherever possible to support this process and so add the determination of wellbore positions”. ISBN 82-
credence to the statistical derivation. 471-0218-8, doctoral thesis 1998, Norwegian
Error models for a particular gyro survey service are University of Science and Technology, 1998:24 IPT-
specific to that service, being influenced not only by the rapport
choice of sensors and their configuration, but by factors
including the tool running configuration, choice of
centralisation adopted and the detailed operating/QC
procedures applied when running the tool.

You might also like