MULTICHANNEL OIL-BASE MUD CONTAMINATION MONITORING USING DOWNHOLE OPTICAL SPECTROMETER

Transcription

1 MULTICHANNEL OIL-BASE MUD CONTAMINATION MONITORING USING DOWNHOLE OPTICAL SPECTROMETER Kai Hsu, Peter Hegeman, Chengli Dong, Ricardo R. Vasques, Michael O'Keefe, and Mario Ardila, Schlumberger Copyright 2008, 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 49 th Annual Logging Symposium held in Edinburgh, Scotland, May 25-28, ABSTRACT Contamination of formation fluid is caused by the filtrate of drilling mud that invades the formation surrounding the borehole during and after drilling. Consequently, a fluid sample taken by a downhole formation tester tool is inevitably a mixture of the mud filtrate and virgin fluid. Contamination can largely reduce the sample quality and make the subsequent pressure/volume/temperature (PVT) analysis unreliable. The objective of real-time contamination monitoring is to continuously analyze the fluid pumped from the formation through the flowline using a downhole optical spectrometer until an acceptable contamination level is measured and then to capture a sample of the fluid. In this paper, we describe a new multichannel oil-base mud (OBM) contamination monitoring technique using optical density measurements recorded at different wavelength channels by an optical spectrometer. Compared to the conventional single-channel approach, the multichannel method has several advantages. First, it can produce a consistent contamination estimate among all the channels of the spectrometer. Second, rather than assuming that the filtrate optical density is zero, the multichannel approach computes the filtrate optical density by exploiting the linear relationship among all the channels. Third, it provides a confidence measure that quantifies the statistical variation of the contamination estimate among all the channels. Finally, the multichannel approach can also estimate the spectra of filtrate and formation oil. To demonstrate the capability of the new method, we apply it to field data from recent field tests and compare the results with the contamination estimate obtained from the laboratory PVT analysis of captured samples. We find very good agreement between the 1 contamination estimates of the multichannel approach and the PVT analysis. INTRODUCTION The use of wireline formation-testing tools to bring formation-fluid samples to the surface for examination was introduced in the early 1950s (Badry et al., 1994). However, fluid sampling in early days suffered several important limitations: the sample often contained a large amount of contaminants (e.g., mud filtrate) and the flow pressure sometimes dropped below bubblepoint, thereby changing the sample characteristics. These drawbacks frequently led to the end-user s frustration caused by the absence of anticipated information or the presence of fluids later found to be useless because of high contamination and phase alteration. Therefore, the key to fluid sampling is to ensure that fluids retrieved from the wells are representative of the reservoir. Introduction of the downhole optical spectrometer (Smits et al., 1995) for contamination monitoring provides an in-situ solution for these particular problems. The objective of real-time contamination monitoring is to use a downhole optical spectrometer to continuously analyze the fluid pumped from the formation through the flowline until an acceptable contamination level is measured and then to capture a sample of the fluid. Reservoir fluid contamination can be either immiscible or miscible. For immiscible contamination, the downhole optical spectrometer is capable of distinguishing water from oil by differences in optical absorption of light at visible and near infrared wavelengths (Smits et al., 1995). Therefore, for the case of water-base mud invading an oil-bearing reservoir, or oil-base mud invading a water-bearing formation, the uniqueness and separation of absorption peaks permit differentiation of oil and water (Hashem et al., 1997). Contamination monitoring becomes much more critical when fluids involved are miscible. This includes the formation crude oil invaded by oil-base mud filtrate and the formation connate water invaded by water-base mud filtrate. Adding a dye to drilling mud can be used

2 to quantitatively determine the contamination level involving the filtrated water and formation connate water. However, for the oil-base mud contamination, a key objective is to distinguish between oil-base mud filtrate and formation crude oil. Achieving this objective is not as straightforward as differentiating oil and water where one can rely on the readily distinguishable absorption peaks in the optical data. In this paper, we decipher the problem of oil-base mud contamination, and to achieve this objective, we propose a new multichannel contamination monitoring technique using optical density measurements recorded at different wavelength channels by a downhole optical spectrometer. spectra based on their respective OD values at λ = 1600 nm. Specifically, each spectrum subtracts its OD value at λ =1600 nm. Note that except for offsetting two end points OD λ, fil and OD λ, oil by their respective OD values at λ = 1600 nm, this normalization procedure does not alter the Beer Lambert relationship of Eq. (1) and the sensitivity to η. Figure 1 shows the normalized spectra of miscible mixtures with different levels of filtrate contamination in volume % (i.e., η x 100%). BEER-LAMBERT LAW The principle of contamination monitoring (Mullins et al., 2000a and 2000b) with optical measurements is based on the Beer Lambert law (Mullins et al., 2000c) that establishes a linear relationship between the optical absorbance (i.e., optical density, OD ) and the concentrations of species under investigation. For a binary mixture of formation oil and mud filtrate, the measured OD λ at the wavelength λ is linearly related to the contamination level by the linear mixing law OD λ η ODλ, fil + ( 1 η) ODλ, oil =, (1) where OD λ, fil and OD λ, oil are the optical densities of mud filtrate and formation oil at the wavelength λ, respectively, and η is the contamination level in volume fraction. Assuming that η changes with respect to the pumping time, the values of OD λ would reflect the changes in contamination level in the sampled fluid in front of the optical window. However, if OD λ, fil = ODλ, oil, the sensitivity to contamination diminishes. Fortunately, there are many wavelengths at which the filtrate and formation-oil optical densities differ significantly, warranting the principle of using the optical measurements for contamination monitoring. We further validate the linear mixing law of Eq. (1) using data acquired by a laboratory optical spectrometer. The experiments were conducted with a live oil in the system initially and subsequently, various levels of synthetic oil-base mud filtrate were injected into the system to form miscible mixtures. The optical spectra for these mixtures, which correspond to different levels of filtrate contamination, were acquired by the spectrometer. A separate experiment was also conducted to acquire the spectrum with 100% oil-base mud filtrate in the system. To mitigate the effect of spectral drift and light scatterings, we normalize all the 2 Figure 1: Normalized spectra of miscible mixtures with different levels of filtrate contamination. Figure 2 shows the optical-density values plotted against the contamination levels at a few arbitrarily chosen wavelengths. The results validate Eq. (1). Specifically, for each wavelength, all the measured OD values with different contamination levels (i.e. 2 % to 25%) fall on the straight line connecting two end points corresponding to zero and 100% contamination. CURRENT APPROACH Based on Eq. (1), one can derive the contamination from the measured OD by the following equation, i.e., λ ODλ, oil ODλ η =. (2) ODλ, oil ODλ, fil However, OD λ, fil and OD λ, oil are normally unknown. Two assumptions are made to derive the current contamination algorithm (Mullins et al., 2000a and 2000b). Firstly, a long-time behavior of the fluid-flow model was established to extrapolate the existing data. The model is inspired by the theoretical models describing the long-time behavior of fluid flow into the flow line via a small sink probe pressed against the borehole wall (Hammond, 1991). That is, the optical

3 assumption is applied to the selected color and methane channel in the current algorithm (Mullins et al., 2000a, 2000b; Dong et al., 2003). The presence of optical scatterings in the data often interferes with the data fitting with the model of Eq. (3). To mitigate this effect, a preprocessing step is often added to the procedure described above. Namely, the optical density of a baseline channel is subtracted from the selected color or methane channel before the model fitting with Eq. (3) is performed. As noted before, except offsetting the end points OD λ, fil and OD λ, oil by some constant values, this differencing procedure does not alter the Beer Lambert relationship. MULTICHANNEL APPROACH Figure 2: Optical density values of Figure 1 versus the contamination levels. density ( OD λ (t) ) observed as a function of pumping time is modeled with the following equation: OD α λ ( t) = C D t, (3) where C is the unknown asymptotic value of OD λ (t), D is an unknown constant, and α is a decay value, usually within a range of 0.2 to 0.8, and typically having a value of approximately 0.5. After analyzing more than 70 field data sets from the Gulf of Mexico and elsewhere, a single functional form t 5/12 evolving in pumping time (i.e., t) was derived empirically as the fluid flow model (Mullins et al., 2000a, 2000b). This empirical model is also validated by numerical modeling and simulation (Alpak et al., 2006). That is, the optical density data ( OD λ ) measured as a function of pumping time is fitted with the fluid-flow model, which enables the extrapolation to obtain OD λ, oil (i.e. C). The second assumption is that OD λ, fil (the optical density of the filtrate at certain wavelength λ ) is nearly zero. Thus, the substitution of OD λ, fil = 0 is made when estimating the contamination using Eq. (2). This 3 The current contamination-monitoring method is based on a single-channel approach. The method is applied separately to a color channel and to the methane channel. As demonstrated by the laboratory spectra of Figures 1 and 2, the optical channel data (from visible to NIR) contains information about the filtrate contamination even though the sensitivity to contamination varies from channel to channel. Therefore, the idea of a multichannel approach is appealing because the multichannel data actually contain much redundant information. Unlike the singlechannel approach that may yield different contamination estimates from the color channel and from the methane channel, the multi-channel approach proposed in this paper produces a single contamination estimate that is consistent with all the channel data. With the introduction of a new generation multichannel downhole spectrometer with high data quality (Dong et al., 2007; Fujisawa et al., 2008), the proposed approach provides considerable improvements in the oil-base mud contamination monitoring during a fluid sampling job. We use a multichannel field data set to demonstrate the idea. Step 1: Preprocessing With the multi-channel optical data as input, the first step involves the determination of useable channels for the rest of the processing. The procedure is illustrated with a multichannel field data set shown in Figure 3, where the optical spectrometer data is shown on the top subplot and the pump volume is shown on the bottom. Both optical and pump volume data are acquired versus elapsed time by the acquisition system and serve as input to the processing. The optical density data in each channel after the start of pumping are checked to determine whether they exceed a threshold value. If the frequency of occurrence in a channel is larger than a

4 preset number, that particular channel is marked as unusable. In the case of Figure 3, the threshold value and the preset number are set to 2.5 and 100, respectively, for this procedure. Note that it is important to exclude the data before the start of pumping because, as shown in Figure 3, the data in all channels frequently exceed the threshold before the start of pumping. This is likely caused by the presence of mud in the flowline before the start of pumping. With this preprocessing procedure, seven channels are determined as useable for subsequent processing. volume-based model is more appropriate than the timebased model for describing the data buildup or builddown behavior. Therefore, all available time-based channels are converted to be volume-based. The conversion uses the pump volume-elapsed time relationship, displayed in Figure 3. The results are shown in Figure 4, in which each channel is the difference channel after subtracting the baseline channel. As a result, there are only six channels because the baseline difference channel is by definition zero. Figure 4 shows all the usable difference channel data (as a function of pumpout volume) available for subsequent processing. Figure 4: Difference channels converted to be a function of pumpout volume. Step 3: Determination of linear relationship among channels Figure 3: Field data set showing optical spectrometer data (a) and pump volume data (b). Once the usable channels are identified, the common baseline channel is subtracted from each channel. In this example, the baseline channel is at wavelength of 1600 nm. The resulting difference channels (as a function of elapsed time) are then converted to be a function of pumpout volume in Step 2. Step 2: Conversion from time-based to pump-volumebased channels If the pumping rate was not constant during the sampling job (which is frequently the case), then the Without loss of generality, the optical channels described in the rest of algorithm are regarded as the optical difference channels after subtracting the baseline channel in the preprocessing step. By taking a particular channel as the reference channel and another channel at a different wavelength, the measured optical densities as a function of pumpout volume (v) at these two channels can be expressed as = η( v) ODi, fil + ( 1 ( v) ) ODi oil, (4) OD i ( v) η, = η( v) ODref, fil + ( 1 ( v) ) ODref oil, (5) OD ref ( v) η, where ref and i denote the reference channel and the channel at a different wavelength, respectively. Note that Eqs. (4) and (5) are the same representation as Eq. 4

5 (1), except that the measured optical density and contamination in Eqs. (4) and (5) are expressed as a function of pumpout volume. By some algebraic manipulation, one can relate these two measurements by ODi ( v) = Ai + Bi ODref ( v), (6) where A i and B i are two constants, and they only depend on the end points OD i, fil, OD i, oil, OD ref, fil, and OD,, i.e., ref oil Ai ODi, filodref, oil ODi, oilodref, fil ODref, oil ODref, fil =, (7) Bi ODi, oil ODi, fil ODref, oil ODref, fil =. (8) In Eqs. (7) and (8), OD i, fil, OD i, oil, OD ref, fil, and OD ref, oil are normally unknown and therefore, A i and B i are unknown as well. Fortunately, A i and B i can be estimated from Eq. (6): Treat the data OD ref (v) and OD i (v) as the independent and dependent variable, respectively, and fit the data with Eq. (6). Figure 5 shows the crossplots of optical density data of the reference channel with the optical density data of five other usable difference channels obtained from Step 2. All the crossplots of data exhibit the linear relationship, validating Eq. (6). In the channel A vs. reference channel crossplot (upper left-hand subplot), a small portion of data at the beginning of pumping actually deviates from the linear trend. This could be caused by the presence of mudcake debris in the flow line initially. Nevertheless, the majority of data follow the linear trend once the cleanup and pumping progresses. The dotted line indicates the fit with the least-absolutevalue criterion or L1-norm (McCullagh and Nelder, 1989) that is known to be a robust estimate preventing the outliers from altering the linear trend. The intercept and the slope of these lines are A i and B i, respectively, given in Eq. (6). Note that the optical densities of filtrate and formation oil of all channels are constrained to be on these lines determined by A i and B i. Furthermore, knowing the optical density of filtrate at one channel, the optical densities of filtrate at all other channels can be obtained. Similarly, for formation oil, knowing its optical density at one channel, the optical densities of formation oil at all other channels can be estimated as well. Figure 5: Crossplots of optical density data of five other usable difference channels (A, B, C, D and E) obtained from Step 2. The analysis to exploit the linear relations among the optical channels is valid no matter which channel is chosen as the reference channel. However, there is a good reason to choose a particular channel as the reference channel. Figure 6 shows the laboratory spectra of three different types of oil-base mud filtrate at 80 o C and 10 kpsi. These spectra have been normalized by subtracting the optical density value at the 1600 nm, as was done with the difference procedure in Step 1. For all three OBM filtrate spectra, the channels in the wavelength range of 800 to 1600 nm would have the OD values very close to zero (i.e. OD ref, fil 0 ) except those around the absorption peaks of 1200 and 1400 nm. The upshot is that if we choose a particular channel (whose filtrate OD value is close to zero) as the reference, then the filtrate OD value at the other crossplot channels can be estimated as ODi fil Ai,, (9) 5

6 which is obtained by substituting OD ref, fil 0 in Eq. (7). The reference channel in Figure 5 was chosen with this criterion. It is interesting to note that the crossplot channel B in Figure 5 is supposed to be another channel whose filtrate OD is close to zero. The zero intercept of linear trend line (i.e. OD i, fil A i 0 ) in this case also confirms the assumption. With A i and ODi ( v) A OD i ref v) = Bi (. (10) B i determined in Step 3, all the usable channels can be converted to the equivalence of the reference channel. Figure 7 shows the plot of all channels in Figure 4 that have been converted to the equivalence of the reference channel. Except for some noise and interference in the data, all channels essentially collapse to a single optical density channel and this further confirms the validity of the analysis. Figure 6: Laboratory spectra of three different types of OBM filtrate at 80 o C and 10 kpsi. In summary, exploitation of the linear relationship [Eqs. (6) to (8)] among all the usable channels enables us to determine the filtrate OD values (i.e. OD i, fil ) using Eq. (7) if the reference channel is selected properly ( OD ref, fil 0 ). Furthermore, knowing the filtrate OD values and B i determined from the slope of crossplot, we can obtain the formation oil OD values (i.e., OD i, oil ) using Eq. (8), provided that the formation oil OD value at the reference channel (i.e., OD ref, oil ) is known. The following sections describe how to determine OD ref, oil from all available data. Step 4: Combined optical channel data Eq. (6) can be rewritten as 6 Figure 7: Plot of all channels shown in Figure 4 that have been converted to the equivalence of the reference channel. Note that there is some discrepancy at the beginning of pumping that is caused by the deviation of linearity of the model as mentioned in Step 3. Once all the channels converge to the equivalence of the reference channel, we can combine them to obtain an improved S/N channel. A simple way to combine them is to take the average of them. Specifically, at each pump volume instance, the OD value of the combined channel is computed as the averaged value of six normalized OD values at that instance. However, the nonlinear methods such as median or trimmed mean (Walpole, 2002) are better choices because these methods can effectively remove the residual scatterings remaining in the data after differencing. A trimmed mean is calculated by discarding a certain percentage of the lowest and the highest optical density values and then taking the average of the remaining optical density values. For example, a mean trimmed 50% is computed by discarding the lower and higher 25% of the optical densities and taking the averaged value of the remaining optical densities. The median is the mean

7 trimmed 100% whereas the average is the mean trimmed 0%. A trimmed mean or median is less susceptible to the effects of extreme optical density values such as scatterings than is the average. The trimmed mean is an attractive option if there are many useable channels available. Throughout the rest of the paper, we use the median to combine the optical channel data. that both spectra are offset from true spectra by their respective optical density value at the baseline channel of 1600 nm. The offsets are caused by the difference operation with the baseline channel in Step 1. Nevertheless, the determination of contamination is not affected by these offsets for the reasons described earlier. Step 5: Determination of formation oil OD at reference channel Modeling the long-time behavior of fluid flow is crucial to extrapolate the combined optical data to obtain the optical density of formation oil. Inspired by the theoretical models developed earlier (Hammond, 1992) and previous studies based on extensive field data (Mullins et al., 2000a and 2000b), we model the buildup or builddown behavior of optical data as given by Eq. (3). One minor modification is that the elapsed time t in Eq. (3) is replaced by the pump volume ν, i.e., OD α λ ( ν ) = C D ν, (11) where the exponent α is a decay value, usually within a range of 0.2 to 0.8, and typically having a value of approximately 0.5. In addition, when fitting data with Eq. (11), we would need to determine the appropriate fitting interval. Since we seek to characterize the most recent fluid flow behavior, the end of the fitting interval is always marked at the last data point. The beginning of the fitting interval, however, is located by adjusting the beginning so that the fitting results reach the minimum of least absolute valve criterion (McCullagh and Nelder, 1989). Figure 8 shows the results of fitting using the model of Eq. (11). The plot shows the combined optical density obtained in Step 4 (in blue) and the fit (in red); the vertical red dotted line indicates the beginning the fitting interval. Note that the fit nearly overlays the data within the fitting interval. The value of C obtained from the fitting is the estimated optical density of formation oil at the reference channel (i.e., OD, ). ref oil Step 6: Estimation of spectra of filtrate and formation oil As noted before, with A i and B i estimated in Step 3 with a properly selected reference channel ( OD ref, fil 0 ) and OD ref, oil obtained in Step 5, we can estimate the spectra of filtrate and formation oil, i.e., OD i, oil and OD i, oil of all other channels. Specifically, they are derived using Equ. (6) to (9). However, note 7 Figure 8: Model fit of the combined optical density data. Step 7: Determination of contamination The final step is to determine the contamination and to calculate a statistical measure associated with the contamination estimate. With the estimated filtrate and formation oil spectra obtained in Step 6, we can estimate the contamination with Eq. (2). For the example shown here, there are six available difference channels (shown in Figure 4) and therefore, there are six contamination estimates (one for each channel) at each volume instance. The median and standard deviation with respect to the median of these estimates at each instance are calculated. The median value is the estimated contamination, and the standard deviation around the median is a characteristic of noise associated with the data. Figure 9 shows the contamination estimate (blue) and the associated noise statistic (i.e., three times the standard deviation) plot (red). At the end of pumping the estimated contamination reaches approximately 9.0 volume percent. The PVT analysis of four fluid samples taken at the end of the sampling job reports the contamination level of 8.8, 10.8, 8.7, and 8.7 weight percent, which are in a good agreement with the results obtained from the downhole spectrometer data. As a final note, the contamination level in volume percent typically differs from its value in weight

8 percent. A detailed discussion about the difference is given in the Appendix. Figure 9: Contamination estimate and associated noise statistic. For comparison, we show the estimated contamination using the single channel approach (Mullins et al., 2000a, 2000b; Dong et al., 2003). In the single channel approach, two separate channels are selected for the processing. One of them is a channel in the wavelength range of 800 to 1600 nm and the other is a channel around the methane absorption peak. For fair comparison, we use the reference channel for the former. Figure 10 shows the estimated contaminations (displayed as a function of elapsed time) obtained with the single channel approach. Since the two channels are processed independently of each other, there is no assurance that two estimates will overlay. At the end of pumping, the estimated contamination levels reach about 12.5 volume percent and 11 volume percent, respectively, for the reference channel (blue) and methane channel (red). Figure 10: Contamination estimates obtained with the single channel approach. FIELD EXAMPLES To demonstrate the capability of the multichannel approach, we apply it to additional field data sets and compare the results with the contamination estimate obtained from the laboratory PVT analysis of captured samples. Selection of field data is based upon the following two criteria: (1) the PVT report of sampled fluid is available and (2) the contamination level at the end of sampling job is low (i.e., < 5 weight percent) based on the PVT report. As a result, seven data sets are chosen for the study. Figure 11 shows the multichannel spectrometer data (top) and the pump volume-versuselapsed time data (bottom) for one of data sets. The pumping starts at the elapsed time of 500 seconds. The buildup and builddown trends of the optical data are clearly seen as the pumping time increases. Only the channels with OD < 2.5 are selected for the processing. The results of the multichannel contamination approach are shown in Figure 12. At the end of the sampling job, after pumping out 90 L of fluid from the formation, a fluid sample was taken. The estimated contamination level from Figure 12 is approximately 2.2 volume percent at the point of sampling. For comparison, the PVT analysis of the captured sample reports the contamination level of 1.3 weight percent. The two estimates agree well. As noted before, a small discrepancy could be due to the PVT results being expressed in weight percent and optical spectrometer results being in volume percent. 8

9 For comparison, we also show the results obtained with the single-channel approach (Figure 13). The anomalous feature that appears around 6700 s is caused by opening the sample bottle to take fluid sample. Around 6700 s, the estimated contamination levels are 3.5 volume percent and 3.4 volume percent, respectively, for the reference and methane channels. Figure 13: Contamination estimates for case 1 obtained with the single channel approach. Figure 11: Optical density (top) and pump volume data (bottom) of case 1. Figure 14 shows the input data for the second example. Before the pumping starts at 400 seconds, the flowline is filled with mud and therefore, the OD values of all channels exceed 4. The cleanup of fluid in the flowline happens rather quickly as evidenced by the abrupt reduction of OD values of multiple channels to the range of 0 to 2. Soon after, the data exhibits the buildup or builddown trend as the pumping continues. Once again, a fluid sample was taken at the end of the sampling job. Figure 15 displays the contamination estimate and the associated noise statistic obtained by using the multichannel approach. The spiky feature at the end of the data is caused by the anomalous escalation of OD values as a result of opening the sample bottle. Nevertheless, the estimated contamination level reaches approximately 2.9 volume percent at the end, which agrees well with the contamination level of 3.2 weight percent reported by the PVT analysis. Figure 12: Contamination estimate and associated noise statistic for case 1 obtained with the multichannel approach. 9

10 Figure 16 shows the estimated contamination (case 2) based on the single channel approach. Around 8200 s before opening the sample bottle, the estimated contamination levels reach about 4.4 volume percent and 1.4 volume percent, respectively, for the reference and methane channel. Figure 16: Contamination estimates for case 2 obtained with the single channel approach. Figure 14: Optical density (top) and pump volume data (bottom) of case 2. Table 1 compares contamination estimates of downhole spectrometer data based on the single-channel approach, multichannel approach, and the laboratory PVT analysis of captured samples for all the cases. The symbol SC-R denotes the single channel approach using the reference channel data whereas SC-M stands for the single channel approach using the methane channel data. The results of multichannel approach are shown under the symbol MC in Table 1. Other than the advantage of a single consistent estimate derived from all channel data, it is clear that the contamination estimates of the multichannel approach show a much better agreement with the results obtained by the PVT analysis of captured sample. Case SC-R (vol%) SC-M (vol%) MC (vol%) PVT (wt%) Figure 15: Contamination estimate and associated noise statistic for case 2 obtained with the multichannel approach. Table 1: Comparison of contamination estimates. 10

11 CONCLUSIONS We present a new multichannel oil-base mud contamination monitoring technique using optical density measurements recorded at different wavelength channels by a downhole optical spectrometer. Compared to the conventional single-channel approach, the multichannel method has several advantages. First, it can produce a consistent contamination estimate among all the channels of the spectrometer. Second, rather than assuming that the filtrate optical density is zero, the multichannel approach computes the filtrate optical density by exploiting the linear relationship among all the channels. Third, it provides a statistical measure that quantifies the variation of the contamination estimate among all the channels. Finally, the multichannel approach is capable of estimating the baseline-adjusted spectra of filtrate and formation oil. A study of seven field data sets with low contamination level (i.e., < 5 weight percent) shows very good agreement between the contamination estimates of the spectrometer data using the multichannel approach and the PVT analysis on the captured sample. REFERENCES Alpak, F. O., Elshahawi, H., Hashem, M., Mullins, O. C., 2006, Compositional Modeling of Oil-Based Mud- Filtrate Cleanup during Wireline Formation Tester Sampling, SPE Annual Technical Conference and Exhibition, SPE Badry, R., Fincher, D., Mullins, O.C., and Smits, A., 1994, Downhole Optical Analysis of Formation Fluids, Oilfield Review, vol. 6, p Dong, C., Hegeman, P. S., Elshahawi, H., Mullins, O. C., Fujisawa, G. and Kurkjian, A., 2003, Advances in Downhole Contamination Monitoring and GOR Measurement of Formation Fluid Samples, SPWLA 44 th Annual Logging Symposium, Paper FF. Dong, C., O Keefe, M., Elshahawi, Hashem, M., Williams, S., Stensland, D., Hegeman, P., Vasques, R., Terabayashi, T., Mullins, O. and Donzier, E., 2007, New Downhole Fluid Analyzer Tool for Improved Reservoir Characterization, presented at Offshore Europe 2007, SPE Fujisawa, G., Jackson, R., Vannuffelen, S., Terabayashi, T. and Yamate, T., 2008, Reservoir Fluid Characterization with a New-Generation Downhole Fluid Analysis Tool, SPWLA 49 th Annual Logging Symposium. Gozalpour, F., Danesh, A., Tehrani, D.-H., Todd, A. C., Tohidi, B., 1999, Predicting Reservoir Fluid Phase and Volumetric Behavior from Samples Contaminated with Oil-Based Mud, SPE Annual Technical Conference and Exhibition, SPE Hammond, P., 1991, One- and Two-phase Flow during Fluid sampling by a Wireline Tool, Transport in Porous Media, vol.6, p Hashem, M. N., Thomas, E. C., McNeil, R. I., and Mullins, O. C., 1997, Determination of Producible Hydrocarbon Type and Oil Quality in Wells Drilled with Synthetic Oil-Based Muds, SPE Annual Technical Conference and Exhibition, SPE McCullagh, P. and Nelder, J. A., 1989, Generalized Linear Model. 2 nd Edition, Chapman and Hall. Mullins, O. C., Schroer, J. and Beck, G., 2000a, Realtime Quantification of OBM Filtrate Contamination in the MDT Using OFA Data, SPWLA 41 st Annual Logging Symposium, Paper SS. Mullins, O. C. and Schroer, J., 2000b, Real-time Determination of Filtrate Contamination during Openhole Wireline Sampling by Optical Spectroscopy, SPE Annual Technical Conference and Exhibition, SPE Mullins, O. C., Joshi, N. B., Groenzin, H., Daigle, T., Crowell C., Joseph, M. T., and Jamaluddin, A., 2000c, Linearity of Near-infrared Spectra of Alkanes, Applied Spectroscopy, vol. 54, p Smits, A. R., Fincher, D. V., Nishida, K., Mullins, O. C., Schroeder, R. J., and Yamate, T., 1995, In-Situ Optical Fluid Analysis as an Aid to Wireline Formation Sampling, SPE Formation Evaluation, vol.10, p Walpole, R., 2002, Probability and Statistics for Engineers and Scientists (7 th Edition), Prentice Hall. ABOUT THE AUTHORS Kai Hsu is an engineering advisor in the Schlumberger Sugar Land Product Center. Since joined Schlumberger, he has worked on R & D projects in multiple technical disciplines, including ultrasonic imaging, sonic-seismic integration, acoustic well logging (wireline and LWD), gamma ray spectroscopy, and most recently, formation testing and sampling. He received his MS and PhD in electrical engineering from the University of Texas at Austin. 11

12 Peter Hegeman is a project manager and engineering advisor in the Reservoir Sampling and Pressure Discipline of Schlumberger Sugar Land Product Center. His interests include well testing, pressure-transient analysis, formation testing, and production systems analysis. He holds BS and MS degrees in petroleum engineering from the Pennsylvania State University. Chengli Dong is a principal research scientist and reservoir domain champion with Schlumberger in both Sugar Land Product Center and North Gulf Coast. He is very active in downhole fluid analysis, formation testing and sampling, and development and applications of new formation tester tools and their answer products. He holds a BS degree in chemistry from Peking University and a PhD degree in petroleum engineering from the University of Texas at Austin. Ricardo Vasques is the current Schlumberger subsea marketing manager located in Clamart, France. He has worked in field and management positions for Schlumberger Wireline in South America, North Sea, and Middle East prior to a position as program manager for development of formation testing tools in Houston. He holds an electrical engineering degree from the Instituto Militar de Engenharia, Rio de Janeiro. Mario Ardila is a principal reservoir engineer with Schlumberger located in Indonesia. He has been very active on projects ranging from production enhancement, sand management, well productivity optimization, downhole fluid analysis, formation testing and sampling. He holds a BS degree in petroleum engineering from the Universidad Industrial de Santander in Colombia. APPENDIX The composition and contamination level of reservoir fluid samples are typically obtained by gas chromatography (GC) in the PVT laboratory. The contamination level specified in the PVT report (Gozalpour et al., 1999) is therefore in terms of weight percent. However, optical spectrometers provide contamination level specified in volume percent. This appendix elucidates the difference between the two unit quantities. contam. volume η ( vol%) = 100% = v 100% (A.1) total volume Therefore, the number of moles of these two fluids in one-liter mixing fluid are α oil ( 1 v) and α fil v. The contamination in terms of mole percent is then given by α filv η ( mole%) = 100%. (A.2) αoil (1 v) + α filv Denote m oil and m fil as the molar mass or molecular weights (g/mole) for the formation oil and filtrate, respectively. Then, one can define the contamination in terms of weight percent m filα filv η( wt%) = 100% moilαoil (1 v) + m filα filv ρ filv = 100% ρoil (1 v) + ρ filv (A.3) where ρ oil ( = moilαoil ) and ρ fil ( = m filα fil ) are the densities of the formation oil and filtrate. The contamination of reservoir fluid samples is measured in weight percent in the PVT laboratory. A logical question to ask is: What is the contamination value that a optical spectrometer measures? For the binary mixing fluid, based on the Beer-Lambert law, the measured optical density at a particular wavelength is ODmeasure = ε oilbαoil ( 1 v) + ε filbα filv, (A.4) where ε oil and ε fil are the optical absorptivity of the formation oil and filtrate at that wavelength, and b is the optical path length. With this, the contamination measured by the optical spectrometer is OD OD ( optics) oil η = measure 100% ODoil OD fil εoilbαoil ( εoilbαoil (1 v) + ε filbα filv) = 100% (A.5) εoilbαoil ε filbα fil = v 100% = η( vol%) Normally, Assume α oil and α fil are the molar density (i.e., moles/liter) of the formation oil and oil-base mud filtrate, respectively. Also assume the volume of filtrate in one liter of the mixing fluid is v which can be interpreted as the contamination in terms of the volume percent, i.e., 12 η( vol%) η( mole%) η( wt%) (A.6) But, if the density of the formation oil is close to that of filtrate (i.e., ρoil ρ fil ), the contamination measured by the optical spectrometer will be close to what is measured by the PVT laboratory,

13 η( optics) = η( vol%) = v 100% η( wt%) (A.7) Figure 17 shows the difference of contamination based on volume percent and based on weight percent versus the contamination in weight percent. The densities of formation fluid for the simulation are 0.3 g/cm 3 (wet gas/condensate), 0.8 g/cm 3 (light hydrocarbon), and heavy 1.0 g/cm 3 (heavy hydrocarbon), whereas the density of oil-base mud filtrate is set at 0.7 g/cm 3. The largest difference between the weight percent and the volume percent occurs when the contamination level is at approximately 40 to 60%. At the two extremes (no contamination or 100% contamination), the two contamination estimates are identical. When the filtrate density is the same as the formation fluid density, there is no difference between the two estimates. The contamination estimate based on the volume percent is larger than that based on the weight percent if the formation fluid density is larger than the filtrate density. Note that when the contamination is low (e.g., below 10%), the difference is approximately 2 to 3% at the worst with a large density contrast between the formation crude (1.0 g/cc) and filtrate (0.7 g/cc). In contrast, if the formation fluid is gas condensate with low density, the contamination estimate based on the volume percent will be lower than that based on the weight percent. The difference can be significant if the density contrast between the condensate and the filtrate is large. Figure 17: The difference between contamination based on volume percent and based on weight percent versus the contamination in weight percent. 13