# SPWLA 51st Annual Logging Symposium, June 19-23, 2010

**QUANTITATIVE INTERPRETATION OF PULSED NEUTRON CAPTURE LOGS IN THINLY-BEDDED FORMATIONS
**

Jordan G. Mimoun and Carlos Torres-Verdín, The University of Texas at Austin, William E. Preeg, Consultant

Copyright 2010, held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors. This paper was prepared for presentation at the SPWLA 51st Annual Logging Symposium held in Perth, Australia, June 19-23, 2010.

INTRODUCTION Since the early days of Neutron Lifetime1 logging instruments (Youmans et al., 1964), PNC tools have become indispensable for cased-hole saturation monitoring. PNC logs can be used to discriminate hydrocarbonfrom water-bearing reservoir units because of the considerable difference of thermal-neutron decay rates in saline water and hydrocarbons. The difference in physical behavior stems from the presence of chlorine in water, which is the strongest absorber of neutrons in common subsurface formations. In addition to identifying pay zones among water-saturated layers, PNC measurements are acquired throughout the life of hydrocarbon-producing wells to monitor variations of water saturation with time (Clavier et al., 1971a). Timelapse monitoring is a widely used application of sigma logs in the petroleum industry and has received much attention. Kimminau and Plasek (1992) formulated a design checklist for interpretation that takes into account uncertainties inherent to PNC measurements. The extension of time-lapse logging in waterflooded reservoirs is the estimation of waterflood residual oil saturation via the log-inject-log procedure (Richardson et al., 1973). Despite recent breakthroughs in resistivity measurements behind casing (Bartenhagen et al., 2001), the need remains for resistivity-independent saturation evaluation in both open- and cased-hole logging operations. Low-resistivity pays as well as non-Archie formations (e.g. carbonates and shaly sands) may limit the reliance on resistivity measurements, hence the need for an alternative method for saturation appraisal and cross-validation of interpretations. This is why the PNC time-lapse technique is routinely used in hydrocarbonproducing wells: sigma changes can be directly associated with fluid-saturation changes. Well logs acquired in thinly-bedded formations can exhibit significant shoulder-bed effects, rendering them spatially smooth. In the specific case of PNC logs, neighboring layers may contribute substantially toward the measured Σ, hence preventing an accurate appraisal of layer-by-layer fluid saturation. This behavior makes the interpretation of sigma logs challenging in economic pay zones. To date, simulating PNC measurements

1

ABSTRACT Pulsed neutron capture (PNC) logs are commonly used for formation evaluation behind casing and to assess time-lapse variations of hydrocarbon pore volume. Because conventional interpretation methods for sigma logs assume homogeneous formations, errors may arise in thinly-bedded formations when appraising petrophysical properties of hydrocarbon-bearing beds. There exist no quantitative interpretation methods to account for shoulder-bed effects on PNC logs acquired in sandshale laminated reservoirs. Due to diffusion effects between dissimilar beds, sigma logs acquired in such formations do not obey linear mixing laws between the sigma responses of pure-sand and pure-shale end members of the sedimentary sequence. We introduce a new method to rapidly simulate sigma logs in thinly-bedded formations. The method makes use of late-time, thermal-neutron flux sensitivity functions (FSFs) to describe the contribution of multilayer formations to the measured capture cross section. It includes a correction procedure based on diffusion theory that adapts a homogeneous, base-case FSF to simulate the response of vertically heterogeneous formations. Benchmarking exercises indicate that our rapid simulation method yields differences smaller than 2% with respect to PNC logs simulated with rigorous Monte Carlo methods for a wide range of geometrical, petrophysical, and fluid properties. The second part of the paper develops an inversion method to reduce shoulder-bed effects on PNC logs and estimate layer-by-layer capture cross sections based on the previously defined FSFs. We successfully test the estimation method on synthetic examples that include a variety of bed-thickness configurations. Inversion consistently improves the vertical resolution and sigmadefinition of PNC logs across beds thinner than 1.5 ft. Testing on field data confirms the efficiency, reliability, and stability of the inversion procedure. Our fast simulation/interpretation algorithm inverts sigma logs over 100+ ft. intervals in less than 4 minutes of CPU time, and is therefore suitable for joint petrophysical interpretation with other open- and cased-hole logs.

Trademark of Lane-Wells Co. (now Baker Hughes).

1

The assumed formation is a homogeneous. the large CPU time associated with Monte Carlo techniques renders them impractical to implement inversion methods in either the assessment of complex lithologies or the joint quantitative petrophysical interpretation with other open. Comparison against experimental results reported by Preeg and Scott (1986) indicates less than 2% error for the Longhorn PNC tool for MCNP-simulated measurements of formation thermal decay time. casing. 1986).SPWLA 51st Annual Logging Symposium. we begin the simulations with basic FSFs for homogeneous formations and subsequently account for heterogeneities by adapting the FSFs to the presence of layer boundaries. Σ obeys a simple mixing law.u. compared against experimental results τexp. thereby enabling the rapid numerical simulation of PNC logs with no further need of Monte Carlo methods. The fast simulation of PNC measurements. In terms of capture cross section. away from the source. where each component is weighted by its relative volumetric concentration (Clavier et al.9 73.77 -0. ΔΣ is the corresponding absolute error on capture cross section.28
Δτ
ΔΣ
[c. shale and fluids) are homogeneously distributed in space.2 213.02 -0.] 0. exhibits a measurement response similar to commercial tools. PNC measurements can be substantially affected by shoulder beds to the extent that it is not possible to precompute such FSFs for all multi-layer cases.7 250
CwBH
[µs] 269.u. 1971b). in some instances it may be approximated by thermal neutron diffusion. Synthetic and field examples of application appraise the accuracy and reliability of our fast simulation and inversion method.5 214.7 73. We also calculate the corresponding multi-layer FSFs with MCNP.2 270. 7-in. To benchmark our FSF-correction method. we generate synthetic PNC measurements using the Monte Carlo N-Particle3 (MCNP) code (X-5 Monte Carlo Team. unlike neutron and density measurements. boreholes filled with salt water.09 0. 30. which we compare against the corrected FSFs.31-c.31 0. Because the purpose of this study is to analyze PNC logs acquired in thinly-bedded formations. Since the earliest days of PNC logging it has been recognized that detecting neutroncapture gamma rays is preferable to detecting neutrons (Wahl et al. with 10-in.7
CwF
[kppm] 0 36. However.
Table 1: Formation thermal decay time at the near detector calculated with MCNP simulations of the Longhorn PNC tool τsim. this represents an error below 0.27 -1. for several values of salt concentration of formation water (CwF) and borehole fluid (CwBH). Even though the underlying physics is governed by the Boltzmann transport equation. 2003). 1970).7 265.2 212. in conjunction with inversion techniques.
2
.9 36.24 0. Instead. which emits fast neutrons through casing and cement into the formation. However.7 73. in this paper we develop a new method for the fast simulation of PNC measurements using Monte Carlo-derived flux sensitivity functions (FSFs). Trademark of Los Alamos National Laboratory.9 36.or cased-hole borehole measurements. τ.. respectively.30 -0. However.6
τsim
[µs] 274 270 270 214 214 212
τexp
[%] -1..9 250 0 73. There also exist finite-difference numerical simulators of PNC measurements based on diffusion theory. Δτ is the relative error between τexp and τsim.05 -0. 2010
with rigorous Monte Carlo methods remains the standard way to quantify PNC tool responses.5 and 21 in.04 0. June 19-23.75 0. In similar fashion to Mendoza et al. The excellent agreement with the calculated Σ for the two methods validates our FSF correction strategy.’s (2010) approach to numerically simulate neutron and density measurements. we exclusively focus our attention to the counting rate at the short-spaced detector due to its better vertical resolution. Σ. opens the possibility of quantitative petrophysical interpretation of sigma logs: we explicitly reduce shoulder-bed effects on PNC logs and improve the estimation of layer-bylayer capture cross sections. This is the reason why we equip the Longhorn tool with two scintillation crystals.06
The tool includes a 14-MeV accelerator source.and 10in. one can write the total capture cross section in a clay-free siliciclastic or carbonate formation as
Trademark of Schlumberger.
[kppm] 36. environmental effects may cause sizable simulation errors: Hamzah (1996) reported errors on the simulated counts above 30% in the cases of 6. This is why the departure curves for TDT2 determined by Locke and Smith (1975) with a diffusion-based numerical simulator were eventually substituted with Monte Carlo-based numerical simulation techniques (Preeg and Scott. LIMITATIONS OF LINEAR MIXING LAWS When all the components (rock matrix.
2 3
TOOL CONFIGURATION We invoke a generic Longhorn PNC tool which. For instance.4%-porosity sandstone. which are located 13. as shown in Table 1. at the near detector (the data processing technique is explained in a subsequent section of this paper). borehole.

ΔΣ is the error between ΣMCNP and ΣLinear. homogeneous formation. This result is due to the inability of the linear mixing law to account for the spatial variations of neutron/photon flux in thinlybedded formations. Consequently. oil-bearing sand layer.5-c.u.] 12. there exists a threshold above which heterogeneities begin to impact the measurement.5
OIL-BEARING SAND
-0. neutrons begin to flow across them again as though they were homogeneously mixed. hence neutrons at thermal energies are the most likely to be absorbed. June 19-23. when laminae become thin enough. Σ ( r0 ) ⎤ Σ ( r0 ) . Σma.3-c. FORMULATION By definition.3
ΔΣ
[%] 52. This behavior indicates that neutrons have a specific spatial variation in layered formations.
(3)
Case
1 2
ΣMCNP
[c.] 23.3
ΣLinear
[c. 2 (right) consists of the same shale layer below a 6. Figure 1 describes two cases of study that
Σ [c.13 ft.u.5 15. compared against estimations from the linear mixing law ΣLinear. For example. Σ(r0)] is the thermal-neutron flux sensitivity function. so that the formation is still a homogeneous mixture. 2010
Σtot = (1 − φ ) Σ ma + φ (1 − S w ) Σ h + φ S w Σ w . at the detector located at position r.5 68. respectively.
(2)
where Csh is volumetric concentration of shale. Thus. which does not solely depend on the relative volume of each component. Orange-colored lines at 0 and 1.6
where r0 is an arbitrary position vector that spans the whole space. r0 . For instance.3 37. The lower the neutron energy. Case No. hydrocarbon. and water. The total capture cross section Σ(r).u. Haley (1995) reported that laminae thicker than 10 cm. describe the positions of detector and source.] 36. neutrons may be captured more promptly in high-Σ layers that act as sinks of neutrons.u. and Σsh and Σss are shale and non-shale component capture cross sections. For the case of an infinite.6
ΔΣ
[c. In shaly formations.]
-0. 1 (left) consists of a 48. sigma obeys a linear mixing law similar to equation (1) only if the shale is uniformly distributed. Σ(r0) is the capture cross section at position r0.5
1.
(1)
where φ is porosity. it follows that the FSF must obey the following normalization:
3
. which represent the importance of each layer toward the effective Σ. which is given by the thermal-neutron flux sensitivity function. Table 2 shows the discrepancies associated with the assumption of a linear mixing law between source and detector. However. ⎣ ⎦
16
0 Relative Depth [ft]
0
24
0. 2. the relative contribution of shale layers to the measurement is lower than expected based on their thickness in both cases.5
48
Figure 1: Case No.8 22. for which one can write
Σtot = (1 − Csh ) Σ s + Csh Σ sh .5
GAS-BEARING SAND
8
emphasize the shortcomings of equation (2) for heterogeneous formations. Sw is water saturation. Comparison against MCNP simulations of the Longhorn PNC tool indicates up to 68% error. These weights. which describes the relative contribution of all the Σ(r0) at position r. We approximate this physical interpretation of Σ with the following Fredholm integral equation of the first kind:
Σ ( r ) ≈ ∫ dr0 FSF ⎡r. as illustrated by Case No.5
0. Σh and Σw are the capture cross sections of the rock matrix.2-c. Consequently. The discrepancy is especially significant when the materials are dissimilar. is the summation over space of all the cross sections Σ(r0) weighted by their flux contribution toward the detector. are likely to cause measurement biases in a 34%-porosity.u. the effective Σ in a layered formation becomes a weighted average of the Σ of all the layers that interact with the neutrons. Table 2: Formation capture cross section calculated with MCNP simulations of the Longhorn PNC tool ΣMCNP. and FSF[r. This situation occurs in the case of structural and dispersed shale. r0. 100kppm salinity sand-shale laminated formation. while they may survive longer in low-Σ layers. monitoring the population of thermal neutrons provides one with the proper importance function. thereby causing deviations from the linear model.u. In thinly-bedded formations. the capture cross section Σ quantifies the ability of a material to capture neutrons.SPWLA 51st Annual Logging Symposium. shale layer overlaid by a 14. respectively.u. respectively.5
32
SHALE 1 1
SHALE
40
1. the more likely capture phenomena will take place. gas-bearing sand layer. are no longer equal to the relative volume of each lamina. Σ(r0) is constant and independent of r0.

we approach the problem with a nonlinear. approximate numerical method described above. τ BH ⎠ F ⎝ ⎝ τF ⎠ where t is time. 1995). which quantifies the relative contribution from all the Σ(z0) at depth z. Also.. This approach enables the dynamic fit of both borehole and formation responses. (2005) reported that the traditional signal processing technique based on fitting the data with two exponential decays is reliable to separate borehole and formation responses. Generating these specialized FSFs using MCNP simulations requires time-consuming computations. Figure 2 compares the thermal-neutron FSFs in homogeneous and heterogeneous formations for the same cases displayed in Figure 1.. radiallyaveraged. Σ ( z0 ) ⎤ Σ ( z0 ) . Instead. which describes the transport of neutrons. we use the Monte Carlo code MCNP that reproduces the transport of neutrons and photons from the emission of fast neutrons and their slowing down through numerous collisions with atomic nuclei. Furthermore.. quantifying these changes is non-trivial. and FSFLT[z. Σ(z0)] is the normalized. it is crucial to ensure that the outcome is reliable. equation (3) is independent of energy in our problem because all neutrons exhibit thermal energies. we improve the simulation statistics via several variance reduction techniques. Σ(z0)] for all depths. which reaches its maximum close to the detector position. 2010
∫ dr
0
FSF ⎡r. One of them uses iterative mesh-based. r0 . 1988). which entails longer CPU times. late-time. This method tracks particles in important spatial regions. From MCNP results we obtain gamma-ray counts at the near detector. We observe that the transition between two distinct materials causes changes in the FSF that can be anticipated on qualitative grounds. We determine the late-time importance of thermal neutrons to ensure cancellation of undesired early-time borehole effects. Such a strategy ultimately constructs an estimated space-and-energy importance function: the flux sensitivity function. The flux undergoes a contraction in the high-Σ shale and a relaxation in the lower-Σ sand layer. Rapid Simulation of PNC Logs Equation (5) is the basis for simulating sigma logs. hence the need of a faster method. M(t). However. energy-dependent weight windows in conjunction with a time-splittingand-Roulette technique for population control purposes. Besides increasing the number of particle histories. such linear regressions may not correctly discriminate closely-spaced signals. For each MCNPsimulated measurement we obtain the corresponding FSF as a byproduct of the weight-window generation. These three assumptions allow us to remove the dependence of the solution of equation (3) on the radial and azimuthal directions. to their capture and subsequent emission of neutron-induced gamma rays that are counted at the detector. z. τBH and τF are the thermal decay times of the borehole and formation signals. which are similar to those acquired in the field. based on the Levenberg-Marquardt algorithm (Marquardt. Because MCNP is a random sampling method. The FSF in a homogeneous formation is a bell-shaped. Monte Carloderived synthetic PNC logs serve as benchmark in the analysis. Σ(z0) is the capture cross section at depth z0. It requires the thermal-neutron flux sensitivity functions FSFLT[z. hence making it possible to neglect invasion effects in the simulation. such as a high-Σ formation and an average saline borehole. energy and/or time. z0. Monte Carlo methods need no averaging approximations in space. Even though some of the recently comercialized PNC tools resort to a database of laboratory measurements (Plasek et al. MCNP-Simulated Measurements Because the Boltzmann equation. June 19-23. 2008). METHOD Throughout this study. respectively. Σ ( r0 ) ⎤ = 1 . Unlike deterministic methods. ⎣ ⎦
(4)
In addition. Although NMR-like inversion techniques have been recently introduced to extract τF from M(t) (Flaum et al. we express the photon counts as a function of time. the problem is now describable with the vertical direction only. 1963). Morris et al. cannot be solved analytically in a borehole-and-formation geometry (Steinman et al. z0. while ascribing lower significance to particles in unimportant regions. we assume non-dipping beds. thermal-neutron flux sensitivity function. ⎣ ⎦
(5)
where z0 is an arbitrary location that spans the vertical direction. To that end. two-exponential-decay fit.
Thus. Even though the gas-bearing sand layer has a lower value of Σ than the oil-bearing layer. at which the PNC tool measures Σ(z).SPWLA 51st Annual Logging Symposium. smooth function. We finally express the total capture cross-section Σ at depth z as
Σ ( z ) ≈ ∫ dz0 FSFLT ⎡ z . All the Longhorn-PNC-tool measurements reported throughout this paper originate from the application of that technique. we assume that fluid reequilibrium takes place shortly upon setting casing. z0 . as ⎛ t ⎞ ⎛ t ⎞ M ( t ) = ABH exp ⎜ − (6) ⎟ + A exp ⎜ − ⎟ . we generate PNC logs using both Monte Carlo simulations and the fast. and ABH and AF are the corresponding amplitudes. the dynamic variations are greater
4
. we resort to numerical solutions.

This behavior causes potential problems in low-porosity and gasbearing formations.
Σ [c. to 37. Synthetic Case No. while Σ exhibits significant variations. Because neutrons scatter longer before absorption. 1 and 2 could not be described by the interpolation of the FSF for Formations Nos. Assuming that diffusion theory is valid within our domain of application simplifies the Boltzmann transport equation.u.13 ft. bed thickness.5
32
SHALE 1 1
SHALE
40
1. Relevant details about this correction procedure. which is correlated to hydrogen index. (2010) for neutron porosity.20 18. Yet. we use the geometric center of the FSFs to define the location of the measurement point. fluid.73 c. we develop a rapid. yet accurate. Orange-colored lines at 0 and 1. FSFs are shifted toward the detector.u. as shown in Figure 2. we use equation (5) to determine Σ at each tool position.5 13. Tables 4 and 5 summarize the lithology. oil. 2).0 16.5 0
1
48
1D Normalized Response [cps/cps]
Figure 2: Comparison of MCNP-generated thermal-neutron FSFs for the layered formation (solid red line) against the FSF for a homogeneous formation (dashed black line). thereby generating field-like PNC measurements that serve as reference. the FSF associated with a layered sequence composed of Formations Nos. For depth-matching purposes. FSFs change as little as 0. the underlying assumptions. we determine the late-time thermal-neutron FSF for a homogeneous sand layer saturated either with fresh water (Formation No. diffusion-approximation correction procedure that yields thermal-neutron FSFs for any layered formation.5
GAS-BEARING SAND
8
16
0 Relative Depth [ft]
0
24
0. carbonate and shale layers saturated with gas.06 0. are not applicable to PNC measurements.3 14.5 0 1
1.5
0. and geometrical properties assumed for this case of study.18 0.54% when Σ varies from 10. Ls is comparable.93 c.7
Ls
[c. changes in Σ cause negligible effects on FSFs for a similar value of Ls. The resulting method would be cumbersome and difficult to implement.40 10. Upon correcting the FSFs for formation heterogeneities. as well as comparisons between MCNPcomputed and diffusion-corrected FSFs are included in the Appendix. 1 (left) and 2 (right). Consequently. especially at high values of porosity.] 5.73 37. Slowing-down length Ls and intrinsic capture cross section Σint were calculated with SNUPAR (McKeon and Scott. which consists of successive sandstone. respectively.93
Σint
ΔFSF
[%] 0. We import the model into MCNP to simulate Longhorn-tool measurements along the cased borehole every 0.
[%] 5 15 25 35
φ
Formation
1 2 1 2 1 2 1 2
[cm] 20. June 19-23.
For each case.4 14.54
5
.25 ft. Figure 3 compares it against the MCNP reference log. Constructing a comprehensive library of FSFs in heterogeneous formations is not practical either. 2010
for the latter owing to the impact of the tool position with respect to the boundary.97 28. 0.4 20.u.33 seconds of CPU time were necessary to achieve less than 2% error on the Σ log.43 9. 1 and 2.SPWLA 51st Annual Logging Symposium. Table 3: Impact of Σ on FSF for different values of porosity φ.86 8.]
-0. In other words. In each case. and tool position with respect to the layered formation are all variable parameters). Table 3 compares MCNP-calculated.5
OIL-BEARING SAND
-0. Thermal-neutron FSFs in homogeneous formations are primarily influenced by the slowing-down length Ls. 1 We examine the capabilities of the method with a synthetic formation. thermal-neutron FSFs for homogeneous formations saturated with fresh and salt water. one should compute as many FSFs as there are heterogeneous cases (Σ. fresh water or saline water. describe the positions of detector and source. thereby allowing us to account for formation heterogeneities analytically. from which we numerically simulate the PNC log. On the contrary.6 13. We use equation (5) in conjunction with the FSFcorrection procedure to rapidly simulate the PNC log for this case.32 7.7 16. for Cases Nos.14 0. Instead. the FSF associated with a homogeneous formation cannot be used to evaluate correctly the substantial changes undergone by PNC logs in heterogeneous formations.u. Iterative refinement techniques such as the one reported by Mendoza et al. 1) or with 200-kppm salt (NaCl) water (Formation No. 1989). respectively. ΔFSF quantifies the relative error on the FSF.

]
50
Figure 3: Comparison of numerically simulated (solid red line) and measured Σ logs (solid green line. well geometry and sampling rate for Synthetic Case No.5 1.u.3 0. The vector of data residuals is given by
⎡ e1 ( x ) ⎤ ⎡ d1 ( x ) − d10 ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0 e ( x ) = ⎢ en ( x ) ⎥ = ⎢ d n ( x ) − d n ⎥ = d ( x ) − d 0 .
6
.8
[%] 5 12 – 20 – 10 15 20 25 25
φs
[%] – – 10 – 10 10 10 – – –
φsh
Csh
[] 0 0 1 0 1 0.8 2. ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎢e ( x ) ⎥ ⎢ d ( x ) − d 0 ⎥ N ⎦ ⎣ N ⎦ ⎣ N
#9
40
16
16
# 10
45
(8)
10 20 30 40 Capture Cross Section Σ [c.SPWLA 51st Annual Logging Symposium. calculated with MCNP) for the model shown in dashed blue line.12 13. infinite-extent formation in the presence of the borehole).40 1 1
[kppm] 150 250 200 50 200 100 150 50 220 0
CwF
Hydrocarbons
Gas Oil – Oil – Gas Oil Oil – –
[c. e(x) is vector of data residuals.25
Units
g/cm3 g/cm3 kppm c.u.u. nature of the hydrocarbons.2 1. June 19-23.86 9.40 1 0. formation water salt concentration CwF. shale porosity φsh. and d(x) is the N-size vector of sigma values numerically simulated with the fast-forward simulation method applied to x.12 15. non-shale porosity φs.65 1 0. water saturation Sw.
#1
2
2
Σ
15
Variable
Formation oil density Formation gas density Borehole fluid salt concentration Borehole fluid capture cross section Borehole diameter Casing diameter Sampling rate
Value
0. layer thickness 2H.
where d0 is a N-size vector of PNC measurements.3 2.
2 2
10
10
#6
(7)
12
12
#7 #8
35
14
14
where x is a vector containing the unknown layer-bylayer values of Σ.2 10 7 0.75 9.50 0.017 225 111.u. The procedure consists of minimizing the quadratic cost function
C (x) = e (x) + λ 2 x .86 11.
This exercise confirms that our method accurately and rapidly simulates PNC measurements with no further need for rigorous Monte Carlo methods.63 8.5 2.50
Σdif
0
Σ Σ
Model Measurement Forward Simulation
0
Table 5: Summary of the assumed hydrocarbon densities.
#
1 2 3 4 5 6 7 8 9 10
Lithology
Limestone Dolomite Shale Sandstone Shale Shaly Sand Shaly Sand Sandstone Sandstone Sandstone
2H
[ft] 2.40 0.0 1. from top to bottom of the formation.] 12.6 0.96
Σint
[c.3 0 0 0
Σ [c.32 14.00 45.41 19. accurate simulation method introduced above to estimate layer-by-layer values of Σ from inversion of PNC measurements.0 1.38 30.u.41 13. borehole fluid properties.5 1. The entry en(x) designates the n-th data misfit at the n-th sampling depth. 2010
Table 4: Summary of the assumed lithology. 1.60 41.70 0. in in ft
#2
4 4
20
#3
6 6
#4
25
Relative Depth [ft]
8
8
#5
30
Inversion of PNC Logs We now make use of fast.00 31.] 8.85 12.82 0.04 45. and λ is a stabilization parameter calculated using Hansen’s (1994) L-curve technique. volumetric concentration of shale Csh. intrinsic capture cross section Σint (from SNUPAR) and expected measured capture cross section Σdif (from MCNP simulations of the pure.]
Sw
[] 0.99 17.38 41.81 22.

alternated with high-Σ shale layers (illite). The numerically simulated log agrees very well with the measured log along depth intervals within shale. Δx. oilbearing and water-bearing layers.3 48. 1963). thickness 2H. Iterative Algorithm Figure 4 summarizes the workflow adopted in this study to appraise unknown layer-by-layer values of Σ.
Figure 5 displays the numerically simulated PNC log and the final estimates of layer-by-layer Σ values.and gasbearing sand layers. Subsequently. 2.SPWLA 51st Annual Logging Symposium.4 48. June 19-23. and the layer-bylayer model. Therefore. we compute the Jacobian matrix numerically and subsequently determine the incremental variation of model parameters by minimizing the quadratic cost function defined in equation (7). Synthetic Case No. we construct a nonlinear system of equations that yields the model parameters x. 2 This synthetic case is intended to reduce shoulder-bed effects on the interpretation method. Additionally.
Table 6: Summary of the assumed lithology.00
[%] 35 25 35 30 35 35 35
φ
S
[] Sw = 1 Sw = 0. The average value of the log is a good alternative choice. which subsequently become the new initial guess to populate the layered model. It consists of an 8-ft. Table 6 summarizes the lithology and fluid properties assumed for this case. Borehole fluid properties. rendered by equation (5).80 1. 2010
Although equation (5) embodies a linear relationship between the measurements Σ(z). well geometry information as well as the hydrocarbon densities are the same as those reported in Table 5. formation water salt concentration CwF.2
Σint
Measured Σ log d0
Match?
NO
YES Final layered model
Figure 4: Flowchart of the fast-forward and inversion method for sigma logs. SYNTHETIC EXAMPLES We now quantify the robustness of the developed method for the rapid simulation and inversion of PNC logs based on two examples of synthetic formations.2 14. We also compare the measured log. The agreement (match) between numerical simulations and measurements is ascertained from the increment in model parameters from one iteration to the next. the layered model is initialized with values of Σ derived from the measured log: successive minima and maxima of the smooth log provide a good first guess. oil. total porosity φ.
From the corresponding FSFs and the measurements. which we verify against a pre-specified threshold.] 48.8 Sw = 1 Sw = 1 Sw = 1
[kppm] 215 200 215 200 215 200 215
CwF
[c.47 48. At every iteration. siliciclastic sequence with water-. The method first detects bed boundaries from the inflection points of a master log (gamma-ray. and intrinsic capture cross section Σint (from SNUPAR) for Synthetic Case No. Convergence of this iterative procedure is diagnosed by comparing the variations of model parameters.2 28.30 0.1 Sg = 0.20 1. otherwise we use PNC logs).9 Sw = 1 Sw = 0. density or resistivity logs if open-hole data are available.15 0. d0. the gas region (layer #2) still shows a small mismatch from 1 to
7
. However. against the numerically simulated log.2 So = 0.2 6.
Initial layered model x
Correct FSFs Modify model x
Simulate Σ log d(x)
#
1 2 3 4 5 6 7
Lithology
Shale Sand Shale Sand Shale Sand Shale
2H
[ft] 1. we describe the mapping between the model and the data with a nonlinear system of equations that we solve using the Levenberg-Marquardt method (Marquardt. Σ(z0). FSFs are not in general constant for all N depths in a heterogeneous formation. from one iteration to the next against a pre-specified threshold.80 1.80 0.u. from top to bottom of the formation. d(x). saturation S. we evaluate the uncertainty of the estimated layer-by-layer values of Σ from the 95% confidence interval of the covariance matrix and plot it as error bars together with the inverted model.

Inversion
Table 8: Summary of the assumed values of total porosity φ. Likewise.89 -1.6 10. difference but only an increase of 2 c.12
ΔΣ
[%] 6.
Table 7: Final layer-by-layer capture cross sections Σest estimated upon matching the measured log with fastforward simulations. formation water salt concentration CwF. the inversion converges to final results in seconds of CPU time.1 c.52 9. thus why the relatively low sensitivity of the measurement to high-Σ layers. Estimated layer-by-layer Σ values (dashed red line) show good agreement with the model (dashed blue line).2 So = 0. especially in thin beds. higher-Σ layers contribute less than expected toward the measurement.31
Σdif
8
.2 33..24 2. ΔΣ is the error between Σest and Σdif. 2010
0
Σ Σ
Model Measurement Initial Guess Forward Simulation Inversion
0
Σ [c.SPWLA 51st Annual Logging Symposium.6 32. Table 8 summarizes the petrophysical information about this case.77 1.u. When using the average value of the measurement log (equal to 33.9 49.95 18.u. once the tool logs the thick shale layers #1 and #7.38
Σint
[c. The inversion algorithm converges independently of the initial guess: initializations from the smooth log or from parsimonious guesses (constant average value of the log as well as shale assumed everywhere) lead to the same outcome.u. June 19-23. This behavior takes place because the rapid simulation method does not gradually shift the non-gas FSF to a gas-related FSF when the logging point is close to the gas-bearing sand.8
[kppm] 100 100
CwF
[c.u.9 52. δΣ95 quantifies the uncertainty in the appraisal (95% confidence interval).6
Σdif
[c. layer #3 shows a peak at 30.79 10.84 13. 3 This synthetic case investigates the influence of bed thickness on the inversion of PNC logs as well as the impact of bed-boundary perturbations.9 6.7 50..u. the low-Σ sands no longer impact the measurement and we infer the model values with less than 4% uncertainty.] 3.7 50. Figure 6 shows the numerically simulated PNC log and the final layer-by-layer estimates of Σ. higher than 19 c. from top to bottom of the formation. We initialized the inversion with a parsimonious guess based on the average value of the measured log (dashed green line).u.2 49. It consists of a turbidite sequence that alternates oil-saturated sand beds and high-Σ shale layers (chlorite). Black error bars describe the 95% confidence interval. well geometry information as well as the hydrocarbon density are the same as those reported in Table 5. saturation S.67 5.13 0.7 10.u. despite the implementation of a correction specific for the presence gas (discussed in the Appendix). The uncertainty on Σ across shale layers is larger than
Synthetic Case No.5 ft.u.u. below the actual value of 49.10 0.0 17. Consequently.6 18.60 0. Σdif is the expected measured capture cross section (from MCNP).47 7. However. 2.
Lithology
Shale Sand
[%] 10 20
φ
S
[] Sw = 1 Sw = 0. the transition from the water leg (layer #6) to the overlying shale (layer #5) does not show the actual 16. This behavior is due to the properties of the FSFs in heterogeneous formations: they decrease in high-Σ layers and increase in low-Σ layers. For instance.] 49.35 2. 3.) as initial guess. for Synthetic Case No.7-c.e.u.9 5.]
1
Σ Σ Σ
1
#1
15
2
2
20
#2
3
Relative Depth [ft]
3
25
#3
4 4
30
across sand layers. as well as the 95% confidence interval for a perturbation of the measured log of 2%.] 52.u.
1 2 3 4 5 6 7
4.44 -0.26
δΣ95
[%] 2.67
#7
8
0
10
20 30 40 50 Capture Cross Section Σ [c.1 c.u.6 49.23 1.]
60
8
Figure 5: Comparison of numerically simulated (solid red line) and measured Σ logs (solid green line).] 40.u.6 c. intrinsic capture cross section Σint (from SNUPAR) and expected measured capture cross section Σdif (from MCNP) for the sand and shale components of Synthetic Case No.] 39. Borehole fluid properties.19 0.7
Σest
ΔΣ
[c. Shoulder-bed effects and spatial smoothing may cause significant misinterpretations. i. Table 7 describes the errors between estimated and actual model parameters.30 2.
#4
5 5
35
#5
6 6
40
#6
#
7 7
45
Lithology
Shale Sand Shale Sand Shale Sand Shale
[c.10 3.

and change the bed thickness by +/.4 13..u. Shift -0.8 1.94
ΔΣ
[%] -0.7 13. as small as +/. Σ increases up to 18.27 6. Reference case is when all bedboundary locations are exactly known.4 41.] -0. leading to layer-by-layer Σ values that are up to 40% in disagreement with original values.25 ft.79 3. changes in bed thickness bias their evaluations from the measured log. Even though the composition of sand layers is the same throughout.8%) in the estimated Σ values.25 ft. Such a behavior would cause a significant misinterpretation in the saturation of thin sand layers. due to shale-sand diffusion effects. The inversion was initialized with a parsimonious initial guess based on the average value of the measured log (dashed green line).u. For the same reasons as those reported in Synthetic Case No.u.2 36.15 -2. 3. +/. When using the average value of the measurement log as starting point (equal to 21. Estimated layer-by-layer Σ values (dashed red line) show good agreement with the model (dashed blue line). The relative error on the sigma log is reported for each case.8 0.2 38. We subsequently shift all boundaries by 0.14 -6.8 0.]
45
11
Figure 6: Comparison of numerically simulated (solid red line) and measured Σ logs (solid green line). While the 1.25 ft.). in the center of the 0.u.77 -3. or -0. cause up to 13. of eleven different natural rock formations.98 -0.2 0.1% error in the numerical simulation of the sigma log.] 13.9 -1. error for sand layers.
0
Σ Σ Σ
Model Measurement Initial Guess Forward Simulation Inversion
Table 9: Final layer-by-layer capture cross sections Σest estimated from the matching of the measured log with fastforward simulations.SPWLA 51st Annual Logging Symposium.31 c. limestone and dolomite. All formations are saturated with 35-kppm salt water.
6
#5
6
7
#6
7
30
#7
8 8
#8
35
9
9
#9
Bed Boundaries
Reference Shift +0. The corresponding inversion results are no longer reliable. thereby emphasizing the importance of bed-boundary selection in the interpretation method.25 ft. Table 10 describes the results obtained for this sensitivity analysis.55 0.6 0.47 18.25 ft.10 -1.56 1.8
Σest
ΔΣ
[c.47
Error
[%] 1.75 c.31
δΣ95
[%] 4.71 -10.9 0.27 15.0 13.07 13.73 8.3 13.14 9.67 15. 2H stands for layer thickness. cause a mismatch in the simulated PNC log of up to 15.5-c.02 -5.07 0. 2. Additionally.9-ft.8-ft. changes in layer thickness.u.8 1. 2010
matched the PNC log within 1.]
1
Σ Σ
1
#1
15
2
2
#2
3
20
3
#3
4
Relative Depth [ft]
4
5
#4
5
25
We now examine the impact of bed-boundary perturbation.51 2. layers #7 and #9.83
0
Σ [c.01.01% error. layer #1 displays the expected value of 13. from top to bottom of the formation.46 -0.
#
1 2 3 4 5 6 7 8 9 10
Lithology
Sand Shale Sand Shale Sand Shale Sand Shale Sand Shale
2H
[ft] 1.01 9. δΣ95 quantifies the uncertainty in the appraisal (95% confidence interval)..40 -2.8 1.u. the loss of measurement sensitivity in shale layers results in a mismatch (up to 10.9%) and a larger uncertainty (up to 9. Slight perturbations. June 19-23.77 9.0.55 c.87 4. Black error bars quantify the 95% confidence interval. Table 9 describes the errors between estimated and the actual model parameters.28 5. 3.5
[c..u.25 ft. for Synthetic Case No.85 -0.12 ft.
Iterations required
13 12 10 9
CPU time
[s] 19.07 0.8 13.09 -4. inversion yields layer-by-layer Σ values that match the actual values within less than 0.0. The threshold on the variations of model parameter used to diagnose convergence is set to 0. such as a depth shift of all bed boundaries of 0. comprising successive beds of sandstone.
FIELD EXAMPLES Field Case No.08 0. ΔΣ is the error between Σest and Σdif (reported in Table 8).2 39.25 ft. 1 This field case is an open-hole test pit that includes 30 ft. Low salinity and absence of shale give rise
9
.1
10
10
# 10
40
11 10
15 20 25 30 35 40 Capture Cross Section Σ [c. as well as the 95% confidence interval for a 1% perturbation of the measured log.7 38.36 6.81 8.7%.0.
Table 10: Impact of bed-boundary perturbation on convergence for Synthetic Case No.25 and -0.8 0.u.

7
2.8 21. for Field Case No.25 0.64 4.3 34.. Table 11 summarizes the corresponding properties of both formation water and borehole fluid. and bed thicknesses. which is the average value of the measurements in this interval. estimated capture cross section Σest. porosity φ.u. The right-hand panel of Figure 7 shows results obtained for the numerical simulation of PNC measurements for this layered formation.33 2. June 19-23. Table 12 describes the estimated layer-by-layer Σ values together with the 95% confidence interval for a 1% perturbation of the measured log.6 13.u.u. in ft
Table 12: Summary of the assumed lithology.1 13.u. and corresponding 95% confidence interval δΣ95.81 3.
Table 11: Summary of the assumed formation water and borehole fluid properties for Field Case No. c.5 18.15
0
-0.] 15.65 2.9 17. the inversion exercise validates the capabilities of our method to simulate and interpret PNC measurements. respectively.
Variable
Formation water capture cross section Borehole fluid capture cross section Borehole diameter Sampling rate
Value
34. 1. Neutron and density logs are also included in the left-hand track.47 2.u.12 1.45
ρb [g/cm ]
3
2.79 1.03 2.3 12. from top to bottom of formation.3 8.6% error in 9 seconds of CPU time. 1. while Table 12 describes the associated lithology.5
Units
c. as shown in the case of layers #3 and #4 of the test-pit model. 2010
to low values of Σ.3 0
0. bed thickness 2H.58 6
[%] 25 0 19 0 17 19 0 26 0 8 0
φ
[c.16 4.17 2.0 14.07
φN [ ]
0. Inversion matches the sigma log with less than 0. together with black error bars that quantify the 95% confidence interval.2
Σest
δΣ95
[%] 2.14 1. Even though this field example does not include either casing or high-Σ shales.2 2.95 10
12
14 16 18 Capture Cross Section Σ [c.4 10.SPWLA 51st Annual Logging Symposium. The dashed blue line describes the initial guess.
#
1 2 3 4 5 6 7 8 9 10 11
Lithology
Limestone Marble Sandstone Dolomite Limestone Sandstone Marble Chalk Dolomite Sandstone Granite
2H
[ft] 3 1 1 1 3.92 3 3 2.15
5
10
Relative Depth [ft]
Σ
Measurement Initial Guess Forward Simulation Inversion
15
Σ Σ Σ
20
25 2.]
20
22
Figure 7: Comparison of numerically simulated (solid red line) and measured (solid blue line) PNC logs (right-hand track) for Field Case No. Estimated layer-by-layer sigma values are indicated with dashed red lines.5 12. porosity. between 10 and 19 c.80 5. 1. The left-hand track displays thermal neutron decay porosity (green) and density porosity (red) logs. Inversion does improve the assessment of Σ values in beds thinner than 1.5 ft.
10
.1 13. We note that in this case inversion does not improve (sharpen) PNC logs across thick beds as the estimated Σ values coincide with the measurement log.

June 19-23. of a shale-sand laminated formation.7 0. Tracks 1 through 3 show the open-hole gamma-ray. interbedded with associated overbank and floodplain siltstone and shale deposits (Pranter et al. The right-hand panel of Figure 8 shows the results obtained from the numerical simulation of the sigma log for 130+ ft. track 2 shows the arrayinduction resistivity logs.]
25
28
Figure 8: Comparison of numerically simulated (solid red line) and measured (solid blue line) PNC logs (track 4) for Field Case No. 2008. late Cretaceous to early Paleogene. 2009).
Table 13: Summary of the average formation and fluid properties assumed for Field Case No.30 0.2
φ [] N
X540
0 10
13
16 19 22 Capture Cross Section Σ [c. Table 13 summarizes the formation properties for this field example. The reason for this behavior is two-fold: first.5
Units
% kppm kppm Ω.m ft
Bed boundaries were determined simultaneously from of all the available open.4 0.and cased-hole logs. 2 This field case targets onshore. track 3 shows the neutron porosity log. Estimated layer-by-layer sigma values are indicated with dashed black lines. Weijers et al. steel-cased. 2010
Field Case No.5 5.000 ft.
Porosity φ Formation water salt concentration CwF Standard deviation for CwF Formation water resistivity at 75ºF Sampling rate
Variable
Value
6 19. Five iterations in 4 minutes of CPU time were necessary to secure an acceptable match between simulated and measured logs with less than 1% error. of lenticular. 2.m] 100 0. The dashed blue line describes the initial guess. Track 1 displays the gamma-ray log. 2. The formation consists of more than 2. they do not improve the Σ values already defined by the measurements. Even though inversion results obtained for this example are in agreement with the PNC log. fluvial point-bar and crevasse-splay sandstone deposits.u..SPWLA 51st Annual Logging Symposium. which is the average value of the measurements in this interval.. The relatively low values of porosity (6%) and salinity (19.5 kppm of NaCl equivalent) prevent reliable interpretations with the time-lapse equation for saturation evaluation.
11
. resistivity and neutron porosity logs. tight-gas sand units in the Williams Fork formation (Piceance Basin. the average bed thickness is greater than the
X420
X440
X460
Σ Σ
Measurement Initial Guess Forward Simulation Inversion
Depth [ft]
Σ Σ
X480
X500
X520 R10 R20 R30 R60 R90 0 100 200 ECGR [gAPI] 1 10 Resistivity [Ω . western Colorado).

Using Monte Carlo methods would require running several-hour simulations for every linear iteration of the inversion method. our interpretation method improves the estimation of layer-by-layer capture cross sections.000 hours of CPU time (assuming 4 hours per run and 8 iterations) for the 130-ft. whereby the corresponding shale-sand diffusion effects are not significant. layered formations in seconds of CPU time. This would represent approximately 8. with a 130-ft. Ideally. Our method requires no more than 4 minutes of CPU time to complete the inversion. hence the measured log already provides an accurate description of the formation. However. where it is necessary to change the starting FSF because Ls greatly impacts the thermal-neutron FSFs. we showed a significant contribution in the appraisal of thin layers (up to 1. The method simulates PNC logs for 10-ft. While common shale. [cm] Vector of data residuals Half-thickness of layer. 2010
vertical resolution of the tool.u. [µs] Vector of layer capture cross sections Vertical depth. second. infiniteextent formation in presence of borehole Estimated Experimental Formation Hydrocarbons Homogeneous Intrinsic Layer k Late time Matrix Non-Shale Shale Simulated Total Water Arbitrary
12
. [cm] Slowing-down length. laminated field case requiring less than 30 seconds of CPU time.and water-saturated sands exhibit similar values of Ls (between 12 and 14 cm. salt concentration of connate water. and gas saturation. 2 and 3.). in all cases we required no a priori knowledge of the formation to perform the inversion. A subsequent step could consist of estimating bed boundaries via inversion methods. [particles/cm3] Saturation. [cps/cps] Position vector Neutron source strength. CONCLUSIONS We developed a new numerical method to simulate sigma logs and to reconcile PNC measurements with layer-by-layer sigma values. Σ spans the range from 10 to 28 c.). Accurate detection of bed boundaries is a critical step in our inversion method.5 ft. 3 indicated significant effects on inverted results due to perturbations of bed
boundaries and/or bed thickness. [particles/cm2]
λ ρb τ φ ϕ
Σ
SUBSCRIPTS
BH dif est exp F h homog int k LT ma s sh sim tot w 0 Borehole Expected measurement for the pure. In Synthetic Cases Nos. [ft] Stabilization parameter for inversion Bulk density.). depending on porosity. Interpretation of PNC logs in such thinly-bedded formations may be challenging due to shale-sand diffusion effects. one could instead consider a single high-Ls FSF to approach the simulation of PNC logs across gasbearing formations. The sensitivity analysis carried out in Synthetic Case No. However. [cm] Relative photon count. [g/cm3] Capture cross section. only. Laminated gas-bearing formations constitute more challenging cases in the interpretation of PNC logs. [kppm] Quadratic cost function for inversion Vector of numerically simulated sigma log Vector of measured sigma log Diffusion coefficient. June 19-23. [%] Thermal-neutron flux.] Thermal decay time. NOMENCLATURE
A C Cw C(x) d(x) d0 D e(x) H Ld Ls M(t) r s S t x z Amplitude of PNC signal.SPWLA 51st Annual Logging Symposium. this behavior does not affect time-lapse interpretation methods because such effects cancel out when differencing the measurements. [µs] Porosity. Except for bed boundaries. [c. [ft] Diffusion length. [ ] Time. [cps/cps] Volumetric concentration. The average value of the measurement log as the initial guess was sufficient to match the measured log within less than 2% error. especially in the cases of thinly-bedded formations and complex lithologies. field example. the starting FSF should be adapted to the local value of Ls. oil. [ ] Salt concentration (NaCl equivalent) of connate water. This behavior is important in the interpretation of field data when there are no open-hole logs available to define bed boundaries. gassaturated formations span a wider range of variation (from 14 to 30 cm. because saturations are typically unknown.u. When used in conjunction with inversion techniques.

and Meunier. Piceance Basin. W. and Portal. pp. Johnstone. no. D. 1971. Mauborgne.. Hess. 75. F. Hoyle. pp. Irvine. J. FSF kppm CPU MCNP® PNC SNUPAR TDT® Count Per Second Capture Unit (1c.. March issue. Interpretation example. Haley. S..R. SNUPAR – a nuclear parameter code for nuclear geophysics applications: IEEE Transactions on Nuclear Science. Bradford. no. pp.. and Logan..... Wyman.. R. J.. Marquardt.. W. 2005. 2007. jointly sponsored by Anadarko. Texas. Pulsed neutron capture log interpretation in laminated formations: A dual-exponential-decay model: SPE Formation Evaluation. T. Stoller. June issue. Mendoza. pp. M. May 25-28. Pathfinder. The thermal neutron
13
.. Kimminau. Preeg. = 10-3 cm-1) Flux Sensitivity Function [1/cm3-eV] Kilo Part Per Million Computer Processing Unit Monte Carlo N-Particle code Pulsed Neutron Capture Schlumberger Nuclear Parameter Code Thermal Decay Tool
ACKNOWLEDGEMENTS We thank ConocoPhillips and Schlumberger for providing field data to test our numerical simulation and inversion method.... pp. BHP Billiton. and Weatherford.1.. Fundamentals and techniques: Journal of Petroleum Technology.E. Scotland. F. Plasek. and Mitchell.. 377-385. Texas. A. Flaum. D. Steinman.. Baker Hughes.H. 2001. May 15-16.C. W. W. Methods for determining residual oil with pulsed neutron capture logs: Journal of Petroleum Technology.. Aramco. 1970. Vargas. H. Petrobras. 593-606.C. February issue.u.. Dallas. 43-57. Stacey.R.A. Edinburgh. March 30-April 1. and Plasek.E. and Quinlan. C. Adolph. New Orleans. 743-755.... Willis. and Scott.C. C. Wahl. October 22-25. pp. 1992. New Mexico Institute of Mining and Technology. 1995. 1975. D. and time-lapse technique: Journal of Petroleum Technology. 1. S. and Weller G. pp 1215-1219.J. R. no. Numerical Simulation of Pulsed Neutron Logging Tools: MS thesis. 11.SPWLA 51st Annual Logging Symposium.
Hansen. Midland. 93-98. H. P. Linear iterative refinement method for the rapid simulation of borehole nuclear measurements. Computing thermal neutron decay time environmental effects using Monte Carlo techniques: SPE Formation Evaluation... 431-441.u. C. D. E.. January-February issue. Richardson.J. Mexican Institute for Petroleum. Nexen. Quantitative interpretation of thermal neutron decay time logs: Part I. vol. D. BG.. Marathon.. RWE. 1996.. June 4-7. Part I – vertical wells: Geophysics. 35-42. March issue. Jorden. The work reported in this paper was funded by The University of Texas at Austin’s Research Consortium on Formation Evaluation. 1963.. Chevron. no.S. W. vol. Applications of pulsed neutron capture logs in reservoir management: Paper SPE 93889 presented at the 2005 SPE Western Regional Meeting. K..F. REFERENCES Bartenhagen. C..J.. An algorithm for least-squares estimation of nonlinear parameters: SIAM Journal of Applied Mathematics. M. Torres-Verdín. C. 1994. 6. D..E. 1989. June issue. Improved pulsed neutron capture logging with slim carbonoxygen tools: Methodology: Paper SPE 30598 presented at the SPE Annual Technical Conference & Exhibition.L. S. Frentrop. Morris. Schlumberger. M.K. 2010. Colorado. pp. T. J..... and David. pp. R. Nuclear Reactor Physics. E9-E29.. WileyVCH. and Smith. and Meunier. interpretation accuracy..E.E. Adolph.B. Statoil. California. pp.. Characterization and 3D reservoir modelling of fluvial sandstones of the Williams Fork Formation. R. pp. Mahdavi.D. and Preeg. 2. Cased hole formation resistivity: Changing the way we find oil and gas: Paper SPE 70042 presented at the SPE Permian Basin Oil and Gas Recovery Conference.A.. Pranter. Hamzah. 1995. M. Clavier. R. Rulison Field. and Scott. ENI. ConocoPhillips. A. vol. 1986. Method for extracting the value of thermal capture cross-section from pulsed neutron tool data by NMR-like inverse Laplace transform: Paper K presented at the SPWLA 49th Annual Logging Symposium. pp.. Quantitative interpretation of thermal neutron decay time logs: Part II.. C.W..E.. Regularization tools: a Matlab package for analysis and solution of discrete illposed problems: Numerical Algorithms.G.. The design of pulsed-neutron reservoir-monitoring programs: SPE Formation Evaluation.-L. Hoyle.J. W. 756-763. 2008. 1. Morris. 2010
ACRONYMS
cps c. 36. Dual-burst thermal decay time logging principles: SPE Formation Evaluation. R. McKeon. vol. ExxonMobil.D. 158-172. Clavier. W. Louisiana. R. vol.E.J..M. 1973. Halliburton.. Bordon.. 1988. USA: Journal of Geophysics and Engineering. June issue.. Nelligan. and Schwartz. C.. 2008. Locke. 5. BP. M.A. pp. D. Computed departure curves for the thermal neutron decay time log: Paper presented at the SPWLA 16th Annual Logging Symposium. May issue. TOTAL. and Preeg. June 19-23.E.. J. 1971. 1-35. R. 20-25.

pp.. Second. MCNP – A General Monte Carlo N-Particle Transport Code. for each sub-problem k. June 19-23. even for the case of high-Σ shale lithology (Formation 3). R. D is diffusion coefficient. ⎠
⎧ ⎛ Hk ⎞ ⎛ zk ⎞ z k ∈ [ H k . 2010
decay time log: SPE Journal. the neutron conservation equation simplifies to d2 1 s − 2 ϕ ( z) + 2 ϕ ( z) = .5) and 0.A. Accordingly... 250-kppm salt water..8 ft. First. we consider the PNC measurements to be acquired sufficiently late in time so that the contribution from the highly-absorbing borehole and casing become negligible.
Material
Formation 1 Formation 2 Formation 3 Borehole fluid Casing
τ [µs]
690 291 93. compared against the FSF computed directly from MCNP for a variety of layered formations. and normalize it in order to honor equation (4).. and thermal decay time for layer k. with beds as thin as 0. zk ∈ ]−∞. common subsurface formations exhibit a sufficiently low capture cross section compared to their total cross section. Tstart is the delay necessary for borehole-effect cancellation. Bergan. τk are half-thickness. late-time thermal-neutron FSFhomog.. Kama. Shemeta. a new nuclear log: Journal of Petroleum Technology. and Cumella.3) and methane.6 37. Y. and Hk.. S. March issue. Neutron lifetime. 1964. Denver. APPENDIX: FSF-CORRECTION PROCEDURE Layered formations introduce model heterogeneities that cause perturbations on the FSFs.H. Hopkinson.
Table 14: Thermal decay time τ for various materials (from SNUPAR). filled with 100-kppm salt water (Sw = 0. this function describes the flux of thermal neutrons in a homogeneous formation. E. and s is neutron source strength.
We subsequently implement the principle of superposition for the linear differential equation (10). J. Figure 9 shows the thermal-neutron FSF calculated with this fast. 2009. 30%-porosity illite filled with 250-kppm salt water. H k ] ⎝ Ldk ⎠ ⎝ Ldk ⎠ ⎪ ⎪ ⎪sinh ⎛ H k ⎞ exp ⎛ zk ⎞ .. Ldk. − H k ] ⎜ ⎟ ⎜ ⎟ ⎪ ⎝ Ldk ⎠ ⎝ Ldk ⎠ ⎩
(11)
where (FSFhomog)k is MCNP-computed thermal-neutron FSF for a homogeneous formation similar to layer k. Youmans. 319-328. (10) dz Ld D where ϕ(z) is thermal-neutron flux at depth z. zk is relative depth for sub-problem k (the reference origin is at the center of layer k). approximate diffusion correction procedure (assuming knowledge of all formation properties). We sum the individual (FSFcorrected)k for all the subproblems k to obtain the FSF associated with the multilayer problem. therefore.. Values of Σ span the range from 10 to 50 c.F. Steel. (9) Ld D where ϕ(r) is thermal-neutron flux at position r. The correction procedure accurately reproduces the effect of multi-
14
. zk ∈ [ − H k . Bigger is better – Hydraulic fracturing in the Williams Fork Formation in the Piceance Basin: Article Search and Discovery 110092 presented at the AAPG Annual Convention. Ld is diffusion length.u. Volume II: User’s Guide.SPWLA 51st Annual Logging Symposium. We calculate such perturbations under a set of assumptions. ⎪ ⎛ Hk ⎞ ⎛ zk ⎞ ⎪ ... given by 1 s −∇ 2ϕ ( r ) + 2 ϕ ( r ) = . Version 5. ∞[ ⎪sinh ⎜ ⎟ exp ⎜ − ⎟. Los Alamos National Laboratory. Table 14 describes the thermal decay times for typical formations. respectively. pp. and Oshry. 365-379.. X-5 Monte Carlo Team. it is valid to assume that scattering is more likely to take place than absorption. Accordingly. Because the physics of PNC measurements involves thermal neutrons only. Weijers. December issue. which enables us to decompose a complex multi-layer problem into elementary single-layer sub-problems. filled with 150-kppm salt water (Sw = 0. A.82-g/cm3 oil. the diffusion equation (Stacey. ⎨1 − exp ⎜ − ⎟ cosh ⎜ ⎟ . which are larger than those associated with both borehole fluid and casing. L. This observation allows us to assume that diffusion takes place solely in the vertical direction with no radial effect due to the borehole (we consider neither dipping beds nor invasion in this study).. Also. Colorado. 2007) accurately describes the conservation of such monoenergetic neutrons under the conditions mentioned above.0 21. H. Such a procedure is equivalent to successively considering each layer as a source of thermal neutrons that diffuse into the surrounding layers. Clean 20%-porosity limestone. ⎝ Ldk ⎠ ⎝ Ldk ⎠ ⎪ . 2003. the diffusioncorrected FSF is calculated from
( FSFcorrected )k
⎛ T = ( FSFhom og ) × exp ⎜ − start k ⎝ τk
⎞ ⎟ × .C. spatial variations of the neutron distribution are linear and we exclusively deal with isotropic scattering.0
Description
Clean 15%-porosity sandstone. June 7-10. diffusion length. The initial distribution is given by the Monte Carlo-derived.

therefore the larger value of Ls compared to oil.5
-0. respectively.D. resulting in less than 1. June 19-23.13 ft.u. Upon correction for presence of gas. in 1991. 2010
layer formations on the FSFs with similar slowingdown lengths Ls (Cases Nos. 2 and 3).5
45
0
1
0
1
0
1
0
1
0
1
0
1
1D Normalized Response [cps/cps]
Figure 9: Comparison of late-time thermal-neutron FSFs simulated with the diffusion-approximation correction (dashed red line) and MCNP (solid blue line). Orange-colored lines at 0 and 1.5
0. 1 to 6 (from left to right). formation
15
. we apply the previously described corrections to another FSFhomog.5
0.u. error on the estimated Σ for the cases considered in this paper. he was Reservoir Specialist and Technology Champion with YPF (Buenos Aires. Carlos Torres-Verdín received a Ph.5
1. Mimoun is a graduate research assistant and a Master’s student in the Department of Petroleum and Geosystems Engineering of The University of Texas at Austin. We initialized the correction procedure using the FSF for a homogeneous formation (dashed black line). 1.5
15
20
0
0
0
0
0
0
25
Relative Depth [ft]
0. He was also a reservoir engineer intern with Beicip-Franlab (IFP) in Paris during summer 2008.5
-0.5
0.5
1. From 1997–1999.5
0. We achieve less than 2% error on the calculation of the multi-layer FSF.5
30
35
1
1
1
1
1
1
40
1. To that end. and well testing.5
-0. degree in Engineering Geoscience from the University of California. Argentina). describe the positions of detector and source.5
0.5
1. despite the multiple transitions across distinct materials. His interests include nuclear well logging.
ABOUT THE AUTHORS Jordan G.
Σ [c. where he worked in the Formation Evaluation Core Group.]
-0. he has been with the Department of Petroleum and Geosystems Engineering of The University of Texas at Austin. Since 1999.5
1.and watersaturated formations.5
-0. Gas formations (represented in yellow) require a specialized treatment due to their lower hydrogen index. He received a degree in Multidisciplinary Engineering from École Centrale de Lille (France) in 2008. During 1991–1997 he held the position of Research Scientist with Schlumberger-Doll Research. He conducts research on borehole geophysics. where he currently holds the position of Zarrow Centennial Professor in Petroleum Engineering. He was an intern with ExxonMobil Exploration Company in Houston during summer 2009. Berkerley. for Cases Nos.
previously calculated for the case of a homogeneous gas-bearing formation with similar value of Ls.5
-0.5
1.7-c. inverse problems. petrophysics. 4 through 6. the calculation procedure leads to equally satisfactory results for Cases Nos.SPWLA 51st Annual Logging Symposium.

and the Atomic Energy Commission. 2004.
16
. and is founder and director of The University of Texas at Austin’s Joint Industry Research Consortium on Formation Evaluation. Associate Editor for the Journal of Electromagnetic Waves and Applications. Preeg received a Ph. Prior to 1980. is recipient of the 2006 Best Presentation Award and of the 2007 Best Poster Award by the SPWLA. degree in Nuclear Science and Engineering from Columbia University in 1970. Vice-president of Engineering in Houston as well as Manager of the Nuclear Department at SDR. Aerojet Nuclear Systems Company.SPWLA 51st Annual Logging Symposium. and Georgia Tech. June 19-23. and integrated reservoir characterization. Torres-Verdín has served as Guest Editor for Radio Science. and is currently an Associate Editor for Petrophysics (SPWLA) and Geophysics. Texas A&M. and the SPE Journal.
William E. and is recipient of SPWLA’s 2006 Distinguished Technical Achievement Award and SPE’s 2008 Formation Evaluation Award. Colorado School of Mines. He is recipient of the 2003. He has also served on advisory committees at The University of Texas at Austin. largely in the area of nuclear radiation transport.D. and 2007 Best Presentation Award by Petrophysics. From 1980 to 2002. he held various positions with Schlumberger including Director of Research at Schlumberger-Doll Research (SDR). 2006. 2010
evaluation. he worked for Los Alamos Scientific Laboratory.