Professional Documents
Culture Documents
10.0000@Www - Onepetro.org@journal Paper@SPWLA 1990 V31n1a2
10.0000@Www - Onepetro.org@journal Paper@SPWLA 1990 V31n1a2
Resistivity Logs
Walter W. Whitman: Colorado School of Mines, Golden, Colorado
Jurgen Schon: Bergakademie Freiberg, GDR
Guy Towle: Colorado School of Mines, Golden, Colorado
Jin-Hoo Kim: Dong-A University, Pusan, Korea
Abstract: This paper presents a method to automate the boundaries and resistivities without consideringthe bore-
inversion of a normal log. Most inversion procedures require hole. The procedure for producing this initial model will
an initial model, usually found by an interactive fit of the be a variation on that described in Pusch et al. (1987).
data with the forward model. The initial model is then im- This method essentially uses the available normal log data
proved iteratively by the inversion scheme. This paper shows to synthesize a focused log (laterolog7) from which the
how to automate the choice of initial model parameters
bed boundaries are inferred. The synthetic focused log is
through the construction of an approximate focused log
(laterolog7)from the original normal log. The bed boundaries then used directly to approximate the resistivities for the
are chosen by the relatively abrupt changes in the ratio of assumed beds.
the two focusing currents. The corresponding bed resistivities To implement the automatic inversion we need both
are then picked directly from the synthetic focused log. a forward model and a general inversion scheme. The
The forward model has horizontal layers of varying resis- forward model employs a finite difference (F-D) approach
tivities, borehole with mud, and invaded zones. A finite dif- to solve the potential equation with an exponentially in-
ference approximation to the potential equation is used in creasing grid spacing both radially and vertically from the
the forward modeling. The inversion procedure follows the current source. For details of the F-D scheme see Kim
ridge regression or Marquardt technique, wherein ill-condi- (1986) or Towle et al. (1988); for a summary see Whitman
tioned matrices are avoided by the addition of a stabilizing et al. (1989). For the general inversion scheme, which is
parameter.
For each of two examples with 16 in normal log data, the
the ridge regression or Marquardt inversion method, see
automatic inversion process is compared with one using hand- Whitman et al. (1989). Comparable forward modeling
picked initial bed boundaries and resistivities. The compar- results have been obtained using a different expanding
ison shows little difference in either the initial guess model grid system, also based on the finite difference method
or the final model, indicating that the automatic initial model (see Rosler and Schon, 1984; Schon and Weller, 1984).
performs quite satisfactorily. The forward modeling in Schon and Weller (1984) in-
cludes both normal and focused logs.
The first portion of this paper describes the generation
INTRODUCTION
of the automatic initial model. In the latter portion of
The purpose of this paper is to describe a method for this paper data from two 16 in normal logs are used to
the automation of the inversion of a normal electrode compare inversion (interpretation) employing the auto-
well log. By inversion we mean constructing an earth matic initial model versus hand-picked initial model.
model from the log data; on the other hand, the reverse Borehole effects are considered for both logs, whereas
of inversion- forward modeling- means generating a log invasion effects are studied only for the second log (they
from an earth model. Generally, parametric inversion appear to be minimal for the first log).
requires a forward model for the earth, using parameters
such as depth at lithologic changes, resistivity of the var- GENERATING THE INITIAL MODEL
ious rock types encountered, mud resistivity, hole di- Generally in inversion processes the initial model is
ameter, and resistivity and depth of invaded zones. In fashioned by an expert. The procedure described in this
particular, after the interpreter specifies approximations paper is not expected to better that type of initial model,
for the various parameters in the assumed earth model, but rather it is expected to yield roughly as good a final
this initial model is turned over to an automatic inversion earth model after the iterative inversion process improves
process, which generates the final earth model. It is the the initial earth model. This process can also provide a
initial model that this paper proposes to automate from good quality control check, as these normal logs, due to
the original log data, so that the entire inversion process their unfocused nature, require both talent and experience
is performed automatically. Of course, final validation to interpret. The only real demand on the interpreter,
remains in the hands of the interpreter. however, is to judge the geologic quality of the final in-
The automated initial model provides approximate bed version (interpretation).
"I
MEASURE ME NT CALCULATION
( N o r m a l log1
individual potential synthesized
tool log values (examples) focusing tool
RN
'
Y y l
1 1 1 1 1 1 1
;,;-2 I 1 1 1 1 1
Ib-I I I I I "111 1
I
10
Because the inversion process will refine a reasonable
initial model to fit the available data, the initial approx-
imation only needs to be not "too far" from a realistic
solution. Usually the positioning of bed boundaries and
associated resistivities can be improved somewhat from
the initial model with interactive modeling feedback. developed (see Siege1 et al., 1982; Schon, 1986; Weller,
However, as the whole inversion process is one of iter- 1986). Also, this principle led to a method to upproxi-
ative improvement, we feel that efforts to refine a rea- mutely synthesize laterologs (Pusch et al., 1987). Pusch
sonable initial model will not improve the final model. et al. (1987) isthe primary reference source for our de-
Constructionof the initial model proceeds in two stages. velopment.
The first and most critical stage is determiningthe number We refer to Figure 1 for the visualization of our syn-
and positioning of possible lithologic boundaries. The thesized laterolog and the actual normal log with a sam-
second stage is assigning appropriate resistivities for the pling of exemplary calculations. Our notation for the syn-
beds. Our experience suggests that the assignment of re- thesized laterolog7 will use A . as the focused current (io)
sistivitiesis less crucial than the boundary determination. electrode, with A representing the upper current (iJ elec-
Indeed, there are bedding sequences that might mask cer- trode and A , representing the lower current (i,) electrode.
tain boundaries due to the specific tool geometry. This The latter are, of course, the adjusting currents to main-
emphasizes the utility of and need for the variety of spac- tain the equipotential surfaces between the upper poten-
ings on a standard electrical survey log. tial electrodes, M , and M,, and the lower potential elec-
To pick the bed boundaries and resistivities for the trodes, M3 and M4, respectively. U will denote potential
initial model, we synthesized an approximate focused log with its subscript(s) implying the reference points. These
(laterolog7) from the measured normal log data. The potentials are with respect to the reference potential dis-
rationale behind this approach is that the focused log a) tant at the surface and denoted by N.
has a relatively sharp response to bed boundaries, and b) The entire approximation depends on three relation-
approximates the true bed resistivity. It will be demon- ships: 1)the equipotential relations, 2) the apparent upper
strated by example that a special current-ratio quotient and lower resistivities, and 3) the principle of superpo-
changes its value rather abruptly as boundaries are crossed. sition. For the details of the construction of the synthetic
For quality control the laterolog7 will be checked against laterolog7, see Appendix A;
the final earth model to corroborate its approximation to A plot of laterolog7 current ratios, specifically log(n,/
the true bed resistivities. n,) versus depth (see Equation (A.3) in Appendix), is used
The theoretical basis for the synthetic laterolog7 is giv- to pick approximate boundary levels for our earth model.
en in Roy (198 l), showing that the response of a focused This is done by choosing those depths which are extrema
tool can be synthesized from the responses of different or turning points on this log curve versus depth (see Figure
types of electrode spacings (normals, laterals). Using this 2 for an illustration of this plot), as the current ratio reacts
principle, various types of synthetic laterologs have been rather sharply to changes in resistivities. Note that ex-
0
Table 1: Earth parameters obtained from a semiautomaticin-
3J I version of the Hilchie log for a model with no borehole.
Layers Resistivities (ohm-m) Thicknesses (ft)
1 1.3 1.7
2 3.2 5.5
3 7.7 4.9
4 6.5 4.1
5 5.6 4.2
6 11.6 7.1
7 6.5 4.7
8 1.1 -
I Depth to top of the layer 1 is 5,379.3 ft.
Table 2: Diagonal elements of the resolution matrix obtained Table 5: Earth parameters obtained from an automatic inver-
from a semiautomatic inversion of the Hilchie log for a model sion of the Hilchie log for a model with borehole (Rm = 1.0
with no borehole. ohm-m, d = 8 in, no invasion).
Layers . Resistivity resolution Thickness resolution Layers Resistivities (ohm-m) Thicknesses (ft)l
1 0.99 0.99 1 1.5 2.7
2 0.99 0.99 2 3.2 4.8
3 0.99 0.92 3 6.6 5.0
4 0.98 0.80 4 6.3 2.2
5 0.98 0.84 5 5.1 6.2
6 1.oo 1.oo 6 11.9 4.5
7 0.99 0.99 7 6.4 6.7
8 0.99 - 8 1.o -
rms error of initial guess = 15.5%; best fit = 6.3%. I Depth to top of the layer 1 is 5,379.3 ft.
3 0.98 0.86 0
I 0 0 0 0 OBSERVE0 DATA
w-
m
4 0.90 0.85 In.
5 0.99 0.96
6 0.97 0.97
7 0.99 0.99 0
m-
8 0.99 - - m
I .
c
I
e .
,0
IY O
0 -
P
In.
Figure 5 for the corresponding comparison with the best Figure 4: Comparison between laterolog7 initial guess model
fit and a schematic of the earth model. There are changes and the Hilchie log data.
I .
I-
Table 7: Earth parameters obtained from a semiautomatic in- n . .
w
0 0o1 -
version of the East Salt Creek data for a model with borehole m
Lo.
(Rm= 0.8 ohm-m, d = 9 in, no invasion).
Layers Resistivities (ohm-m) Thicknesses (ft)'
m
0 ,
1 3.0 3.2 m
Lo.
2 5.7 4.0
3 3.7 3.8
4 11.0 3.8
5 3.7 3.4
6 15.0 4.0 CURRENT RATIO
7 10.7 4.2 Figure 6: Bed boundaries for the East Salt Creek log picked
8 20.0 -
from extrema of the log of the ratio of the synthetic upper and
Depth to top of the layer 1 is 5,900 ft. lower focusing currents.
0 v
I
"
0
i
0 -
6900.0 FT
I
3.2 FT
4.0 FT
3.7 FT
+9-
0.8
0"n-n
*
3.0
5.7
3.1
10.1
onn-n
onn-n
onn-n
onn-n
m 4.3 FT '
w . 2.9 FT 3.1 O M - H
4.0 FT 14.6 0nH-n
4.4 FT 10.9 o w n
- m
c
-_
0
w . Dv
20.0 own
- .
U
I .
I-
& .
w
0 w
-
-
o
w .
cn
0
I 0 0 0 0 OBSERVED OIITA
m
0 - 0 -
m
w .
m
In.
2-
: z 10 15 20 - 0 7
- .
I - Y ) .
$ 0 5 &I
RPPARENT R E S I S T I V I T Y IOHfl-fl) r -
I-
Figure 7: Laterolog7 approximateresistivitiesfor the East Salt &
w o
.
P 01-
Creek log computed from the apparent upper resistivities. m
w .
I
6900.0 FT
3.1 FT
--9. -1
3.0 own
4.0 FT 6.7 OHH-R
Layers Resistivity resolution Thickness resolution 3.8 FT 0.8 3.7 om-n
3.9 FT 10.7 OHH-H
1 1.oo 0.98 3.1 FT 3.3 OHH-H
4.8 FT 14.0 OHM-H
2 0.99 0.92 3.6 FT r . l 6 . 2 " 1 13.9 OHH-H
3 0.99 0.92 I 19.9 OHH-H
4 0.95 0.94 Dv
2.1 OHH-H
5 0.98 0.94
6 0.97 0.76
7 0.99 0.74
8 1.oo -
----.
rms error of initial guess = 13.6%; best fit = 2.4%. - BEST FIT n o m
BEST F I T LOO I
the Hilchie log, differences between the best fit models of
the hand pick and the automatic pick do not seem sig-
nificant, with the resolution appearing relatively good in
both cases.
In the final case we consider the effects of invasion using
an automatic initial model with invasion allowed in the
seventh layer; from the data this appears to be the most
likely layer. If we had been less restricted by computer
capacity we would have allowed invasion in all layers.
As it was, in the seventh layer we arbitrarily chose an
initial model with invasion depth of twice the borehole
diameter, and with a resistivity twice that of the mud (the
formation factor would be an appropriate general mul-
tiplier). The remainder of the initial model was deter-
mined automatically using the approximate synthesized
focused log (see Figure 10). The results of the best fit Figure 11: Comparison between best fit model with borehole
and invasion and the East Salt Creek log data using laterolog7
0 picks.
m
W
Ln
----.
I
INITIAL OUESS HOOEL
- INITIAL OUESS LOO inversion (see Figure 11 with its schematic of the earth
Q e Q Q OBSERVED DATA
model) may be found in Tables 1 1 and 12, which display
the resistivities and layer thickness and the resolution
matrix diagonal, respectively, yielding rms initial model
error of 14.8% and best fit error of 2.6%. Compared with
the previous model including borehole effects only, the
resistivities and thicknesses changed somewhat in the sixth
and seventh layers: This is no surprise, because there is
evidence of possible invasion in the log. Thus, not allow-
ing invasion in the previous model forced the inversion
to rationalize the log data. Note that the resistivity effect
in the model with no invasion has been more or less
averaged between the invaded zone and the virgin zone.
The resolution decreases for some parameters, probably
because the additional complexity of the model empha-
i sizes the essential interpretation ambiguities, given that
0
v I
m 1 ' ~ ' ~ " " 1 ' ~ " l ' ' ' '
6 10 15 20 26
we have data from only a single tool. In fact, the depth
APPARENT RESISTIVITY IOHPI-ti) and resistivity of the invaded zone are sufficiently poorly
Figure 10: Comparisonbetween laterolog7 initial guess model determined as to warrant little confidence in their values.
with borehole and invasion and the East Salt Creek log data. This is consistent, however, with the general concept that
Table 11: Earth parameters obtained from an automatic in- Table 12: Diagonal elementsof the resolution matrix obtained
version of the East Salt Creek data for a model with borehole from an automatic inversion of the East Salt Creek data for a
and invasion (Rm = 0.8 ohm-m, d = 9 in). model with borehole and invasion (Rm = 0.8 ohm-m, d = 9
____
in).
Layers Resistivities (ohm-m) Thicknesses (ft)'
3.0 3.1 Layers Resistivity resolution Thickness resolution
5.7 4.0 1 0.99 0.90
3.7 3.8 2 0.96 0.76
10.7 3.9 3 0.96 0.75
3.3 3.1 4 0.84 0.82
14.0 4.8 5 0.93 0.84
2.1/13.92 1.4/3.6 6 0.91 0.79
19.9 - 7 0.2610.87 I 0.2V0.77
Depth to top of the layer 1 is 5,900 ft. 8 1.oo -
2 Ri = 2.1 ohm-m, Di = 16.2 in.
rms error of initial guess = 14.8%;best fit = 2.6%.
I Resolution for Ri at layer 7 is 0.26, and that for Di is 0.21.
v GDR: Proc. 27th Intl. Geophys. Symp., Bratislava, v. 2, p. are possible for all distances equal to A M . For multiples of A M ,
82-93. we approximate as illustrated in the following example:
Towle, G. H., Whitman, W. W., and Kim, J. H., 1988, Electric
modeling log with a finite difference method: The Log Analyst, uM3(A2) = (iO/2kN)R, 64-51
v. 29, no. 3, p. 184-195. where the reciprocal of R is the average conductance spanned
Weller, A., 1986, Berechnung Geoelektrischer Potentialfelder by the new length of twice AM (using the conductances from
mit dem Differenzenverfahren: Freiberger Forschungshefte the short normal with electrode spacing AM). Therefore,
C405, Leipzig.
Whitman, W. W., Towle, G. H., and Kim, J. H., 1989, Inversion
R-l = + 2)
(%)[RN-l(i + RN-I(i+ 3)], (A4
of normal and lateral well logs with borehole compensation: where we again refer to Figure 1, and note that the 2 multiplier
The Log Analyst, v. 30, no. 1, p. 1-1 1. in the denominator of Equation (A.5) recognizes that we are
dealing with a normal tool of twice the length A M , hence, its
geometric factor doubles. Note that Equation (A.6) assumes an
APPENDIX homogeneous medium, whereas most earth environments are
essentially heterogeneous. However, this assumption is only used
The following describes the process of synthesizing a laterolog7
to calculate an initial model that will be iteratively improved
from a normal resistivity log. We begin with three relationships
in the inversion process. In fact, this type of approximation for
upon which the entire approximation depends: 1) the equipo-
an initial model is quite typical for parametric inversion pro-
tential relations, 2) the apparent upper and lower resistivities,
cedures, such as the Marquardt inversion.
and 3) the principle of superposition.
Using the principle of superposition, as expressed in Equation
The equipotential relations (to maintain the focus) are
(A.3), the equipotential relations in (A. I), together with Equa-
U, = U,, - U,, = 0, and tions (A.4) and (A.5), we can calculate the two current ratios n ,
(A.1) and n,. Using (A.2) we can compute the apparent upper and
U*fIM2= UM, - UM2= 0. lower resistivities for our synthesized tool. Specifically, the com-
putations proceed as follows. As there is another pair of equa-
The apparent upper and lower resistivities (for the upper fo-
tions from the principle of superposition for the other guard
cusing system between A , and M , and for the lower focusing
positions, M , and M4,these can be combined and reduced to
system between A , and M3)are
two, using the equipotential relations in (A. 1). After computing
R , , = k~U,,.,,/i,), and all of the potentials associated with the individual currents, as
('4.2) in Equations (A.4) and (A.9, these values can be used in the
Ra,/ = k x u M 3 / i 0 ) , final pair of superposition relations. Note that all terms will now
have i, in them, which can be canceled, leaving only n , and n,
where k,is the shape factor for the focusing tool geometry.
as unknowns in the resulting two combined superposition equa-
The principle of superposition (for the upper and lower po-
tions. The apparent resistivity log can now be computed from
tential guards) is expressed as
Equation (A.2), using Equation (A.3) with relations typified by
U,, = UM,,A,)+ n,(U,,,,J + n2(UM,,,)), and Equations (A.4) and (A.5) (note that i, cancels out).
(A.3)
uM3 = uM3(A,) + ndUM3(Al))+ n2(UM3(A2))?
where the second subscript in parentheses denotes the current
electrode; n , and n , are the current ratios i,/i, and i2/io,re-
spectively, dictated by the equipotential relations (A. 1). It is
important to note that by using the current ratios in Equation
(A.3), all potentials are related to the same current, i,.
We are now in a position to compute the various potentials
on the synthetic tool and the associated current ratios. The result
of these computations will be a) the current ratios n , and n,,
and b) the apparent resistivity log for the focusing system. For
the calculations we will assume that distances between elec-
trodes will be spaced at A M (or multiples thereof-see Pusch et
al., 1987), the normal tool spacing from which we obtain all our
data.
As a first partial result we use:
uMl(Ao, = (idkN)RN(i),and
(A.4)
= (io/kN)RN(i + 11,
UM~(A,)
where RN(i)is the resistivity of the normal array at depth point
A B O U T THE A U T H O R S
i (see Figure l), and kN is the shape factor for the normal tool Walter W. Whitman received a Bachelor of Engineering Phys-
array (kN= 47rAM). We note that calculations similar to (A.4) ics degree from Cornell University in 1959. In 1964 he was
awarded a Ph.D. in Mathematics, also from Cornell. His first ical measurements in the region of a borehole, in the far field
position was that of Instructor and then Assistant Professor of and in the laboratory. He is also developing models of porous
Statistics at the University of California at Berkeley. This was rocks to characterize their behavior.
followed by appointment as Associate Professor of Mathematics
at Colorado School of Mines. He was shortly promoted to Full
Professor at the youngest age in the some one hundred year
history of CSM.
In 1982 he was appointed a Full Professor of Geophysics at
Colorado School of Mines, his current position. His initial ex-
periences in well-logging were as collaborator circa 1969 with
the late G. R. Pickett. The initial effort was published in The
Log Analyst, concerning the Ratio Test for crossplots.
As a member of the Society of Exploration Geophysicists
(SEG), he has also published expository essays in The Leading
Edge, a general forum for Geophysicists,and is currently Chair-
man of the Editorial Board of The Leading Edge. He is the
American Association for the Advancement of Science (AAAS)
representative from the SEG, and a member of the Steering
Committee for the AAAS Consortium of Affiliates for Inter-
national Programs, representing the physical sciences.
As a member of the SPWLA, his interests include inversion
techniques and statistical applications. He is a member of the Jin-Hoo Kim graduated from Seoul National University, Ko-
Publications Committee of The LogAnalyst, and has been con- rea, in 1978 with a Bachelor of Science in Earth Science. He
ducting annual seminars based on the SPWLA Transactions at was employed by the Earth Science Department of Seoul Na-
Colorado School of Mines. He is Director of the CSM Center tional University while continuing his geophysics studies as a
for Well Logging and Petrophysics, which commenced opera- postgraduate. He received his Ph.D. in geophysics from Colo-
tions in the Fall of 1988. rado School of Mines, Golden, Colorado, in 1986. While at
CSM, he was involved in the Borehole Geophysics Project for
two years as a research assistant. Currently, he is an Assistant
Professor of Geophysics at Dong-A University, Pusan, Korea.
He is a member of the SPWLA, SEG, and ASME. He also serves
as a member of the editorial staff for the Korea Committee for
Ocean Resources and Engineering (KCORE).