144 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO.

1, JANUARY 2016

Continuous Real-Time Measurement of Drilling

Trajectory With New State-Space Models
of Kalman Filter
Qilong Xue, Henry Leung, Fellow, IEEE, Ruihe Wang, Baolin Liu, and Yunsheng Wu

Abstract— Rotary-steerable drilling provides unique features drillstring attitude (inclination and azimuth) measurement
such as an extended reach and accurate well trajectory control. is usually carried out when the drillstring is not rotating.
These characteristics can notably increase drilling efficiency However, as drilling technology improves, continuous mea-
and safety. One of the main technical difficulties of rotary-
steerable drilling is dynamically measuring the spatial attitude surement of the well trajectory becomes increasingly impor-
at the bottom of the rotating drillstring as the drillstring rotates. tant. It is also essential in a rotary-steerable system (RSS).
We developed a strap-down measurement system, with a triaxial An RSS [5]–[7] is a mechatronics tool developed for direc-
accelerometer and triaxial magnetometers installed near the bit. tional drilling. It can drill a more economical and smoother
The inclination and azimuth can be measured in real time even borehole. Since the introduction of RSS, rotary-steerable
as the drillstring rotates. Although the magnetic system is the
norm, we can use this system to achieve continuous measurement technology has achieved notable progress in reliability and
while drilling; to achieve this, a novel state-space model is has become a standard drilling tool in many worldwide
proposed here and the Kalman filter is used to estimate the applications. The application of RSS is restricted to high-cost
states. Simulation and experiments show that a continuous-survey offshore sites and is becoming more common in cost-sensitive
system with a Kalman filter approach can improve measurement land work, especially in shale-gas and shale-oil drilling [8].
precision and reduce errors produced by drillstring vibration.
Despite popular use of the RSS, field-trial results of
Index Terms— Continuous measurement while drilling continuous measurement of inclination and azimuth have not
(MWD), directional drilling, Kalman filter, state-space model, been well documented in the literature. However, measuring
strap down.
the posture of downhole tools as the drillstring rotates is
I. I NTRODUCTION essential because of its closed-loop control structure [9], [10].
A bottom-drilling tool shows complex dynamics while rotating
D IRECTIONAL drilling technology involves directing
a wellbore along a predefined trajectory, which
dramatically reduces costs and saves time during drilling
owing to the combined effects of nonlinear vibrations, such
as vertical vibration, horizontal vibration, eddy, and sticky
operations [1]–[3]. During the last several years, more and slip [11], [12]. The effects of such vibrations cause the
more attention had been paid to the development of directional measurement sensors to generate large errors. This is a huge
well-drilling technologies. Technology for directional- challenge for signal processing [9], which is completely
drilling navigation is currently based on an integrated different from the measurement-while-drilling (MWD) survey
magnetometer and accelerometer triad [4]. To compute the systems in current oil and gas industry.
bottom-hole assembly (BHA) position, data of the earth’s Continuous MWD is studied under laboratory
magnetic field and the force of gravity are measured. The conditions using a gyroscope-based system [13]–[15].
surveying system is executed along the well trajectory at ElGizawy et al. [13], Elgizawy et al. [14], and
stationary survey stations. In drilling engineering, the bottom Jurkov et al. [15] proposed an advanced inclination and
direction sensor package based on an inertial navigation
Manuscript received April 15, 2015; revised August 3, 2015; accepted system (INS). They verified the reliability of the algorithm
August 4, 2015. Date of publication October 6, 2015; date of current version
December 7, 2015. This work was supported in part by the Fundamental through simulation, which used INS to achieve continuous
Research Funds for the Central Universities under Grant 2652015063 and in MWD with high accuracy. The influences of vibration
part by the Public Welfare Fund Project through the Ministry of Land and and temperature on MWD were also analyzed [16]–[18].
Resources under Grant 201411054. The Associate Editor coordinating the
review process was Dr. George Xiao. Yanshun et al. [19] and Chen et al. [20] conducted a
Q. Xue and B. Liu are with the Key Laboratory on Deep GeoDrilling similar study by developing an MWD instrument based
Technology, Ministry of Land and Resources, School of Engineering and on a predigested inertial measurement unit. However, they
Technology, China University of Geosciences, Beijing 100083, China (e-mail:
[email protected]). did not consider the downhole complex situations, severe
H. Leung is with the Department of Electrical and Computer Engineering, vibration, and high temperature. These are great challenges
University of Calgary, Calgary, AB T2N 1N4, Canada. for measurement of accuracy and sensors’ lifetime. Drillstring
R. Wang is with the College of Petroleum Engineering, China University
of Petroleum, Qingdao 266580, China. vibration can greatly affect the life of the gyroscope, and an
Y. Wu is with State Grid Haixing Power Supply Company, increasing temperature can cause drift error in the gyroscope.
Cangzhou 510623, China. The approaches mentioned above can increase the accuracy
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. of measurements to a certain extent. However, effort still needs
Digital Object Identifier 10.1109/TIM.2015.2479096 to be made in order to improve the performance of the MWD
0018-9456 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
XUE et al.: CONTINUOUS REAL-TIME MEASUREMENT OF DRILLING TRAJECTORY 145

Fig. 1. (a) Multimodel measurement system at the drill bit. (b) Construction of measurement system.

TABLE I x, y, and z axes, respectively. Moreover, m x , m y , and m z are

C HARACTERISTICS OF S ENSORS defined as survey signals of the triaxial magnetometers on the
x, y, and z axes, respectively. Assume that the earth’s magnetic
field strength is M. Obviously, M = (m 2x + m 2y + m 2z )1/2 .
Under a certain sample frequency (100 Hz), the measuring
signal can be considered as a time series.
Assume that the acceleration of gravity is G, and
gx , g y , and gz are defined as survey signals of gravity
acceleration on the x, y, and z axes, respectively. Then,
G = (gx2 + g 2y + gz2 )1/2 . The survey signals (ax , a y , az ) of
the triaxial accelerometers include not only the acceleration
of gravity but also the acceleration of the drillstring vibration.
Electronic instruments consisting of three-axis accelerometers
instrument. Qilong Xue et al. [9] developed a strap-down and three-axis magnetometers are shown to relate to the
multimode surveying system. They gave more consideration measured inclination and azimuth. The drillstring posture is
to the actual situation of the drilling process and used field determined via the following known equations:
test data to study the measurement algorithm. This system is ⎛ ⎞
more conducive to field applications. Moreover, improvement gx2 + g 2y
in accuracy is crucial, especially for the continuous MWD Inc = arctan ⎝ ⎠ (1)
gz
of an RSS, for precise and efficient measurement of wellbores 

drilled for oil and gas exploration. Within the signal-processing G · (gx m y − g y m x )
community, Kalman filter still remains a very active topic. This Azi = arctan  . (2)
m z gx2 + g 2y − gz (gx m x − g y m y )
paper is concerned with improving the accuracy and stability
of inclination and azimuth measurements of the accelerometer The three-axis inclination and six-axis azimuth equa-
and magnetometer. We developed new state-space models that tions [22] are used in the static MWD surveys as industry-
were applied to the Kalman filter model. The algorithm greatly standard survey methods. The azimuth direction is determined
improved the accuracy of well-trajectory measurements and in a stationary mode by using three-axis magnetometers, while
is expected to be applied to ordinary magnetic surveying the inclination and the tool face angle are determined using
systems, which are more widely used in drilling engineering. three-axis accelerometers. Drilling has to pause frequently
at surveying stations in order to allow the inclination and
II. M EASUREMENT S YSTEM azimuth to be surveyed. The well trajectory is then computed
We develop a strap-down MWD system here that incorpo- between the two surveying stations based on some mathe-
rates three-axis magnetometers and three-axis accelerometers matical assumptions; It is assumed that the drilled distance
arranged in three mutually orthogonal directions [9], [10], is a smooth arc. In (1) and (2), only the acceleration of
as shown in Fig. 1(a). The magnetic measuring tools are gravity is concerned, and the drillstring vibration signals are
installed in the interiors of nonmagnetic drill collars. These considered as noise. The determination of azimuth while
special drill collars are usually designed from monel metal drilling (while the sensors are rotating) was not demonstrated
to avoid external interference with measurements taken by owing to limitations in the capabilities of downhole process-
magnetic MWD surveying tools [3], [9]. Fig. 1(b) shows the ing. The survey signals of triaxial accelerometers include not
structure of the downhole measurement system, and Table I only the acceleration of gravity but also the acceleration of
shows the characteristics of the sensors. ax , a y , and az are drillstring vibration. Unlike the conventional MWD, the incli-
defined as survey signals of the triaxial accelerometers on the nation and azimuth should be dynamically solved when the
146 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO. 1, JANUARY 2016

Fig. 2. Energy spectrum of lateral vibration signals (left) and longitudinal vibration signals (right).

drillstring rotates. Accelerometers are hypersensitive to

vibrations, so the violent vibration of a BHA leads the mea-
sured signals to be submerged in noise signals. The continuous
MWD in this paper is based on a strap-down multimodel
surveying system [9].
According to the data of field test, we completed a Fig. 3. Solving process for the system.
frequency analysis of the vibration signals to evaluate if the
white noise hypothesis. Frequency analysis is indispensable
The measurement equation of the system is given by
when discussing the most advanced approach that involves
dynamic spectral analysis and analyzing how the spectrum z k = Hk x k + ηk (5)
of a signal evolves in time. This analysis is performed with
Gabor transforms [25], which are close relatives of wavelet where z k is the measurement vector of the system output,
transforms. Gabor transforms are particularly useful in ana- Hk is the observation or design, and ηk is the measurement
lyzing drillstring vibration signals. The transform is employed noise. The system noise wk and the measurement noise ηk
here to determine the sinusoidal frequency and phase are unassociated zero-mean white-noise processes with deter-
content of local sections of a signal as it varies with time. The mined autocovariance functions.
function is first multiplied by a Gaussian window function. The We developed an algorithm with new state-space models
resulting function is then transformed by the Fourier transform. by analyzing drillstring dynamics. In the solution system,
We use a time–frequency box to express the energy distrib- we define the input vector of KF-1 as
 
ution of the Gabor function. The time–frequency spectrum of ax a y az
the vibration signal is plotted in Fig. 2. The Gabor transform X=
mx m y mz
can help us discriminate the rotational movement pattern.
which is measured by the three-axis magnetometers and
Define G u,ξ (t) as the time window function, where u is center
three-axis accelerometers. The input vector of KF-2 is the
point, σ is the width, and the Fourier transform of G u,ξ (t) can
be expressed as inclination and azimuth, defined as M = IA , where I is
Ĝ u,ξ (ω) = Ĝ(ω − ξ )eiu(ω−ξ ) . (3) the inclination of the borehole and A is the azimuth of the
borehole. Using the solving process as shown in Fig. 3, we
As shown in Fig. 2, from the energy spectrum of develop two KFs for the entire drilling process. After KF-1,
lateral vibration signals (left) and longitudinal vibration we can obtain the more precise signals of gravity acceleration
signals (right), we find that the spectrum does not particularly gx , g y , and gz , that defined as the output of KF-1. Using
have the energy concentration point, and the signals can be the gravity accelerations, we can obtain the inclination and
considered as random. Therefore, the hypothesis of white azimuth by the equation developed when the drillstring rotates.
noise in the new Kalman filter is satisfied. KF-2 is then used to further smooth the drilling trajectory. The
output of KF-2 is defined as M  = [I  A ]T , which is more
III. N EW K ALMAN F ILTER A PPROACH
precise.
The application of KF [23], [24] requires that both the sys-
tem and the measurement models of the underlying process be
A. State-Space Model for KF-1
linear. A discrete-time linear state-space system is described by
The sensors installed at the center of the drillstring, and
x k = k,k−1 x k−1 + G k−1 wk−1 (4) the measurement signals of the x and y axes, will exhibit
where k,k−1 is the state transition matrix, x k is the state a sine wave during rotation. Theoretically, accelerometer
vector, G k−1 is the noise distribution matrix, wk−1 is the and magnetometer signals have the same rule. In the actual
process noise vector, and k is the measurement epoch. drilling process, the vibration of the drillstring affects less
XUE et al.: CONTINUOUS REAL-TIME MEASUREMENT OF DRILLING TRAJECTORY 147

on the magnetometer signals. That is, the magnetometer as follows:

signals are used to calibrate the accelerometer signals.  2  2
 2N p 2N p
From the laboratory testing, we can conclude that the TM =0 g x d T M + TM =0 g y d T M
changes of signals of fluxgate in line with signals of gravity I = tg −1  2N p (9)
acceleration, as shown in Fig. 6, will be provided in the − TM =0 gz d TM
following. A = tg −1
Assuming that the angular velocity is ωx,y,z and the   2N p 
G(m y gx − m x g y )d TM
TM =0
sampling interval is t, then ×  2N p   2N p
TM =0 m z g x +g y d TM − TM =0 gz (m x g x + m y g y )d TM
2 2
m x,y (k) M sin ωx,y t
= (10)
m x,y (k − 1) M sin ωx,y (t − t)
G sin ωx,y t g x,y (k) where TM is the magnetic toolface angle: TM =
= = . (6) tg −1 (−m y /m x ). Toolface angle used for near-vertical wells.
G sin ωx,y (t − t) gx,y (k − 1)
Magnetic toolface is the angle, or azimuth, of the borehole
 g (k)  survey instrument within the wellbore measured clockwise
x
In the KF-1, we define the state vector as x g (k) = g y (k) , relative to magnetic north and in the plane perpendicular to
gz (k) the wellbore axis; the north, east, south, and west directions
as shown in Fig. 2, from the vibration signals of ax , a y , and az ,
have magnetic toolface angles of 0°, 90°, 180°, and 270°,
we can obtain the signals of gravity acceleration x g (k). Thus
a  respectively. Magnetic toolface may be corrected to reference
x
the measurement vector of the system output as z(k) = a y , either grid north or true north.
az Although the drillstring rotational speed is a way to deter-
when the drillstring rotates, the measurement signals of the mine whether the string is rotating or not, it reliability of this
x- and y-axis will exhibit a sine wave except the signal of the approach is not high. Instead, using the standard-deviation
z-axis. So the transformation matrix is defined as statistical methods to determine the drillstring movement is
⎡ m (k) ⎤ more effective because it reflects the degree of dispersion
x
0 0 among the individuals within the group. Using 50 data points
⎢ m x (k − 1) ⎥
⎢ m y (k) ⎥ as a time window, assumed them to be x 1 , 
x 2 , . . . , x 49 , x 50 , we
Hk = ⎢ 0⎥ .
⎣ 0
m (k − 1) ⎦ get the standard deviation σ = ((1/N) i=1 N
(x i − x̄)2 )1/2 .
y
0 0 1 The drillstring can be considered static when the standard
deviation σ is close to zero.
The system noise wk and the measurement noise ηk are
uncorrelated zero-mean white-noise processes. Therefore, we
obtain the state-space model of the KF-1 as follows: C. State-Space Model for KF-2
⎡ m (k) ⎤ In the drilling process, the movement states that σ = 0 or
x
0 0 σ = 0 will appear alternately. When σ = 0, the results are
⎢ m x (k − 1) ⎥
⎢ m y (k) ⎥ more accurate because the vibration is slight when the drill-
x g (k) = ⎢ 0⎥ x g (k − 1)
⎣ 0
m (k − 1) ⎦ string is not rotating. We develop another Kalman filter (KF-2),
y
as shown in Fig. 2, to smooth the trajectory of the drilling.
0 0 1
⎡ ⎤ (KF-2 was divided into KF-2.1 and KF-2.2, respectively, as
wx (k − 1) shown in Fig. 5.)
+ ⎣ w y (k − 1) ⎦ (7) In a normal drilling process, the drilling operation has to
wz (k − 1) frequently stop at measurement stations in order to measure the
z(k) = Hk x g (k) + η(k). (8) inclination and azimuth. The well trajectory is then computed
between the two surveying stations based on mathematical
assumptions. For instance, it may be assumed that the drilled
B. Calculating the Inclination and Azimuth
distance is a straight line, smooth arc, or polygonal line; each
When the drillstring rotates, the above equations are not requires a different calculating method. Assuming that the
applicable. The sensors installed at the center of the drillstring, 3-D coordinates of the measuring Nth point of the actual
and the measurement signals of the x- and y-axis, will exhibit drilling trajectory are (x N , y N , z N ), the measuring (N + 1)th
a sine wave during rotation. point is (x N+1 , y N+1 , z N+1 ), and the well depth, vertical
Through KF-1, we obtain gx , g y , and gz which are defined depth, inclination, and azimuth are L N , H N , θ N , and ψ N ,
as survey signals of gravity acceleration on the x, y, and z and L N+1 , H N+1 , θ N+1 , and ψ N+1 , respectively. The well
axes, respectively.
Then define the input vector of the system trajectory between the two points can then be defined as
 g
as X = m m m .
x g y g z follows:
x y z ⎧
Define the rotational speed as R. If R = 0, ⎪
⎪ L = L N+1 − L N
⎨
(1) and (2) are used to calculate the inclination and azimuth. H = H N+1 − H N
(11)
R = 0 indicates that the drillstring is rotating, and compu- ⎪
⎪ θ = (θ N + θ N+1 )/2
⎩
tational formulas of inclination I and azimuth A are given ψ = (ψ N + ψ N+1 )/2.
148 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO. 1, JANUARY 2016

Fig. 4. Trajectory prediction using method of inclined plane cirque.

If the vertical depth H is a known, the 3-D coordinates of

the measuring (N + 1)th point can be defined as
⎧
⎨x N+1 = x N + H tan θ cos ψ Fig. 5. Solving process for KF-2.
y N+1 = y N + H tan θ cos ψ (12)
⎩
z N+1 = z N + H.
We can obtain the spatial coordinates of each point by during t). As shown in Fig. 4, we can use the measurement
recursive calculation, thereby obtaining the entire drilling of acceleration on the z-axis to calculate the displacement.
trajectory. az is the signal of the triaxial accelerometers on the z-axis,
As shown in Fig. 4, we can use the wellbore trajectory which is combined with the gravitational acceleration and
extrapolation method to establish a recursive relationship vibration acceleration. The measurement of acceleration on
between two adjacent measurement points. Assume L is the the z-axis time series is defined as az (k).
depth of the drilling, γ is the angle of the drilling trajectory So before calculating the displacement L, we should
and its tangent. Then exclude the impact of gravity as follows:
L(k) = L(k − 1) + L(k) (13) f gz (k) = az (k) − G · cos(Ik−1 ) (18)
γ = arccos [cos(Ik−2 ) cos(Ik−1 ) + sin(Ik−2 ) where f gz (k), the acceleration time series function by remov-
× sin(Ik−1 ) cos(Ak−1 − Ak−2 )] (14) ing the acceleration of gz , can be calculated from az (k) and
γ inclination Ik−1 corresponding to the same time.
γ (k) = L(k). (15)
L(k − 1) Then, we can calculate the depth (L) of drilling using
From (13) to (15), we can use two points to estimate the acceleration on the z-axis time series f gz (k). Define the state
 L(k) 
next point, so the well trajectory can be smoothed. As shown in
vector as L(k) =  L̇(k) , the measurement vector of
Fig. 5, we can calibrate the drilling trajectory using Kalman  L̈(k)
filter 2.2, wherein the kth measurement point is defined as the system output as z(k) = f gz (k), we develop a state-space
P(k) = [Ik Ak ]. The system inputs are P(k-2) and P(k-1), model for KF-2.1 as follows:
KF-2.2 estimated P(K) combined with measured values and ⎡ ⎤ ⎡ 1 2⎤ ⎡ ⎤ ⎡ ⎤
the theoretical calculation values. L(k) 1 t t L(k − 1) ε(k)
⎣  L̇(k) ⎦ = ⎣ ⎢ 2 ⎥ ⎣ ⎦ ⎣ ⎦ (19)
KF-2.2 can smooth the well trajectory using inclination 0 1 t ⎦  L̇(k − 1) + ε̇(k)
and azimuth  L̈(k)  L̈(k − 1) ε̈(k)
as inputs. Assume that the state vector is 0 0 1
x(k) = A I k and the measurement vector of the system ⎡ ⎤
k
 ! L(k)
f gz (k) = 0 0 1 ⎣  L̇(k) ⎦ + η(k). (20)
output is y(k) = AIk , which is calculated from (8) and (9).
k m
  L̈(k)
The transformation matrix is defined as H (k) = 10 01 . By preprocessing the measured signals, a dynamic azimuth-
According to (12)–(14), the state-space model for KF-2.2 can and inclination-solving algorithm is established based on
be obtained in (16) and (17), as shown at the bottom of this dynamic analysis of the bottom rotating drilling tool. Based
page. on the theoretical model, we develop a Kalman filter to
As shown above, we should determine L as the improve the solver accuracy as well; the state-space equations
KF-2.2 input (the distance that the drill bit moves forward used for the Kalman filter have been established based on

⎡  ⎤
  sin(γ + γ (k))  
⎢ arccos cos(I k−2 ) cos(γ + γ (k)) − (cos(I k−2 ) cos γ − cos(I k−1 )) ⎥
Ik
=⎢ sin(γ
 )  ⎥ + w I (k) (16)
Ak ⎣ cos(γ (k)) − cos(Ik−1 ) cos(Ik ) ⎦ w A (k)
Ak−1 + (Ak−1 − Ak−2 ) · arccos
sin(Ik−1 ) sin(Ik )
y(k) = H (k)x(k) + v(k) (17)
XUE et al.: CONTINUOUS REAL-TIME MEASUREMENT OF DRILLING TRAJECTORY 149

Fig. 6. (a) Laboratory bench. (b) 3-D simulation model of the test bench. (c) Experimental equipment at inclinations of −5°, −2°, 0°, 2°, and 5°
(shown by the red line).

the drilling trajectory predicted. The dynamic measurement When the drillstring continuously rotates with a con-
algorithm with the Kalman filter is a new model that can stant speed and is located at a particular borehole inclina-
greatly reduce the solution errors. tion and azimuth, accelerometer (x, y, z axes) and fluxgate
(x, y, z axes) measurement data are obtained as shown in
IV. E XPERIMENTAL R ESULTS Fig. 7(b) and (c), respectively.
We put the entire system on a laboratory bench [Fig. 6(a)]
Dynamic measurement algorithms developed in this paper
to conduct an accuracy test of the measurement system.
were tested through laboratory bench and field measurement
Fig. 7(b) and (c) shows the measurement data for the mag-
data, respectively.
netometers (x, y, z axes) and accelerometers (x, y, z axes).
Here, we can calculate the signals of the magnetometers
A. Laboratory Testing and accelerometers if we know the inclination according
The combined measurement system was tested first in a to (9) and (10). Assumed that the reference of the measure-
laboratory environment, as shown in Fig. 6. Measurement data ments data is the reverse theoretical calculation results, the
were obtained under different inclination and rotating speed relative errors of measurement data as shown in Fig. 8, it can
conditions. In the experimental system, we used the encoder be seen that the accelerometer noises are relatively much larger
to measure the drillstring rotational speed and positioned the than the magnetometer signal noises, the peaks in Fig. 8 show
inclination mechanically. that the measurement error increases when drillstring appear
As shown in Fig. 6(b), the power simulation motor (with stick-slip vibration. The main reason is that accelerometers
digital 1 representation), will work to provide electricity, are hypersensitive to drillstring vibrations. In the field test,
although this time the drillstring not continuously rotates, the the vibration of the drillstring is more violent. Using the
power simulation motor will generate vibration making the sta- components of gravity acceleration on the x, y, and z axes can
tic solver results as above are not very accurate. The platform help us calculate the inclination and then combine the fluxgate
rotating motor (with digital 7 representation) working makes measurement signals to obtain the azimuth [Fig. 7(d)].
the drillstring continuously rotates, we could use (1) and (2) In engineering applications, the inclination error of 0.1
to dynamically solve the inclination and azimuth. In the will suffice. As shown in Fig. 6(c), when the experimental
experiment, we set the laboratory test at different inclinations equipment is at inclinations of −5°, −2°, 0°, 2°, and 5°,
and different rotational speeds, as shown in Fig. 6(c). the test results are shown in Table II. Through laboratory tests,
150 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO. 1, JANUARY 2016

Fig. 7. Measurement system. (a) Construction of measurement system. (b) Laboratory survey signals of triaxial magnetometers on the x, y, and z axes.
(c) Laboratory survey signals of triaxial accelerometers on the x, y, and z axes. (d) Continuous survey method, gx , g y , and gz are the gravity accelerations
on the x, y, and z axes, in (8) and (9), we are only concerned about the acceleration of gravity.

Fig. 8. Relative errors of accelerometers and magnetometers measurement Fig. 9. Schematic of field test. The electronic circuit is installed on a
data. compressive cylinder and placed in the axis of the drill collar, the sampling
frequency of measurement and the control system is 100 Hz, and the
TABLE II measurement data can be stored in real time.
E RROR OF I NCLINATION
satisfactory when the drillstring rotates. The accelerometer
signals of the field tests are completely different from those
obtained via laboratory survey. The azimuth is determined
using three-axis magnetometers, while the inclination is
determined using three-axis accelerometers. Therefore, the
drillstring vibration dramatically amplifies the error of the
continuous survey when the drillstring rotates. The algorithms
developed in this paper will solve this problem by improving
the accuracy of continuous MWD with the proposed state-
space models. Fig. 9 shows the schematic of the field test.
the theoretical models can be verified as entirely feasible when From the measurement data stored in real time, we can
the drillstring rotates. Experiments show that the dynamic obtain the vibration signals measured by the accelerometer
solving methods in this paper meet engineering requirements. (x, y, and z axes), which contain the gravity acceleration on
the x, y, and z axes.
B. Field-Drilling Testing We use the real field-test data to demonstrate the feasibility
In the field, a few methods such as a low-pass filter and of our algorithm. The results are shown in Fig. 10. During
moving average filter are considered, but the results are not the first step of the solving process for the system (Fig. 3),
XUE et al.: CONTINUOUS REAL-TIME MEASUREMENT OF DRILLING TRAJECTORY 151

Fig. 10. Signals of triaxial accelerometers on the x, y, and z axes, as shown in subparts (a), (b), and (c) respectively, before KF-1 and after KF-1.

Fig. 11. Solving results of inclination and azimuth at XuanYe 1.

the signals of the triaxial accelerometers (ax , a y , az ) going

though KF-1 get the gravity acceleration (gx , g y , gz ).
The results of KF-1 are shown in Fig. 10. The red line is the
output of KF-1, that is, the gravity acceleration. Because the
accelerometer on the z-axis is not influenced by the rotation
of the drillstring, the curve tends to be a straight line when the
inclination does not change [Fig. 10(c)]. We cannot estimate
the error of this process, because the data came from the field
test, which impossible to know the exact measurements.
In the actual drilling process, we need to find a way
to measure how good the measurement accuracy is. For
a continuous drilling trajectory, there is no method that Fig. 12. Inclination error of solving results.
could be used to accurately measure the wellbore trajectory.
So we could only use the MWD static measuring points for measurement system was deployed at multiple formations.
comparison. It should be feasible to evaluate the advantages The field tests were carried out on wells using automatic
of the dynamic measurement algorithm. Tests of more than vertical or rotary steerable drilling. Three- and four-rib
ten wells were conducted throughout China. The dynamic actuators were adopted. A lot of raw measurement data
152 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO. 1, JANUARY 2016

TABLE III
BASIC PARAMETERS OF F IELD T EST

Fig. 13. Solving result of inclination at Anshun1.

were accumulated. These data were input into our algorithm are all used for calculation, and at the same time, filtered
for authentication. Conclusions on bit movement rules were signals on the x and y axes were stored; on condition of
drawn by analyzing and summarizing large amounts of data. rotation string, real time filtered signals on the z-axis and
Two wells in China (Xuanye1 and Anshun1) are selected stored signals of the x and y axes with nonrotating string are
here to evaluate the algorithm. The results for inclination adopted. In addition, stick-slip state of the down-hole drilling
and azimuth are shown in Figs. 11 and 12. The basic para- tool is considered as a nonrotating stationary state.
meters of field tests are shown in Table III. The proposed Inclination and azimuth show great fluctuations when the
Kalman filter method can significantly improve the precision drilling string rotates. The results obtained using the method
of the survey and reduce vibration interference in the solution we developed are shown again in Figs. 11 and 12. Comparing
results. static measurement points, it can be seen that the dynamic
We can also use the periodic static measurements to improve measurement inclination error is less than 1°, and the azimuth
the dynamic solver accuracy. Dynamic solution approach to error is relatively large, between 5° and 20°, because it is
the bottom of the rotating drillstring attitude is proposed in [9]. calculated under small inclination angles. When the angle of
With nonrotating string, filtered real time signals on three axes inclination is small, accelerometer signals of G x and G y are
XUE et al.: CONTINUOUS REAL-TIME MEASUREMENT OF DRILLING TRAJECTORY 153

small as well; this leads to an increased impact of noise. The dynamic measuring algorithm proposed is analyzed
Moreover, the error accumulated in formula (2) causes the through downhole measurement instance data. The results
azimuth error to become larger. show that the algorithm can effectively reduce inclination-
1) Drilling at Xuanye1: This drilling field test was and azimuth-solving errors. Field measurement data analysis
conducted on May 19, 2010. We use the static measurement shows that the algorithm can effectively solve the dynamic
data to evaluate the continuous solving results. From inclination and azimuth when the inclination is large.
Figs. 11 and 12, we can see that the inclination error is
less than 1°, additionally, the error is gradually reduced. ACKNOWLEDGMENT
Unfortunately, we did not get the static field measurement The authors would like to thank the Drilling Technol-
data of azimuth. ogy Research Institute, Shengli Petroleum Administration of
2) Drilling at Anshun1: This drilling field test was Sinopec Corp, for providing data and materials.
conducted on January 23, 2012. We use the static measurement
data to evaluate the continuous solving results. From Fig. 13, R EFERENCES
we can see that the inclination error is less than 1°, and that
[1] S. Brzezowski and J. Fagan, “Analysis of alternate borehole survey
the azimuth error is between 5° and 20°. systems,” in Proc. 39th Annu. Meeting Inst. Navigat., Houston, TX,
USA, 1983, pp. 71–78.
V. C ONCLUSION [2] S. D. Joshi and W. Ding, “The cost benefits of horizontal drilling,” in
Proc. Amer. Gas Assoc., Arlington, VA, USA, 1991, pp. 679–684.
In the directional drilling and rotary-steerable drilling [3] W. A. Rehm, A. Garcia, and S. A. Cia, “Horizontal drilling in mature
technology, it is a challenge to measure the spatial attitude oil fields,” in Proc. SPE/IADC Drilling Conf., New Orleans, LA, USA,
(inclination, azimuth, and tool face) of the bottom-drilling Feb. 1989, pp. 755–764.
[4] J. L. Thorogood and D. R. Knott, “Surveying techniques with a solid-
tool accurately in real time while the drillstring is rotating. state magnetic multishot device,” SPE Drilling Eng., vol. 5, no. 3,
The drillstring vibration seriously affects the accuracy of pp. 209–214, Sep. 1990.
the dynamic solution. In drilling engineering, the bottom [5] T. Warren, “Rotary steerable technology conclusion: Implementation
issues concern operators,” Oil Gas J., vol. 96, no. 12, pp. 23–24,
drillstring attitude is usually measured when the drillstring Dec. 1998.
does not rotate. However, with recent progress in drilling [6] Y. Tetsuo et al., “Robotic controlled drilling: A new rotary steerable
technology, continuous measurement of the well trajectory drilling system for the oil and gas industry,” in Proc. IADC/SPE Drilling
Conf., Dallas, TX, USA, 2002, pp. 744–758.
has become increasingly important. It also becomes necessary [7] S. Poli, F. Donati, J. Oppelt, and D. Ragnitz, “Advanced tools for
in the RSS and automatic vertical drilling system. In recent advanced wells: Rotary closed loop drilling system—Results of proto-
years, the requirement for a continuous survey that captures type field testing,” SPE Drilling Completion, vol. 13, no. 2, pp. 67–72,
Jun. 1998.
the actual trajectory between stationary surveying stations has [8] J. Sugiura, A. Bowler, R. Hawkins, S. Jones, and P. Hornblower,
become an urgent need. This allows for a better estimation of “Downhole steering automation and new survey measurement method
the casing pipe and cementing of the borehole. In addition, significantly improves high-dogleg rotary-steerable system perfor-
mance,” in Proc. SPE Annu. Tech. Conf. Exhibit., New Orleans,
this provides an actual estimation of the curvature along the LA, USA, 2013, pp. 1–10.
well trajectory. Therefore, in this paper, the trajectory between [9] Q. Xue, R. Wang, F. Sun, L. Huang, and L. Han, “Continuous
the two surveying stations is continuously surveyed using a measurement-while-drilling utilizing strap-down multi-model surveying
system,” IEEE Trans. Instrum. Meas., vol. 63, no. 3, pp. 650–657,
triad of accelerometers and a triad of magnetometers. The Mar. 2014.
calculation algorithm is based on a strap-down computation [10] Q. Xue, R. Wang, L. Huang, and F. Sun, “Dynamic solution approach
mechanism and Kalman filtering. We developed downhole to the inclination and amizuth of bottom rotating drill string,” in Proc.
SPE Western Regional Meeting, Monterey, CA, USA, 2013, pp. 1–7.
measurement signal-processing methods and an algorithm to [11] R. Wang, Q. Xue, L. Han, F. Sun, and W. Yue, “Torsional vibration
dynamically solve the spatial attitude of the bottom rotating analysis of push-the-bit rotary steerable drilling system,” Meccanica,
drillstring. Initially, we established a theoretical system of vol. 49, no. 7, pp. 1601–1615, Jul. 2014.
[12] Q. Xue, R. Wang, and F. Sun, “Study on lateral vibration of rotary
dynamic solving and laid a theoretical and technical founda- steerable drilling system,” J. Vibroeng., vol. 16, no. 6, pp. 2702–2711,
tion for continuous MWD system and RSS development. Sep. 2014.
As the core of an RSS, MWD urgently needs to improve [13] M. ElGizawy, A. Noureldin, J. Georgy, U. Iqbal, and N. El-Sheimy,
“Wellbore surveying while drilling based on Kalman filtering,” Amer. J.
its accuracy and reliability. However, due to the complexity of Eng. Appl. Sci., vol. 3, no. 2, pp. 240–259, Feb. 2010.
BHA vibration and dynamic real-time measurement, precise [14] M. Elgizawy, A. M. Noureldin, and N. El-Sheimy, “Continuous wellbore
solution has become a key technology affecting the rapid surveying while drilling utilizing MEMS gyroscopes based on Kalman
filtering,” in Proc. SPE Annu. Tech. Conf. Exhibit., Florence, Italy, 2010,
development of MWD. For the downhole signal filtering pp. 5416–5428.
of sensors, a specific algorithm needs to be proposed by [15] A. S. Jurkov, J. Cloutier, E. Pecht, and M. P. Mintchev, “Experimental
analyzing the motion state of the bottom-hole drillstring. feasibility of the in-drilling alignment method for inertial navigation in
measurement-while-drilling,” IEEE Trans. Instrum. Meas., vol. 60, no. 3,
After filtering, the measurement signals are input into a pp. 1080–1090, Mar. 2011.
space attitude dynamic measurement algorithm model that [16] A. Noureldin, D. Irvine-Halliday, and M. P. Mintchev, “Accuracy limi-
is based on motion analysis. The use of stick-slip vibration tations of FOG-based continuous measurement-while-drilling surveying
instruments for horizontal wells,” IEEE Trans. Instrum. Meas., vol. 51,
phenomena can improve the accuracy of the solution. Using no. 6, pp. 1177–1191, Dec. 2002.
the dynamic measurement theory model and a trajectory pre- [17] A. Noureldin, H. Tabler, D. Irvine-Halliday, and M. Mintchev, “Testing
diction, a new space-state equation can be established. A more the applicability of fiber optic gyroscopes for azimuth monitoring for
measurement-while-drilling processes in the oil industry,” in Proc. IEEE
effective Kalman filter model was developed. This further Position Location Navigat. Symp., San Diego, CA, USA, Mar. 2000,
reduced the errors in the results of a dynamic solution. pp. 291–298.
154 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 65, NO. 1, JANUARY 2016

[18] E. Pecht and M. P. Mintchev, “Observability analysis for INS alignment Ruihe Wang was born in 1957. He received
in horizontal drilling,” IEEE Trans. Instrum. Meas., vol. 56, no. 5, the M.Sc. degree in petroleum engineering and
pp. 1935–1945, Oct. 2007. the Ph.D. degree in drilling engineering from the
[19] Z. Yanshun, W. Shuwei, and F. Jiancheng, “Measurement-while-drilling China University of Petroleum, Qingdao, China,
instrument based on predigested inertial measurement unit,” IEEE Trans. in 1989 and 1995, respectively.
Instrum. Meas., vol. 61, no. 12, pp. 3295–3302, Dec. 2012. He is currently a Professor and the Vice President
[20] C. Chen, Z. Yanshun, and L. Chunyu, “Surveying method of measure- with the China University of Petroleum, and direct-
ment while drilling based on the inertial sensor,” in Proc. 1st Int. Conf. ing the Oil and Gas Engineering Research Institute.
PCSPA, Harbin, China, Sep. 2010, pp. 1192–1195. As a Cross-Century Academic and Technical Leader
[21] P. H. Walters, “Method of determining the orientation of a surveying with China National Petroleum Corporation, Beijing,
instrument in a borehole,” U.S. Patent 4 709 486, Dec. 1, 1987. China, an expert who enjoys the State Council
[22] M. K. Russell and A. W. Russell, “Surveying of boreholes,” Special Allowance, he focuses on oil and gas engineering fluid mechanics,
U.S. Patent 4 163 324, Aug. 7, 1979. rock mechanics, and new technology and methods on rock breaking of drilling.
[23] A. Noureldin, T. B. Karamat, and J. Georgy, Fundamentals of Inertial
Navigation, Satellite-Based Positioning and Their Integration. Berlin,
Germany: Springer-Verlag, 2013.
[24] G. Minkler and J. Minkler, Theory and Application of Kalman Filtering.
Palm Bay, FL, USA: Magellan Book Co., 1993.
[25] J.-J. Ding, “Time frequency analysis and wavelet transform class note,”
Dept. Elect. Eng., Nat. Taiwan Univ., Taipei, Taiwan, Tech. Rep., 2007.

Qilong Xue was born in 1983. He received the

bachelor’s degree in electrical engineering and
the Ph.D. degree in drilling engineering from the Baolin Liu was born in 1959. He received the
China University of Petroleum, Beijing, China, Ph.D. degree in engineering from the China Uni-
in 2006 and 2014, respectively. versity of Geosciences, Beijing, China.
He was with PetroChina, Beijing, from 2006 to He is currently a Professor and the Director of the
2008. He was a Joint Researcher with the University Key Laboratory on Deep GeoDrilling Technology
of Calgary, Calgary, AB, Canada, from 2012 to of the Ministry of Land and Resources, China Uni-
2013, where he was involved in real time measure- versity of Geosciences. His current research inter-
ment of bottom drilling tool’s attitude and rotary ests include scientific drilling, environmental science
steerable drilling system. He is currently an Assistant drilling, computer controlled drilling, and new tech-
Professor with the China University of Geosciences, Beijing. His current nology drilling engineering.
research interests include measurement and control technology of drilling Dr. Liu is a member of the Professional Committee
engineering. of Mineral Exploration Engineering.

Henry Leung (F’15) received the Ph.D. degree in

electrical and computer engineering from McMaster
University, Hamilton, ON, Canada.
He was with Defence Research Establishment
Ottawa, Ottawa, ON, Canada, where he was involved
in the design of automated systems for air and
maritime multisensor surveillance. He is currently
a Professor with the Department of Electrical
and Computer Engineering, University of Calgary, Yunsheng Wu was born in 1977. He received the bachelor’s degree in electric
Calgary, AB, Canada. His current research system and automatism from Hebei Radio and Television University, Hebei,
interests include chaos, computational intelligence, China, in 1999.
information fusion, data mining, robotics, sensor networks, and wireless He has been with State Grid Haixing Power Supply Company, Cangzhou,
communications. China, since 2000.