OTC Copyright 2001, Offshore Technology Conference

OTC 13000 Reliability Through Data Reconciliation Robert van der Geest, ABB; William H. Broman Jr., Tina L. Johnson, and Ray H. Fleming, BP; and John O. Allen, ABB Copyright 2001, Offshore Technology Conference This paper was prepared for presentation at the 2001 Offshore Technology Conference held in Houston, Texas, 30 April 3 May 2001. This paper was selected for presentation by the OTC Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as presented, have not been reviewed by the Offshore Technology Conference and are subject to correction by the author(s). The material, as presented, does not necessarily reflect any position of the Offshore Technology Conference or its officers. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the Offshore Technology Conference is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of where and by whom the paper was presented. Abstract We propose a flexible simulator of an oil well that may be used for several forms of online, real-time reconciliation of data from existing measurement equipment. It turns out that data reconciliation used as part of the instrumentation philosophy on an oilfield constitutes a dissimilar, redundant level of instrumentation. This additional level of instrumentation may be used to make the existing measurement hardware more reliable by verifying the validity of measurement data. In addition, the technology may be used to replace permanently installed hardware in times of failure. Greater reliability is expected to contribute to greater system uptime by reducing the need for workovers. Introduction Data reconciliation is a technique that has traditionally been used in process control to verify measured data by reference to a process model. In this paper we consider a more active form of data reconciliation in which we use a model of the process to estimate a number of unknown variables on the basis of other, known variables in the process. Such estimated variables may be seen as virtual measurements of the process; therefore estimators based on data reconciliation are sometimes referred to as virtual sensors. They are also known as software sensors or soft sensors, which distinguishes them from their hardware counterparts. It turns out that soft sensors may be used with great success in the operation of chemical processes for supplementing, and in many cases even replacing, permanently installed hardware. There is a big potential for soft sensors in offshore oil production. First of all, even though the added value of permanently installed sensors is undisputed, downhole equipment is particularly expensive to install because of the nature of offshore operations. Therefore it is appealing to limit the number of downhole sensors to what is minimally required in order to get sufficient insight into the production process, and to use inferential techniques to make the picture complete. Secondly, it is widely recognized that downhole measurement equipment and the corresponding communication systems are prone to failure during installation in the well, during special events in the life of the well, or after a few years in operation due to harsh living conditions. Furthermore, it is prohibitively costly to repair or replace downhole equipment in offshore wells. Software sensors, as opposed to hardware sensors, typically do not break down. In fact, even though a soft sensor depends on the availability of hardware measurements that may be subject to errors, it does not necessarily depend on the availibility of a particular piece of hardware. A soft sensor will work as long as the total set of available data contains enough information for the system to make up its mind. In this paper we propose a flexible simulator of an oil well that may be used for several forms of online, real-time data reconciliation. This includes the following applications. a) Verification of the validity of measurement data under normal operating conditions. This corresponds to the classical notion of data reconciliation as a data verification tool. b) Estimation on the basis of available data of a number of interesting process variables that are not being measured directly. This is an example of data reconciliation as a tool to supplement existing equipment. c) Estimation of measured variables at times when measurement equipment fails. This application is an example of data reconciliation as a tool to take over some of the functionality of permanently installed hardware. Data reconciliation used as a part of the instrumentation philosophy on an oilfield constitutes a dissimilar, redundant level of instrumentation that makes the existing hardware more reliable. Greater reliability is expected to contribute to greater system uptime by reducing the need for workovers. The proposed model of the process must be tuned to production data for the identification of field-specific parameters. This means that, again, software sensors depend on hardware sensors for the collection of measurement data, but they do not necessarily depend on the availability of a particular piece of hardware, or on the availablity of data during a particular period in time for that matter. A soft sensor may be tuned as long as the available data set is rich enough;

2 VAN DER GEEST, BROMAN, JOHNSON, FLEMING, AND ALLEN OTC 13000 such a data set may be collected whenever it is convenient, e.g., during well tests. We have tested the proposed technology against production data from the BP Troika field in the Gulf of Mexico. Fortunately for the sake of this project, but unfortunately for Troika, the instrumentation on Troika had its ups and downs during the trial period. The results of these tests against live data confirm the potential of data reconciliation as an offshore reliability tool. Soft sensors The basic principle behind soft sensors is that measurement data contains more information than it does at first sight, through knowledge about the process captured in a process model. A very simple example of this principle from everyday life is the speedometer in a car: you typically do not measure speed directly, but you use a tachometer to count the number of revolutions of the axle, and you estimate the speed of the car through a model of the wheel. Similarly, you measure the pressure at several points in an oil well which gives the pressure drop over various sections of the well. Through a flow model the pressure drop may be related to interesting properties like the density of the flow and the flow rate. When you get to think of it, many seemingly hardware sensors (like a speedometer) turn out to be soft sensors. An example from the oil industry of a piece of hardware that is in fact a soft sensor is a venturi downhole flow meter of the type that is being used on Troika. (An overview of the downhole instrumentation in the Troika wells may be found in Fig. 1.) A venturi flow meter is a small computer that collects pressure measurements on both ends of a contracting tube. The computer estimates the flow rate by comparing the observed pressure loss to a model of the pressure loss in the venturi. The soft sensor proposed in this paper, Wellocate, is based on a model of the entire well, that is, a somewhat less well-known environment than the venturi, which is relatively well understood. Nevertheless, it turns out that the model is accurate enough to produce valuable estimates under a variety of operating conditions. The key point in our soft sensor is that we use all measurements in the well simultaneously instead of considering them individually. The combined use of data provides the redundancy that forms the basis for increased robustness. Our well model consists of a state-of-the-art compositional model of the fluid based on the equation-of-state of Peng and Robinson, see Ref. 4, and a multiphase model of the flow through pipes based on first principles, see the work of Barnea et al., Ref. 1. We would like to point out at this stage that there is a fundamental difference between such a model-based method and methods based on interpolation of tables and curves. Interpolation may produce good results in the vicinity of the points that the tables and curves are tuned against. On the other hand, a model based on first principles is expected to represent what is actually happening on the inside instead of what is observed on the outside. Therefore we claim that the model-based method is better suited to reproduce the process over the entire range of operating conditions. A comparative simulation study of our soft sensor against an interpolation method for the estimation of production rates in an oil well may be found in Ref. 3. That paper also includes a detailed description of the model. Field data We have tested the proposed technology on two wells in the Troika field in the Gulf of Mexico, wells TA-1 and TA-5, during a trial period from April 1 st, 2000 until September 15 th, 2000. During these months Troika was made up of five deep water subsea wells producing out of a single reservoir. Fluid data. Both TA-1 and TA-5 produce a relatively light, sweet crude with a relatively high gas-oil-ratio (GOR). Moreover, in July 2000, TA-5 started making water. Our soft sensor is based on a full compositional characterization of the oil based on fluid laboratory experiments carried out on Troika. In addition we have used information about the density of the produced water based on laboratory experiments. All of these experiments were carried out before the wells started to produce. Tubing data. TA-1 has a 13554 ft long tubing string running down from the wellhead on the seabed at 2756 ft true vertical depth (TVD) to 15294 ft TVD. This tubing string consists of an inclined 10873 ft long pipe with an outer diameter (OD) of 5.5 and an inner diameter (ID) of 4.892 running down from the wellhead to 12613 ft TVD, as well as a vertical 2681 ft long pipe with an OD of 4.5 and an ID of 3.920 running down from 12613 ft TVD to 15294 ft TVD. TA-5 has a 14789 feet long tubing string running down from the wellhead on the seabed at 2756 ft TVD to 16462 ft TVD. This tubing string consists of three sections: a vertical 20 ft long pipe with an OD of 4.5 and an ID of 3.920 running down from the wellhead to 2776 ft TVD, an inclined 10749 ft long pipe with an OD of 5.5 and an ID of 4.892 running down from 2776 ft TVD to 13189 ft TVD, and an inclined 4020 ft long pipe with an OD of 4.5 and an ID of 3.920 running down from 13189 ft TVD to 16462 ft TVD. Instrumentation data. The wells on Troika are equipped with downhole pressure and temperature sensors (W0n-DFM-P3 and W0n-DFM-T3, where n = 1,,5). In addition, there are venturi downhole flow meters installed with pressure and temperature sensors at the inlet (W0n-DFM-P1 and W0n- DFM-T1) and in the vena contracta (W0n-DFM-P2 and W0n- DFM-T2). These flow meters give the volume rates of oil and water at standard conditions (W0n-DFM-FQIO and W0n- DFM-FQIH2O). Fig. 1 shows an overview of the downhole instrumentation in the Troika wells. On the well head, there are pressure and temperature sensors upstream (W0n-UP and W0n-UT) and downstream (W0n-DP and W0n-DT) of the production choke. Finally, the position of the production choke is recorded (W0n-PCV-P1). An overview of the instrumentation on the well head of the Troika wells may be found in Fig. 2. In TA-1 the bottom hole gauge is located at 15129 ft TVD

OTC 13000 RELIABILITY THROUGH DATA RECONCILIATION 3 and the downhole flow meter is located at 12768 ft TVD. All instruments on TA-1 were functioning properly at the beginning of the trial period. However, at some point in July 2000 the bottomhole gauge (W01-DFM-P3 and W01-DFM- T3) broke down. Furthermore, since TA-1 has not been making any water throughout the trial period, the downhole flow meter only predicts the flow rate of oil (W01-DFM- FQIO); the flow rate of water (W01-DFM-FQIH2O) is not being used. The bottom hole gauge in TA-5 is located at 16335 ft TVD and the downhole flow meter is located at 13496 ft TVD. At the beginning of the trial period all instruments on TA-5 were operational, except for the choke position (W05-PCV-P1) which has been erratic throughout the entire trial period. Moreover, the flow rate of oil and the flow rate of water (W05-DFM-FQIO and W05-DFM-FQIH2O) must not be read separately but they must be interpreted together as the flow rate of liquid. Finally, at some point in May 2000, the bottomhole gauge in TA-5 (W05-DFM-P3 and W05-DFM-T3) broke down. TA-1 and TA-5 have an identical venturi downhole flow meter with an inlet diameter of 2.7984 and a diameter in the vena contracta (the venturi throat) of 2.1. Production data. For both wells we have identified a number of time intervals when production was relatively stable, at the beginning of the trial period when all instruments were operational, including the downhole flow meter. Logged data from these time intervals may be used as virtual well test data. We have done six of such virtual well tests on TA-1 on April 5 th and 12 th, May 4 th, 9 th, and 15 th, and June 13 th ; we have done four virtual well tests on TA-5 on April 5 th and 12 th and May 4 th and 9 th. The well tests on May 9 th are particularly informative, since both wells were shortly shut in on that day. Test results Case 1: Base case. For tuning purposes we have constructed a base case with full instrumentation for each of the wells. Both of these cases consist of the following building blocks. 1) The source with a known flow rate of oil and a known composition of the oil. (There was no water at the beginning of the trial period.) 2) The bottomhole pressure and temperature sensor with 3) The lower tubing from the bottomhole sensor to the downhole flow meter. 4) The venturi inlet pressure and temperature sensor with 5) The venturi tube. 6) The vena contracta pressure and temperature sensor with known values 7) The upper tubing from the downhole flow meter to the 8) The well head pressure and temperature sensor with Note that the production choke and the sensor downstream of the choke are not being used. Inclusion of a choke model could make the system more flexible and more robust: this may be an interesting extension to the work described in this paper. On the other hand, the choke position on TA-5 (W05- PCV-P1) is erratic. We have also observed some problems with the choke position on TA-1 during the trial period. The roughness of the pipe wall and the overall heat transfer coefficient between the fluid and the formation are two parameters in the tubing model that need to be identified by tuning. It is natural that a model does not give a perfect explanation of reality. Therefore we introduce a factor on the static pressure drop in the tubing model to explain all unmodelled details. Such a fudge factor is perfectly acceptable as long as its value remains close to one, in other words, as long as it captures small discrepancies rather than fundamental flaws in the model. The main parameter in the venturi model that needs to be identified by tuning is the loss coefficient that represents the loss in kinetic energy in the flow as the result of acceleration of the fluid into the venturi. In addition we introduce a pressure offset over the venturi in order to eliminate systematic errors in the pressure sensors. The following procedure is used to estimate the value of the model parameters. 1) Make an initial guess of the parameters. 2) Using these parameters, calculate the mismatch between the observed pressure and temperature on the one hand, and the pressure and temperature calculated with the model on the other hand. 3) If the mismatch in 2) is small enough, then stop. 4) Calculate the gradient of the mismatch in 2) with respect to changes in the parameters. 5) Using the gradient information in 4), make a new guess of the parameters, aiming to minimize the mismatch in 2). 6) Go to step 2). The resulting parameters are listed in Tables 1 and 2. The resulting pressure mismatch between observed and calculated data is listed in Tables 3 and 4. These results show that the model matches the observed behaviour in the well tests with great accuracy. We would like to spend a few words on the difference in roughness between the lower tubing and the upper tubing in TA-5 that may be seen in Table 2. Obviously these values do not represent the actual roughness of the pipe wall: then they would not be so far apart. However, it turns out that when we use these values, the friction model in the soft sensor gives a very accurate description of the loss of energy at the pipe wall over a range of operating conditions. This shows that the friction model is not perfect and should be studied more. However, this does not mean that the current model may not be used with great success, as we will show in the rest of this paper. Apparently you do not need a perfect model to do the job. Case 2: Redundant channel on TA-1. The next challenge for our soft sensor is to run in parallel with fully operational downhole instrumentation in TA-1 at the beginning of the trial period. The goal is to obtain the same rates as the downhole

4 VAN DER GEEST, BROMAN, JOHNSON, FLEMING, AND ALLEN OTC 13000 flow meter at times when we believe that the DFM is right, and to do better at times when we do not believe in the DFM. This case consists of the following building blocks. 1) The source with an unknown flow rate of oil, but with a known composition of the oil. 2) The venturi inlet pressure and temperature sensor with 3) The venturi tube. 4) The vena contracta pressure and temperature sensor with 5) The upper tubing from the downhole flow meter to the 6) The well head pressure and temperature sensor with The following procedure is used to estimate the flow rate of oil. 1) Make an initial guess of the flow rate. 2) Using this flow rate, calculate the mismatch between the observed pressure and temperature on the one hand, and the pressure and temperature calculated with the model on the other hand. 3) If the mismatch in 2) is small enough, then stop. 4) Calculate the gradient of the mismatch in 2) with respect to changes in the flow rate. 5) Using the gradient information in 4), make a new guess of the flow rate, aiming to minimize the mismatch in 2). 6) Go to step 2). It turns out that Wellocate predicts the accumulated oil production from April 1 st through June 30 th with an error of only 0.3% in comparison with the DFM. The results from this test in the month of May are depicted in Fig. 3. Obviously, the soft sensor follows the DFM very accurately, except on May 9th when the well was shut in: the soft sensor goes to zero, whereas the DFM predicts the wrong flow rate. Fig. 4 shows a close-up of the results for the first eight hours of May 24 th. Strangely enough, the soft sensor follows the pressure drop over the DFM much more accurately than the DFM does. These two examples show how our soft sensor may be used with success as a data verification tool for existing downhole equipment. Case 3: Bottom hole pressure sensor in TA-1. The bottom hole sensor in TA-1 got into trouble in July and after a few ups and downs it finally dropped out altogether. We have used our soft sensor with the following set of building blocks to resuscitate the lost sensor. 1) The source with an unknown flow rate of oil, but with a known composition of the oil. 2) The bottomhole pressure and temperature sensor with unknown values (!) 3) The lower tubing from the bottomhole sensor to the downhole flow meter. 4) The venturi inlet pressure and temperature sensor with 5) The venturi tube. 6) The vena contracta pressure and temperature sensor with 7) The upper tubing from the downhole flow meter to the 8) The well head pressure and temperature sensor with The following procedure was used to identify the bottom hole pressure. 1) Make an initial guess of the flow rate of oil and the bottom hole pressure and temperature. 2) Using these variables, calculate the mismatch between the observed pressure and temperature on the one hand, and the pressure and temperature calculated with the model on the other hand. 3) If the mismatch in 2) is small enough, then stop. 4) Calculate the gradient of the mismatch in 2) with respect to changes in the variables. 5) Using the gradient information in 4), make a new guess of the flow rate of oil and the bottom hole pressure and temperature, aiming to minimize the mismatch in 2). 6) Go to step 2). The combination of the flow rate of oil and the bottom hole pressure and temperature in a vector of unknown values leads to the beautiful results depicted in Fig. 5. Wellocate gives very accurate bottomhole pressure readings that coincide with observed values whenever those are available. This is a good example of our soft sensor succesfully replacing permanently installed hardware in case of failure. Case 4: Redundant channel on TA-5. We have run our soft sensor in parallel with the DFM in TA-5 at the beginning of the trial period when all instrumentation was operational. The following set of building blocks was used in this exercise. 1) The source with an unknown flow rate of oil, but with a known composition of the oil. (There was no water at the beginning of the trial period.) 2) The venturi inlet pressure and temperature sensor with 3) The venturi tube. 4) The vena contracta pressure and temperature sensor with 5) The upper tubing from the downhole flow meter to the 6) The well head pressure and temperature sensor with The same procedure as in Case 2 was used to estimate the flow rate of oil. The results of this exercise for the month of April are depicted in Fig. 6. It turns out that Wellocate predicts the accumulated production of oil in April with a difference of only 0.6% as compared to the DFM. Case 5: Water breakthrough in TA-5. On July 27 th TA-5 experienced water breakthrough which was detected at the topside. Unfortunately the bottom hole sensor in TA-5 was unavailable at that stage so that the downhole flow meter was unable to detect the water. Fig. 7 shows the flow rate of water registered by the DFM against the daily production of water (the Test Rate) observed at the topside. Since TA-5 was the only well on Troika to make any water during the entire trial

OTC 13000 RELIABILITY THROUGH DATA RECONCILIATION 5 period, the water rate from this well could be determined from samples on the separator with reasonable accuracy. We have used our soft sensor to estimate the flow rates of both oil and water. The following set of building blocks was used in this exercise. 1) The source with two unknown flow rates (oil and water), but with a known composition of the oil and a known density of the water. 2) The venturi inlet pressure and temperature sensor with 3) The venturi tube. 4) The vena contracta pressure and temperature sensor with 5) The upper tubing from the downhole flow meter to the 6) The well head pressure and temperature sensor with The unknown flow rates of oil and water were estimated by means of the following procedure. 1) Make an initial guess of the flow rates. 2) Using these flow rates, calculate the mismatch between the observed pressure and temperature on the one hand, and the pressure and temperature calculated with the model on the other hand. 3) If the mismatch in 2) is small enough, then stop. 4) Calculate the gradient of the mismatch in 2) with respect to changes in the flow rates. 5) Using the gradient information in 4), make a new guess of the flow rates, aiming to minimize the mismatch in 2). 6) Go to step 2). Fig. 8 shows the estimated flow rate of water against the observed daily production of water. It turns out that Wellocate predicts water breakthrough on the exact same day as TA-5 started making water. The soft sensor also estimates the production volumes of water with a fair amount of accuracy. Note that the soft sensor expects to find both oil and water, but that the model has not been tuned against observed water rates. The ability of the sensor to distinguish water from oil is based on information about the composition of the oil and the density of the water from laboratory experiments that were carried out prior to the trial period. This case shows in what way our soft sensor may be used to supplement and replace existing equipment in determining the value of interesting variables in the process. Concluding remarks The ensemble of available sensors in an oil well turns out to be a sensor on its own, albeit a virtual one. As a rule of thumb, this data reconciliation or consistency sensor provides one additional layer of instrumentation. The results in this paper show that the sensor may be used for the following purposes. 1) Validating hardware measurements, such as measured flow rates, by checking consistency of the data in combination with other measurements. 2) Estimating the value of interesting process variables that are not being measured by existing equipment, such as the density of the fluid or the water cut. 3) Providing a backup for existing equipment, such as a bottomhole pressure gauge, in case of failure. Applications such as these make our soft sensor an interesting reliability tool for instrumentation in offshore oil production. It turns out that our soft sensor is able to simultaneously estimate a number of different kinds of properties, like pressure, temperature, and flow rates. It seems that the sensor provides good estimates as long as the number of unknown variables is less than the number of available sensors; it would be an interesting research topic to investigate if this is a strict bound. In the longer term we would like to move this technology from a method to supplement equipment to a tool to replace permanently installed hardware. This means that we have to qualify this system with the appropriate authorities as part of an alternative philosophy for offshore instrumentation. As a part of this process we need to build up additional confidence in the system by collecting more live results from offshore oil production. Acknowledgements The authors would like to thank the partners on Troika for the kind permission to use field data in this project. The authors thank Svein Morud and Rune Killie at ABB Corporate Research, and Kjetil Stenersen at ABB Automation in Norway for their help with the simulation studies in this paper. References 1. Barnea, D., A unified model for predicting flow pattern transitions for the whole range of pipe inclinations, Int. J. Multiphase Flow (1987), vol. 1, pp. 1-12. 2. Morud, S., and Van der Geest, R., Wellocate field data test, Troll C, well N14, Internal Report NOCRC MOG200011, ABB Corporate Research (2000). 3. Van der Geest, R., Morud, S., and Zaostrovski, A., Oil well allocation: the ultimate interpolation problem, in: Preprints ADCHEM 2000 - International Symposium on Advanced Control of Chemical Processes, IFAC - International Federation of Automatic Control (2000), pp. 443-448. 4. Whitson, C., SPE Phase Behavior Monograph, Society of Petroleum Engineers (1999). SI Metric Conversion Factors (inch) 2.54* E+00 = cm ft 3.048* E-01 = m ft² 0.9290304* E-01 = m² BBL 0.15899 E+00 = m³ psia 6.894757 E+00 = kpa BTU/minute 1.75843 E+01 = W ºR 0.55556 E+00 = K *Conversion factor is exact (ºF 0.55556) - 32.0 = ºC

6 VAN DER GEEST, BROMAN, JOHNSON, FLEMING, AND ALLEN OTC 13000 Appendix: Tables and Figures TABLE 1 TUNING PARAMETERS TA-1 Model Parameter Value Unit Roughness of the wall 4.6E-6 m Lower Heat transfer coefficient 15.0 W/m²K Tubing Factor static pressure drop 1.0178 - Venturi Upper Tubing Pressure offset -0.880 psia Loss coefficient 0.224 % Roughness of the wall 4.6E-6 m Heat transfer coefficient 5.8 W/m²K Factor static pressure drop 1.0067 - TABLE 2 TUNING PARAMETERS TA-5 Model Parameter Value Unit Roughness of the wall 0.866E-6 m Lower Heat transfer coefficient 6.7 W/m²K Tubing Factor static pressure drop 1 - Venturi Upper Tubing Pressure offset -1.033 psia Loss coefficient 4.5 % Roughness of the wall 37.5E-6 m Heat transfer coefficient 6.7 W/m²K Factor static pressure drop 1 - TABLE 3 MISMATCH TA-1 Tag 5/4 12/4 4/5 9/5 15/5 13/6 Unit P1 0.0 0.0 0.0-0.9 0.0 0.0 % P2 0.3 0.7 1.2-1.9-1.9-0.3 % UP -0.2-0.1-0.1 1.4-0.1 0.4 % DFM-P3,DFM-T3 DFM-FQIO DFM DFM-P1,DFM-T1 DFM-FQIH2O PT PT PT DFM-P2,DFM-T2 Fig. 1 - Configuration Downhole Instrumentation UP,UT PCV-P1 DP,DT PT PT Fig. 2 - Configuration Wellhead Instrumentation TABLE 4 MISMATCH TA-5 Tag 5/4 12/4 4/5 9/5 Unit P1 0.5 0.2-0.7 0.0 % P2 1.5 1.0-1.1 0.0 % UP 0.3 0.3-0.1 0 %

OTC 13000 RELIABILITY THROUGH DATA RECONCILIATION 7 05/01 05/08 05/15 05/22 05/29 Flow rate of oil [BOPD] Fig. 3 - Oil production TA-1 Downhole Flow Meter (grey line) 6 00:00 01:00 02:00 03:00 04:00 05:00 06:00 Flow rate of oil [BOPD] 07:00 5.5 5 4.5 4 3.5 Pressure drop DFM [psia] 3 Time Fig. 4 - Oil production TA-1 vs.-pressure drop DFM Downhole Flow Meter (dark grey line) Pressure Drop DFM (light grey line)

8 VAN DER GEEST, BROMAN, JOHNSON, FLEMING, AND ALLEN OTC 13000 P3 [psia] 8500 8250 8000 7750 7500 7250 7000 6750 6500 07/11 07/18 07/25 Fig. 5 - Bottom hole pressure TA-1 Bottomhole Sensor (grey points) 04/01 04/08 04/15 04/22 04/29 Flow rate of oil [BOPD] Fig. 6 - Oil production TA-5 Downhole Flow Meter (grey line)

OTC 13000 RELIABILITY THROUGH DATA RECONCILIATION 9 07/01 07/08 07/15 07/22 07/29 08/05 08/12 08/19 08/26 09/02 09/09 Flow rate of water [BWPD] Fig. 7 - Water production TA-5 Downhole Flow Meter (black line) Test Rate (grey dots) 07/01 07/08 07/15 07/22 07/29 08/05 08/12 08/19 08/26 09/02 09/09 Flow rate of water [BWPD] Fig. 8 - Water production TA-5 Test Rate (grey dots)