You are on page 1of 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317833456

Inversion of dielectric logging using mixing model as a regularization function

Conference Paper · June 2017

CITATION READS
1 91

4 authors:

Babak Kouchmeshky Alberto Mezzatesta


Baker Hughes Incorporated Baker Hughes a General Electric Company
17 PUBLICATIONS   151 CITATIONS    68 PUBLICATIONS   145 CITATIONS   

SEE PROFILE SEE PROFILE

Roberto Arro Otto Fanini


Baker Hughes Incorporated R&D
4 PUBLICATIONS   16 CITATIONS    78 PUBLICATIONS   130 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

MULTICOMPONENT INDUCTION LOGGING TOOL View project

All content following this page was uploaded by Babak Kouchmeshky on 23 June 2017.

The user has requested enhancement of the downloaded file.


SPWLA 58th Annual Logging Symposium, June 17-21, 2017

INVERSION OF DIELECTRIC LOGGING USING MIXING MODEL AS A


REGULARIZATION FUNCTION
Babak Kouchmeshky, Baker Hughes, Alberto Mezzatesta, Baker Hughes, Roberto Arro, Baker Hughes, Otto
Fanini, Baker Hughes

Copyright 2017, held jointly by the Society of Petrophysicists and Well Log regularization of the petrophysical parameters, and
Analysts (SPWLA) and the submitting authors.
This paper was prepared for presentation at the SPWLA 58th Annual Logging structure the electromagnetic properties of the
Symposium held in Oklahoma City, Oklahoma, USA, June 17-21, 2017. formation using mixing models.

ABSTRACT In order to compare the two inversion methods,


their performance is put into test using an
Dielectric log can be used to obtain petrophysical example. For a realistic example experimental data
properties of a formation like water saturation, on dispersive electromagnetic properties of a core
water resistivity at DC, and formation texture. is used. The experimental real and imaginary part
Dielectric tool operates at a wide frequency range of dielectric dispersion are further used to obtain
from a few MHz to 1 GHz. Since the gathered data the relative magnitude and phase of an
is inevitably contaminated with noise, studies that electromagnetic wave travelling in the formation
quantify the effect of noise on the interpretation of and picked up by two receivers. This synthetic tool
the dielectric log are very useful. response is then perturbed by noise. Monte Carlo
method is used to study the effect of the noise on
This work provides an overview of inversion the inverted petrophysical properties of the
methods used for interpretation of dielectric log. formation. The probability distribution of the
First the classical method of inversion is inverted petrophysical parameters obtained from
introduced in which the task of quantifying the noisy data using two different inversion
petrophysical parameters is divided into two steps. methods are compared to each other. It is shown
The first step involves inverting for the that the proposed method is more tolerant to the
electromagnetic properties of the formation noise, and provides a probability distribution for
(permittivity and conductivity) from magnitude each petrophysical parameter that leads to an
and phase of electromagnetic wave recorded at the expected value for these parameters that is very
receiver. These electromagnetic properties are close to the actual values set in the lab, and also
obtained for each frequency of operation to obtain results in a much smaller standard deviation in
their variation with respect to frequency comparison with the classical inversion method.
(dispersion). It is this dispersive behavior of
permittivity and conductivity that is used in the
second step to obtain petrophysical properties of INTRODUCTION
the formation. For this step, valid mixing models
are required to relate petrophysical parameters of Dielectric log can provide invaluable information
formation and dispersive electromagnetic on the petrophysical parameters of a formation. It
properties. It is shown that the presence of noise in measures the magnitude and phase of the
the data can lead to ill-posed inverse problems electromagnetic wave propagating through the
where multiple answers can be present. formation. This data is used to infer
electromagnetic properties of a formation (Cox
In order to avoid having an ill-posed inverse and Warren 1983, Rau et. al. 1991, Hizem et. al.
problem, use of regularization method is 2008). Usually the magnetic permeability is
assumed to be that of vacuum (μr = 1) while the
suggested. An inversion methodology is proposed
relative permittivity is generally a complex value
that combines the two steps of inversion into one.
and is related to the real permittivity and
In doing so, only one inversion problem needs to
conductivity of a formation through the following
be solved that directly obtains the petrophysical equation.
parameters. It is shown that the new inversion
method can seamlessly take advantage of the
1
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

σ(ω)
ε̃r (ω) = ε′r (ω) + iε′′
r (ω) = εr (ω) + i (1) and receivers where multiple depths of
ωε0
investigation are studied.
where ε̃r is the complex value representing relative The structure of the paper is as follows. First, two
permittivity, εr is the real part of relative dielectric different inversion methods for interpretation of
property, σ is electrical conductivity, ω is angular dielectric log are explained in detail. The
frequency, and ε0 is dielectric constant of vacuum. differences between the two methods are
highlighted, emphasizing the advantage of one-
The electromagnetic properties of the formation step inversion method in incorporating
exhibit a variation with respect to frequency that is regularization. Next, these two methods are
called dispersion. It is this dispersive behavior that applied in an example with synthetic data
can be used in inferring petrophysical properties of perturbed by noise. Finally, the ability of the
a formation. Dielectric log uses contrast in the methods to provide accurate petrophysical
permittivity of water and the other elements parameters from a noisy input data is discussed.
present in the formation to obtain information on
water saturation, water resistivity at DC, and
TWO-STEP INVERSION METHOD
formation texture. Dielectric log interpretation
provides an advantage over traditional resistivity
In this method the inversion of dielectric log is
logging in formations where the water resistivity is
divided into two steps (Hizem et. al., 2008). The
unknown or varies significantly over short
distances. first step involves inverting for the
electromagnetic properties of the formation,
The accuracy of the petrophysical parameters namely permittivity and conductivity, using the
obtained from this log is affected by noise level. magnitude and phase response of the tool. The
The aim of this paper is to study the possibility of inversion is done at multiple frequencies. At each
increasing the noise tolerance by modifying the frequency the output of inversion is permittivity
inversion methods. The current paper addresses and conductivity of the formation. This step can be
the effect of noise in the recorded dielectric log summarized by the following equation.
data on the inverted petrophysical parameters. The
existence of noise in the recorded data leads to an 𝑎𝑟𝑔𝑚𝑖𝑛𝜀𝑟(𝜔),𝜎(𝜔) ‖𝑴(𝜔) − 𝒇(𝜀𝑟 (𝜔), 𝜎(𝜔))‖2 (2)
ill-posed inverse problem. In order to address this
problem and make the inverse problem at hand where ‖. ‖2 is ℓ2 norm, M is a vector containing
more tolerant to noise, usually some sort of the measured magnitude and phase at each
regularization is performed. frequency, and f is a vector containing the
calculated magnitude and phase at each frequency
Two different inversion methods are presented in as a function of permittivity, ε, and conductivity,
this paper. The major difference between the
σ, of formation. Usually constraints are added to
methods is reflected in their ability to take
the inversion such that the obtained values for
advantage of the regularization methods and hence
permittivity and conductivity have the following
their tolerance to the level of noise. The first
conditions.
method presented is the classical method used in
inversion of dielectric log (Hizem et. al., 2008).
This method divides the task of inversion into two 𝜀𝑟 (𝜔) > 1 , 𝜎(𝜔) > 0 (3)
steps. The first step inverts for electromagnetic
properties of the formation using the recorded Collecting the inverted electromagnetic properties
relative magnitude and relative phase of of formation at multiple frequencies results in a
electromagnetic wave. The results of this step at dispersive relation obtained for both permittivity
different frequencies are used as an input to the and conductivity. This dispersive relation is used
second step of inversion where petrophysical as the input for the second step of inversion where
properties are obtained. The second method the petrophysical properties of formation are
combines these two steps into one and in the obtained as follows
process enables the use of regularization leading to
an inversion problem that is more tolerant to noise 𝑎𝑟𝑔𝑚𝑖𝑛𝑿 ‖𝑷 − 𝒈(𝑿, 𝜑)‖2 (4)
level. It should be noted that the methods deployed
in this paper can be easily applied to an array where P is a 2 × N vector with N being number of
distribution of sensors with multiple transmitters
2
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

recorded frequencies and contains permittivity and petrophysical parameters.


conductivity obtained for each frequency from the
first step 𝑿0 = [𝜀𝑚0 , 𝜎𝑤0 , 𝑆𝑤0 , 𝑅𝑎𝑏0 ] (8)

𝑷 = [𝜀𝑟 (𝜔1 ), 𝜎(𝜔1 ), 𝜀𝑟 (𝜔2 ), 𝜎(𝜔2 ), … , 𝜀𝑟 (𝜔𝑁 ), 𝜎(𝜔𝑁 )] (5) Prior knowledge about formation or petrophysical
values obtained from adjacent depth can be used in
Alternatively, X is a vector containing selecting values for elements of X0 . It should be
petrophysical parameters of the formation. These noted that selecting a large value for α would lead
properties include water saturation, Sw , water to an optimal value of X satisfying above equation
conductivity at DC, σw , relative permittivity of dry that is very close to X0 . Conversely, selecting a
rock, εm , and in more detailed models, the texture very small value for α reduces the effect of
of formation is included as aspect ratio of regularization and results in the optimal value of X
inclusions, R ab . As it can be seen porosity, φ, is satisfying only the noisy data presented in P (Eq.
also needed but not included as a parameter of 5).
inversion as it needs to be measured independently
from other logging tools. If porosity is included as One drawback of this method of inversion is that
a parameter of inversion for dielectric log, it can separate inversions are needed for the
lead to an ill-posed problem as shown by aforementioned steps. This can lead to less noise
Kouchmeshky and Fanini (2016 a). In Equation 4, tolerance as the inherent errors associated with
g is a 2 × N vector that contains permittivity and numerical convergence of multiple inverse
conductivity values calculated for a formation with problems add up. As noted, the results of first step
petrophysical parameters X and following a above are electromagnetic properties at different
predetermined mixing model. For calculating the frequencies that will be used for the inversion
elements of vector g from petrophysical problem in the second step. As such, no
parameters represented by vector X and porosity φ regularization at the first step is recommended
knowledge of the mixing model appropriate for the because enforcing any arbitrary regularization
formation is required. This information can be function at step 1 would effectively alter the shape
obtained through lab measurement on rock cores of dispersive function, representing variation of
from the formation under study. As shown by electromagnetic properties with respect to
Kouchmeshky et. al. (2016 b) using the quality of frequency. It is precisely this dispersion function
fit to logged data at frequency range of operation that is used in the second step to obtain
for dielectric tool is not sufficient as an indicator petrophysical properties. The most natural choice
to choose the appropriate mixing model. For a for regularization function for the first step would
better judgment on the performance of a mixing be the same mixing model that is used in the
model, access to a wider frequency range through second step. The difficulty with this approach is
lab measurements is required. The inversion in the that to be able to use the mixing model one would
second step is subjected to the following require determination of the petrophysical
constraints properties which are only taken into account in the
second step of inversion. Following this
𝜀𝑚 > 1 , 𝜎𝑤 > 0, 0 ≤ 𝑆𝑤 ≤ 1 (6) discussion, we propose a one-step inversion
method for dielectric logging in the following
In reality, the collected data is subjected to noise. section.
This can lead to an ill-posed inversion problem. In
order to avoid this scenario, regularization ONE-STEP INVERSION METHOD
methods are typically used. Inversion for the
second step can be modified to include As discussed in the previous section, the presence
regularization of the petrophysical parameters as of noise in collected data makes application of
follows regularization methods very appealing. These
methods increase the tolerance to noise and ensure
𝑎𝑟𝑔𝑚𝑖𝑛𝑿 (‖𝑷 − 𝒈(𝑿, 𝜑)‖2 + 𝛼‖𝑿 − 𝑿0 ‖2 ) (7) unique answers to an otherwise ill-posed inverse
problem. We saw in the previous section that the
where α is the regularization coefficient and X0 is natural choice for regularization functions for the
a vector that contains expected values of first step of inversion, was the same function used
3
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

as mixing model for the second step. This led to conductivity values is needed. The petrophysical
the following procedure for the inversion parameters obtained from the inversion can be
used with the selected mixing model to generate
𝑎𝑟𝑔𝑚𝑖𝑛𝑿 (‖𝑴∗ (𝝎) − 𝒉(𝑿, 𝜑, 𝝎)‖2 + 𝛼‖𝑿 − 𝑿0 ‖2 ) (9) the electromagnetic properties of the formation
(dispersive relation).
where h is a 2 × N vector that contains the
calculated magnitude and phase at each frequency, It should be noted that prior knowledge on the
M ∗ is a 2 × N vector that contains the measured formation or lab studies on the cores obtained
magnitude and phase for all frequencies present in from formation may indicate that more than one
vector ω = [ω1 , ω2 , … , ωN ], vector X contains mixing models are applicable. In that case, both
petrophysical parameters of formation, φ is two-step and one-step inversion methods can be
porosity, vector X0 contains expected values of repeated for allowable choices of mixing model,
petrophysical parameters and α is the and the best fit to the recorded data can be selected
regularization coefficient. to represent the formation.

In calculating the elements of vector h the EXAMPLE


following steps should be taken. First, a predefined
mixing model is chosen. Then, the petrophysical In this example synthetic data representing a noisy
parameters assigned to vector X and porosity, φ, tool response in a formation with known
are used to find permittivity and conductivity of petrophysical properties is used to study the
formation at angular frequencies in ω. The result performance of the two different inversion
is the electromagnetic properties of the formation methods outlined in the previous sections. The
represented as a dispersive relation for permittivity petrophysical parameters obtained from the
and conductivity calculated for the petrophysical application of each of these two inversion methods
parameters, X. These electromagnetic properties are compared with the actual petrophysical
are then passed to the forward model that predicts parameters used to generate the tool response to
the behavior of the tool in the formation. This can compare the noise tolerance of each method.
be a model that simplifies the transmitters and
receivers as dipoles and assumes the formation to To make sure the data represent a realistic
be homogenous, or it can be a more detailed model situation, experimental data on a rock core are
taking into account the effect of finite size of the used (Golikov et. al., 2015). Figure 1 shows the
sensor, borehole effect, etc. Using the predicted real and imaginary part of dielectric dispersion for
electromagnetic properties of the formation as a sandstone core at a wide frequency range. The
inputs to the forward model, leads to calculating core’s porosity was 19% and it was fully saturated
values for the tool response in the format of with water. A fitted mixing model to the
relative magnitude and phase for each frequency. experimental data is also shown in the same figure.
The resulting magnitude and phase populate the Our previous study (Kouchmeshky et. al. 2016b)
2 × N elements of vector h. on this core data showed that Maxwell-Garnett can
be used as a mixing model that provides a good fit
𝒉 = [𝑚𝑎𝑔𝜔1 , 𝑝ℎ𝑎𝑠𝑒𝜔1 , 𝑚𝑎𝑔𝜔2 , 𝑝ℎ𝑎𝑠𝑒𝜔2 , … , 𝑚𝑎𝑔𝜔𝑁 , 𝑝ℎ𝑎𝑠𝑒𝜔𝑁 ] (10) to the dispersive real and imaginary part of
permittivity (or permittivity and conductivity) as
where mag ωi and phaseωi are respectively the well as an accurate estimation of water saturation
relative magnitude and relative phase between two and water resistivity at DC. Details on this mixing
receivers at frequency ωi . In this method the model are provided in the appendix. Hence, this
petrophysical parameters are directly obtained mixing model is selected for the inversion in both
from one inversion. Reducing the number of of the methods used in the current example.
inversion problems would increase the tolerance to
noise. Additionally, the regularization can be The dispersive electromagnetic properties of the
easily applied on the petrophysical parameters and formation are assumed to follow the results
since the mixing models are used in the forward obtained from the core data. Next, the transmitters
model to generate the electromagnetic properties and receivers of the dielectric tool are simplified
of the formation from petrophysical properties, no as perfect magnetic dipoles in a homogenous space
extra regularization on the permittivity and with aforementioned electromagnetic properties.
4
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

Using these assumptions, the tool response can be The response of the tool at different frequencies
obtained from the following relation from 20 MHz to 1 GHz are obtained from previous
equation and subjected to noise as below.
𝐵𝑟2 𝑟 3 𝑘𝑟 +𝑖
= ( 1) 2
𝑒 𝑖𝑘(𝑟2−𝑟1) (11) ∗ = 𝑚𝑎𝑔 × (1 + 𝑢 )
𝐵𝑟1 𝑟2 𝑘𝑟1 +𝑖 𝑚𝑎𝑔𝜔 𝑖 𝜔𝑖 𝑖 (12)

where transmitter and receivers are represented by ∗ = 𝑝ℎ𝑎𝑠𝑒 × (1 + 𝑣 )


𝑝ℎ𝑎𝑠𝑒𝜔 (13)
𝑖 𝜔𝑖 𝑖
points in space, k = ω(με̃)0.5 is the complex wave
number with ω as angular frequency and μ and ε̃ where mag ωi and phaseωi are the magnitude and
as magnetic permeability and complex permittivity phase of the tool at frequency ωi , while mag ∗ωi
respectively, rm is the magnitude of the vector 𝐫m and phase∗ωi are the tool response perturbed by the
connecting transmitter and receiver m, and Brm is
noise. ui and vi are independent random variables
the component of magnetic field along vector 𝐫m representing the error associated with recording
at receiver m. It is assumed that all transmitter and the tool response and follow a Gaussian
receivers can be represented by collinear points distribution N(μ, σ) where μ = 0 and σ = 0.05.
and that magnetic moments of transmitter and There are totally 2 × N independent random
receivers are all parallel to the direction of vector variables representing the noise
connecting the transmitter and receivers, 𝐫m . E = [u1 , v1 , u2 , v2 , … , uN , vN ].

Monte-Carlo simulation was used to study the


effect of noise on the inverted petrophysical
parameters using the two inversion methods
described in previous sections. 10000 realization
of the vector E were used. For each realization, the
perturbed tool response is calculated and used to
obtain the inverted petrophysical parameters σw
and Sw representing water conductivity and water
saturation, respectively.

Figure 2 shows the relative magnitude and phase


between receivers as the tool response in a
homogenous medium with electromagnetic
properties shown in Figure 1. The perturbed
response from a realization of random vector
representing noise is also shown. As discussed
before and observable from the figure, the
elements of the random vector representing noise
are independent from each other.

Figure 3 shows the distribution of petrophysical


parameters obtained from two-step inversion
method. The actual value for the petrophysical
parameters is shown as a dotted line. As it can be
seen, the inverted petrophysical parameters with
the highest probability, correspond to values that
are far from the actual ones. In the case of the
inverted water saturation, Sw , although the value
corresponding to the highest probability matches
the actual water saturation, the probability of
finding an inverted water saturation that is far
Figure 1, Experimental dispersive permittivity and fitted MG from its actual value is not negligible. As it can be
mixing model for a sandstone core. Top, real part of seen, the two-step inversion method results are not
permittivity. Bottom, imaginary part of permittivity. satisfactory as it is not tolerant to the noisy tool
5
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

response used in this example.

Figure 3, Inversion with 2 steps; Top, probability distribution


of the inverted water salinity, ssw. Bottom, probability
Figure 2, Relative magnitude and relative phase measured at distribution of the inverted water saturation, Sw. Dashed line
discrete frequencies. Both actual response (solid line) and a shows the actual petrophysical parameters.
realization of perturbed response (dashed line) are shown.

Also, the inverted petrophysical parameters using


the one-step method is displayed in Figure 4 where
the location of actual petrophysical parameters is
also demonstrated by a dotted line. It can be seen
that the inverted petrophysical parameters with
highest probability correspond to the actual
petrophysical parameters. Furthermore, the
probability of finding inverted petrophysical
parameters far from actual parameters is small, as
shown in Figure 4. Hence, it can be concluded
that the one-step inversion method is much more
tolerant to the error in the recorded tool response
than the two-step inversion method.

6
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

step. The inversion in the second step provides


petrophysical parameters of the formation. In
doing so, petrophysical models (mixing laws) that
relate the petrophysical parameters and
electromagnetic properties of formation need to be
validated by experimental results on cores.
Appropriate constraints for the petrophysical
parameters are also used in the inversion problem
for the second step. Additionally, regularization of
the solution to the inversion problem of the second
step can be implemented using the prior
knowledge obtained either from past experience or
from values obtained from adjacent zones in the
borehole.

It should be noted that application of the


Figure 4, Inversion with one-step using mixing model as regularization on the solution of the inversion
regularization; Top, probability distribution of the inverted problem in the first step is not recommended. The
water salinity, ssw. Bottom, probability distribution of the reason for this is that the outputs from the first
inverted water saturation, Sw. Dashed line shows the actual inversion at different frequencies form a dispersive
petrophysical parameters. behavior and it is exactly this dispersive relation
that is the subject of the inversion in the second
DISCUSSION step. Any attempt for arbitrarily regularizing the
solution that alters the dispersive behavior of
The goal of this paper is to increase the noise permittivity and conductivity can add unnecessary
tolerance of the inversion method for dielectric noise to the process and deteriorate the overall
logging. The existence of noise in the data can noise tolerance of the method. The only plausible
lead to an ill-posed inverse problem where the method of regularizing the permittivity and
uniqueness of the solution cannot be guaranteed. conductivity as the solution of the inverse problem
In order to avoid this problem, regularization in the first step is to use the same validated mixing
methods are usually incorporated into inversion. model that relates dispersive behavior of
permittivity and conductivity to the petrophysical
We have discussed two different inversion parameters of the formation in the inversion
methods for dielectric logging. The first method problem of the second step. However, since the
divides the task of inversion into two steps. The application of the mixing model requires having
first step uses the magnitude and phase of the petrophysical parameters as the independent
electromagnetic wave propagating in the formation variables and they are not available for the
to provide electromagnetic properties of formation inversion problem of the first step, these models
at different frequencies through an inversion cannot be used for regularizing the solution of the
problem. The solution to inversion problems inversion problem in the first step.
would be those values of permittivity and
conductivity at each frequency that can reproduce Drawbacks of the classical two-step inversion
the measured magnitude and phase of method for interpretation of dielectric log
electromagnetic wave propagation in the discussed in the previous section lead us to
formation. Appropriate constraints on the values of propose another approach for inversion problems.
permittivity and conductivity are implemented to In the new method, only one inversion problem
guarantee values for permittivity and conductivity needs to be solved where the observable
that are physically plausible. The values obtained parameters in the inversion problem are the
at the end of this step for permittivity and magnitude and phase of electromagnetic wave, and
conductivity at different frequencies exhibit a the solutions are the petrophysical parameters of
variation with respect to frequency that is called inversion. The objective function is still the
dispersion. It is this dispersive behavior that would difference between measured and calculated
be the input to the inversion problem in the second magnitude and phase of electromagnetic wave, but
7
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

the independent variables are the petrophysical APPENDIX


parameters. In order to provide the relation
between independent and observable variables, the MAXWELL-GARNETT MODEL
forward model in this method is divided into two
steps. The first step uses the validated mixing Maxwell-Garnett (MG) is based on how
model to convert the petrophysical properties of introduction of inclusions into a host material
formation to electromagnetic properties of (background) would alter the distribution of
formation, and the second step uses the electric field. The effective relative permittivity in
electromagnetic properties of formation to Maxwell-Garnett model is calculated as below
calculate the magnitude and phase of [Maxwell-Garnett, 1904, Sihvola et. al., 1989,
electromagnetic wave. Since a mixing model is Seleznev et. al., 2004]
used to generate the permittivity and conductivity,
at each frequency they follow the intended
dispersive behavior. It should be noted that
1 ε̃b
selecting the mixing model beforehand does not ∑ f (ε̃ −ε̃b ) ∑i=x,y,z
3 j j j ε̃b +Nij (ε̃j −ε̃b )
impose any drawback, as the classical two-step ε̃f = ε̃b + 1 Nij (A1)
1− ∑j fj (ε̃j −ε̃b ) ∑i=x,y,z
3 ε̃b +Nij (ε̃j −ε̃b )
method of inversion also requires the selection of
mixing model for the second step of inversion. In
addition the fact that only one inversion problem Where index j is over the inclusion types, index i
needs to be solved rather than two, reduces the is for three principal axes, ε̃b is the permittivity of
error associated with numerical convergence of the background, ε̃j is the permittivity of inclusions, f
inverse problem, and requires less computational is volume fraction of inclusions and N is the
resources. The proposed method is also able to depolarization factor that depends on the shape of
provide the dispersive conductivity and the inclusions. It is assumed that the inclusions are
permittivity of the formation for the petrophysical dilute and they are homogenously distributed with
parameters of inversion as they are calculated as random orientations. Assuming ellipsoidal shapes
part of the forward model used in the inversion. for the inclusions (with the main axes of a, b=c
Finally, reducing the number of inversion along local axes x, y and z respectively), the
problems to one imposes no drawbacks on the depolarization factors for inclusion j along
number of parameters used or the type of forward principal axes can be written as
model implemented. For example, if properties of
bore-hole fluid or geometric parameters of the
formation are to be added as independent variables
Rab ∞ −1.5
in the inversion, the process can be done Naj =
2
∫0 (s + R2ab ) (s + 1)−1 ds (A2)
seamlessly using the proposed method.
Nbj = Ncj = 0.5(1 − Naj ) (A3)
CONCLUSION
where R ab = a/b is aspect ratio of inclusion. The
Two different methods for inverting dielectric parameters for this mixing model are ε̃m , SW and
logging data are proposed. These methods take the σW , R ab .
relative magnitude and phase of an
electromagnetic wave traveling in the formation
and extract relevant petrophysical properties. The
difference between methods is in their ability to REFERENCES
include a petrophysically based regularization
function in their search for the best parameters that Cox P.T., Warren W.F., “Development and testing
could explain the observed response. The methods of the Texaco dielectric log”, SPWLA 24th annual
are applied to synthetic data perturbed by noise logging symposium, 1983
and the advantage of the method that includes the
petrophysically based regularization (One-step Golikov, NA, Elcov, T.I., Melkozerova, S.N.,
method) is shown. Novikov,I.V., “Measurement of complex dielectric
permittivity of fluid-saturated rock samples“,
Internal report from Trofimuk Institute of
8
SPWLA 58th Annual Logging Symposium, June 17-21, 2017

Petroleum Geology and Geophysics at Siberian Evaluation in Baker Hughes, Inc. supporting Drilling
Branch of the Russian Academy of Sciences, 2015 Systems, Wireline Systems, and Drill Bit divisions.
Currently, he has responsibility over areas of product
Hizem, M., Budan H., Deville B., Faivre O., research and development processes, technical
Mosse L., Simon M., ”Dielectric dispersion: A standards, service and product technical compliance,
new wireline petrophysical measurement”, SPE performance, integrity and content, patent portfolio
annual Technical Conference and Exhibition, 2008 management, and product and technology roadmaps.
During his career he has worked in innovation and
Kouchmeshky, B., Fanini, O., “Dielectric logging state-of-the-art projects in multiple industries such as
for heavy oil reservoirs”, World Heavy Oil
oilfield services, semiconductor, broadcast, printing,
Congress, WHOC16-400, 2016
communication, and textile industries. Fanini has
accumulated 35 years of engineering experience, over
Kouchmeshky, B., Fanini, O., Nikitenko, M.,
“Validating mixing models for dielectric logging”, 50 patents issued, and over 60 technical publications.
SPE Russian petroleum technology conference, He holds a Master of Science degree in Electrical
182096, 2016 Engineering from Texas A&M University, an MBA
degree from Houston Baptist University, and a
Maxwell-Garnett, J.D., "Colures in metal glasses Bachelor degree in Electrical Engineering from the
and in metal films", Transactions of the Royal Universidade Federal do Rio de Janeiro (UFRJ) in Rio
Society, CCIII, London, 385-420, 1904 de Janeiro, Brazil. Fanini is a member of IEEE, SPE,
SPWLA, ASQ, and INCOSE.
Rau R., Davies R., Finke M., Manning M.,
“Advances in high frequency dielectric logging”, Alberto G. Mezzatesta is the Manager of Research
SPWLA 32nd Annual logging symposium, 1991 NMR Science and Integrated Interpretation within the
BHI Houston Technology Center. With more than 35
Seleznev, N., Boyd, A., Habashy, T, "Dielectric years in the Oil Industry, Alberto has been involved
mixing laws for fully and partially saturated with E&P, Consulting, Technology Development, and
carbonate rocks", SPWLA 45th Annual logging the Academia. During his tenure of 24 years with BHI,
symposium, 2004 Alberto has been involved with Formation Evaluation,
Interpretation Development, Logging Tool Design and
Sihvola, A., and Kong, J.A., “Effective Development, Geoscience Applications, and Reservoir
permittivity of dielectric mixture”, IEEE Engineering. Alberto received a Petroleum Engineering
transaction on Geoscience and Remote sensing, degree from the National University of Cuyo,
26(4), 420-429, 1989
Argentina, and a Ph.D. degree in Chemical Engineering
from the University of Houston. He is an active
member of the SPE and SPWLA and has authored and
ABOUT THE AUHTOR co-authored several publications and patents.
Babak Kouchmeshky received his PhD from Cornell
Roberto Arro is a Geoscientist at the Houston
University school of Engineering in 2009. Between
Technology Center since June 2014. He received his
2009 and 2011 he was a post doc at the national center
Telecommunications Engineer grade from the La Plata
of hypersonic structures in UTA. He joined Baker
National University, Argentina in 1980, then, he
Hughes in 2011 where he is currently an R&D
initiated his professional career in Baker Hughes
engineer. He has worked on various aspects of sensor
(previously Dresser Atlas) in the same year, completing
development including design, simulation,
his Field Engineering progression in 1988 as General
experimentation, and interpretation for acoustic and
Field Engineer. He has acted as Operations Manager
electromagnetic tools. His current focus is on the
and District Manager at various locations in Argentina.
development of a dielectric logging tool. He is the
During 2007-2008 he took a M.Sc course in Applied
author of several patent applications on the sensor
Statistics in the Universidad Nacional del Comahue,
design, inversion algorithms, and petrophysical
Argentina.
interpretation for wireline tools.

Otto Fanini currently holds the position of Global


Principal Engineer (Electrical) for Drilling &

View publication stats

You might also like