Automatic Control of Managed Pressure Drilling

Transcription

1 2010 American Control Conference Marriott Waterfront, Baltimore, MD, USA June 30-July 02, 2010 WeA12.5 Automatic Control of Managed Pressure Drilling Øyvind Breyholtz, Gerhard Nygaard, and Michael Nikolaou* Abstract Managed pressure drilling (MPD) has emerged as a powerful technology for safe and efficient drilling of hydrocarbon wells. However, it poses interesting control challenges that must be addressed before the anticipated impact. These challenges stem from the strongly multivariable nature of the problem, the presence of critical constraints, and the inherent uncertainty in both models and measurements. In this work we give a brief overview of managed pressure drilling, and make the case that model predictive control is a natural approach to controlling MPD. We demonstrate the proposed approach on an example of model predictive control application to dual-gradient drilling, a particular variant of MPD. I. INTRODUCTION Drilling wells for the production of hydrocarbons using offshore platforms is a technically challenging and expensive operation. The drilling operation combines elements of a mechanical system and a unit-operations process. All concerns associated with mechanical systems, such as mechanical integrity and resiliency, vibration control, equipment health monitoring, and maintenance are pronounced with offshore drilling systems. At the same time, process control issues, such as drilling fluid flow and consistency, pressure management in the borehole, and management of surface facilities (such as pumps, mixers, and storage tanks) are equally important, for reasons of efficiency and foremost safety. This is not surprising, given that a drillstring (the long string of thread-joined pieces of pipe hanging from the derrick and transferring motion and weight to the drill bit at the bottom) traverses a few thousand feet through sea water and rock formation. To enhance flexibility, efficiency, and safety, managed pressure drilling (MPD) has emerged as a powerful technology proposition for precise control of wellbore pressure. The SPE/IADC official definition of MPD is "an adaptive drilling process used to more precisely control the annular pressure profile throughout the well bore. The objectives are to ascertain the downhole pressure environment limits and to manage the annular hydraulic pressure profile accordingly". MPD typically employs a closed, pressurized mud circulation system, in contrast to conventional systems, in which mud is returned through an open line at atmospheric pressure. Because MPD treats the Øyvind Breyholtz and Gerhard Nygaard are with the International Research Institute of Stavanger in Norway. Michael Nikolaou is with the Department of Chemical Engineering in the University of Houston, TX, USA. *To whom all correspondence should be addressed. Nikolaou@uh.edu. Tel: mud circulation system as a closed vessel, rather than as an open system, it offers higher flexibility and precision than traditional pressure adjustment based on mud weight and mud pump rate adjustments alone. MPD comes in many variants (Hannegan, 2007), such as PMCD (Pressurized Mud Cap Drilling), CBHP (Constant Bottomhole Pressure), RC (Reverse Circulation), and DGD (Dual Gradient Drilling, several metho). To become possible, MPD requires equipment such as drill string non-return valves (NRV), rotating control device, choke manifold, surface separation system, and various pumps. While such equipment offers unprecedented capabilities, it also generates operability challenges that have to be addressed before widespread acceptance of the technology (Rehm et al., 2008). These challenges emanate, in no small part, from (often non-trivial) interactions among different pieces of equipment and the need for coordinated control of their operation. Operating processes that are well accepted in industry (and mostly based on human intervention) have to be adapted and further developed for use in MPD. More importantly, processes based mostly on humans reach their limits when applied to MPD, given the limited capability of humans to handle reliably situations that involve several interacting variables simultaneously. Inability to handle such situations satisfactorily would result in loss of productive time, which would nullify a main reason for using MPD. A solution to this potential predicament is the use of enabling automation tools. Such tools would provide reliable integration of MPD-related activities, allowing humans to concentrate on higher-level decisions, while leaving the reliable execution of lower-level decisions to automation. MPD automation has been limited in the field to date (Santos and Catak, 2007; Fredricks et al., 2008; Bjørkvoll et al., 2008; Godhavn, 2009) but its potential is growing (Thorogood et al., 2009). Automation has proven its reliability, safety, and ability to outperform humans in oil refining (Saputelli et al., 2005; Saputelli et al., 2006; Nikolaou, 2008) and its potential in traffic management (Godhavn et al., 1996) and aerospace control (Maciejowski and Jones, 2003). The basic principle of this approach is that humans have a crucial role in setting system objectives and constraints, while the underlying automation system ensures that objectives and constraints are met. The overall implications of this approach for making MPD more reliable and effective are discussed in Breyholtz et al. (2010). In this work, we give an overview of an integrated control structure for automatic control of MPD. We discuss the multi-level nature of the problem and provide a structure that identifies specific control tasks and associated variables /10/$ AACC 442

2 An example focusing on one particular variant of this automation approach, namely automated control of bottomhole pressure and hook position in a DGD system is presented. The automation system uses model-predictive control (MPC) as automation technology to coordinate the main mud pump flow rate, subsea pump flow rate, and drill string velocity. II. MANAGED PRESSURE DRILLING The basics of hydrocarbon well drilling are described in standard textbooks and on the internet. A schematic of a generic MPD system is shown in Figure 1. The drilling operation relies on the rotation of drill bit at the bottom of a drill string, hoisted from a derrick on a drill rig at the surface. Controlling the weight placed on the bit by the drillstring is important for ensuring high rate of penetration into the rock formation. It has been shown that weight-onbit control poses significant control challenges (Nikolaou et al., 2005; Awasthi, 2008) Essential for the drilling operation is the pumping of drilling mud through the drillstring. The drilling mud serves multiple important purposes, such as lubrication of the rotating drillstring, removal of the cuttings from the bottom of the borehole through the annulus to the surface, and maintenance of pressure in the borehole within desirable limits (to avoid either migration of flui from the reservoir into the well or damage of the rock formation). The latter is the main reason for MPD. What is different between standard drilling and MPD is the use of a closed annulus and of valves and pumps in addition to the standard main pump. While the configuration shown in Figure 1 offers obvious flexibility by providing a number of manipulated variables by which pressures and flows can be controlled, it also creates obvious control challenges. Because of interactions among all variables, multivariable control is necessary. In addition, because operational constraints (mainly on pressure along the borehole but also on flowrates, to ensure removal of cuttings) are of great importance, model-predictive control is a natural choice for a preferred multivariable control strategy. The structure of a multi-level (cascaded) multivariable control system is discussed in Breyholtz et al. (2010). In this paper we focus on a particular variant of MPD, namely dual-gradient drilling (DGD) with lower riser return system. III. SYSTEM DESCRIPTION The MPD variant we deal with in this work could be typically classified as a dual-gradient drilling (DGD) with low riser return system. DGD is a relatively new drilling technology, which was introduced in the 90 s. The term DGD includes several different drilling concepts, where the common ground is that the pressure profile in the annulus is not linear, but follows a piecewise linear or curved profile, due to the use of flui of varying density along the annulus. Density variation can be based on different dilution-based approaches, such injection of light-weight solid additives to reduce the density, proposed by Maurer Technology (Cohen & Deskins, 2006), and injection of light-weight flui into the riser to lower the efficient fluid density above the injection point, proposed by Louisiana State University (Lopes & Bourgoyne, 1997). Diversion-based concepts, where the mud returns do not travel through a riser, but are either dumped at the seafloor or returned back to the rig through a separate return line have been proposed by Deep Vision (Forrest & Bailey, 2001), SubSea MudLift Drilling Joint Industry Project (Schumacher et al., 2002), Shell (Gonzalez, 1998), AGR (Brown et al., 2007), and Ocean Riser Systems (Fossli & Sangesland, 2006). Common for all DGD concepts is that they use mud with higher than normal density. The system depicted in Figure 2 falls in this category. Figure 2. Dual-gradient drilling flow schematic. Figure 1. Managed Pressure drilling. Model Predictive Control is well developed (Maciejowski, 2002; Mayne, 2000) and a detailed description will be omitted here. In this application MPC coordinates the flow rate q from the main mud pump, the subsea pump flow pump 443

3 rate q sub, and the drill string velocity ν, to control both the bottomhole pressure (BHP) x 1 and the position of the hook x 2. Therefore, the manipulated input vector is u q u = u = q = Main mud pump 1 pump 2 sub Sub sea pump u 3 ν Drill string velocity and the controlled output vector is y1 BHP y = y = 2 Hook position (2) From the above vectors, it can be seen that the system at all times has one additional degree of freedom, which means that in the control application there is one more manipulated input than outputs to be controlled. This is exploited in the control application as an ideal resting value for the main mud pump. This will ensure that the desired flow rate from the main mud pump is obtained during steady state conditions. set set u1 = qpump (3) Enforcing pressure constraints is the primary reason for using MPD. For success of the drilling operation it is critical that the pressure everywhere in the wellbore, p well, must be below the formation fracturing pressure, p frac, and above both the formation collapse pressure, p coll, and the reservoir pore pressure, p res, at all times, namely max( pcoll (, tx), pres (, tx)) pwell (, tx) pfrac (, tx) (4) where x is the position along the open hole section and t is the time. For some drilling operations it can be impossible to reach the target depth with conventional technology without exceeding the pressure window. In the MPC application studied here we focus on pressure at a single location, namely bottomhole pressure (BHP), for which we impose the constraints BHPmin BHP BHPmax (5) The same methodology can be easily extended to include pressure at multiple locations, as discussed in Breyholtz et al. (2009b). In addition, upper and lower boun are placed on the second controlled output (hook position), namely y2,min y2 y2,max (6) Upper and lower boun are also placed on all manipulated inputs and their rates of change. The nonlinear MPC model employed is described in detail in Breyholtz et al. (2009c). The nonlinear solver in the MPC controller is a single-shooting multi-step quasi- Newton method (Meum et al., 2008). IV. SIMULATIONS All simulations in this paper are performed with a highfidelity model (Nygaard & Gravdal, 2007) serving as a virtual process. This model has been proven through several onshore and offshore tests. The test scenario for the (1) controller presented in this paper, is based on a close to vertical 9600m deep well in the Gulf of Mexico. A pressure sensor at the MWD tool will transmit the BHP readings continuously to the surface through a telemetry pipe (Hernandez, et al., 2008). This system has no limitations at times of low or no circulation as mud-pulse telemetry systems experience. All parameters in the low-order model are experimentally fitted to the high-order model before the simulations are carried out, based on step-response experiments. It is conceivable that model parameter identification can also be performed under operating conditions, provided informative enough data are available. The introduced multivariable controller is tested on a case where one segment of the drill pipe is extracted out of the borehole. The operation is performed by changing the set point on the hook from 0 m to 27m. The hook position is included in the controller, and the controller will automatically try to change the hook position to its new position while trying to reduce pressure variations in the bottom hole to a minimum. After the hook is located at 27 m and the pressure has settled to its desired value, the setpoint of the hook is reset from 27 m to its original value of 0 m. Note that four different MPC tunings are considered, corresponding to different selections for the state weighting matrix in the MPC objective. For meaningful comparison between scenarios with and without MPC, the drill string velocity profiles resulting from implementation of MPC are also chosen for scenarios without MPC. Note also the relatively high weighting on the third manipulated input (drill string velocity), which is used to prevent velocity alone from being the dominant manipulation by which to control BHP. TABLE 1. CASE DESCRIPTION Case # 1a 2a 3a 4a 1b 2b 3b 4b Symbol Mode Observed maximum drill string velocity No MPC [ ] u1 = ˆ qpump = 1000 l/ min MPC [ ] 1 = ˆ pump = 1000 / min set set u q l ν

4 V. RESULTS The results from all simulations are presented in Figure 3 through Figure 6. As described in Table 1 the simulations plotted with a dotted line are simulations where the drill string is moved without coordinated control (MPC), and the solid-drawn lines are the simulations with coordinated control. The simulations are also paired in colours according to drill string velocity. Figure 3 illustrates the drill string velocity in, and the velocity is the same for both with and without MPC. Figure 4. Hook position. In Figure 5 the volumetric flow rate from the main mud pump in both cases (with and without MPC) is plotted in [l/min]. For all simulations without MPC the mud pump flow rate is kept stable at 1000 [l/min]. For the cases with MPC, the controller will automatically change the flow rate from the main mud pump to compensate for the drill string induced pressure-fluctuations. Figure 3. Drill string velocity. Positive direction is tripping in to the well. The velocity curves for the drill string in the no-mpc cases (1a-4a) have been selected similar to the MPC cases (1b-4b) to simplify the comparison of the controller performance. Therefore the dotted and solid lines in Figure 3 are selected to be identical. It is important to notice that the driller is not able to set the drill string velocity directly, but can only change the hook position setpoint in the MPC case. The drill string velocity will be a result of the tuning parameters in the weighting matrices for the states and the manipulated inputs, which are selected such that stability and performance are guaranteed. In the no-mpc cases, the driller s setpoint to the drill string position is transformed to a velocity curve to minimize the pressure disturbance. As can bee seen, the velocity curve is a relatively smooth curve, similar to how an experienced driller would adjust the drill string movement. A smooth velocity curve reduces the acceleration of the drill string, which again reduces the pressure disturbances in the wellbore. In Figure 4 the hook position is plotted. This plot is closely related to the velocity plot, with the dotted and solidlines selected to be identical. Figure 5. Volume flow rate from main mud pump. In Figure 6 the volumetric flow rate from the sub sea pump is plotted in [l/min]. For all simulations without MPC the flow rate is kept stable at 1000 [l/min]. It can be observed that the MPC application also actively uses the sub sea to compensate for the drill string movements. Figure 6. Volume flow rate from subsea pump. Mud from this pump is circulated to surface through a separate return line. In Figure 7 the BHP is plotted, and the variations in the pressure with and without multivariable control for the different velocities can be observed. 445

5 every time they feel the operation of the controlled system is in jeopardy. VII. CONCLUSION The ability of MPC to control both hook position and BHP through coordinated manipulation of mud pump flow rate, subsea pump flow rate and drill string velocity, while satisfying various important constraints was demonstrated on computer simulations. Figure 7. Variation in bottomhole pressure during tripping in and out of hole. VI. DISCUSSION Based on previous operational experiences (Nygaard et al., 2007) the drill string movement will induce pressure disturbances if not compensated for. In this study, we initially present the no-mpc solution, which transforms the driller s position signal to a drill string velocity signal that reduces the acceleration of the drill string, thus avoiding pressure spikes. As seen in Figure 7, this velocity curve is not sufficient to avoid large pressure changes. Active compensation of the drill string-induced pressure disturbance by manipulating the main pump is shown to result in much lower pressure changes. Note that small pressure oscillations can be observed in the MPC case. Such oscillations can be reduced by increasing the weight on manipulated inputs in the MPC objective, at the expense of have slower response. In this DGD case, using the level in the riser to compensate for pressure disturbances induced by drill string movement would be too slow to be effective. Therefore, it is essential that the drilling fluid main pump is used actively to compensate for pressure disturbances. The results in Figure 7 show that using MPC offers an effective framework for integrating manipulation of the main pump flow rate with drill string movement for better rejection of pressure disturbances, while maintaining all relevant constraints. In conventional drilling operations, actively using the flow rate of the drilling fluid main pump to compensate for pressure disturbances is not commonly practiced. This is because the pressure transients induced by varying the main pump flow rate, may be too complex to understand and coordinate manually with other variables used for control purposes. Certainly, the idea of transferring control from humans to a computer will be met with scepticism and potentially resistance. This is indeed what happened in the initial application of similar concepts in other industries, such as oil refining (Nikolaou, 2008). However, experience has shown that computers can outperform humans in multivariable control, in a reliable and safe way. A key for acceptance of computer-based automation is to provide enough human-machine communication for human operators to feel comfortable with what the computer is doing and avoid resorting to the manual mode of operation VIII. FUTURE WORK This work demonstrated how MPC can be used as a multivariable control framework for coordinated control of multiple variables in MPC. While this work focused on simultaneous control of two variables, the same framework can be used to integrate control of several more important variables, such as weight on bit and rate of penetration. Additional constraints can be included, such as avoidance of stick-slip situations. Further, MPC variants that rely on better disturbance estimation or model adaptation may also be used. Finally, additional operating scenarios related to drilling will be examined. IX. NOMENCLATURE BHP Bottom Hole Pressure DGD Dual-Gradient Drilling MPC Model Predictive Control MPD Managed Pressure Drilling MWD Measurements While Drilling NPT Non-Productive Time X. ACKNOWLEDGMENT The results presented in this paper are based on work performed in the project MaxWells at the International Research Institute of Stavanger AS (IRIS). The project is funded by StatoilHydro ASA and the Research Council of Norway. REFERENCES Allgöwer, F., Findeisen, R., Nagy, Z. K., Nonlinear Model Predictive Control: From Theory to Application. J. Chin. Inst. Chem. Engrs. 35(3): Awasthi, A. (2008). Intelligent oilfield operations with application to drilling and production of hydrocarbon wells, Ph.D. Dissertation, Chemical Engineering Department, University of Houston. Bjørkevoll, K. S., Molde, D. O., Fjeldberg, H., Utilize Managed Pressure Drilling Equipment and Technique to Cement a Severly Depleted HPHT Reservoir in the North Sea. SPE presented at the SPE Russian Oil and Gas Technical Conference and Exhibition, October. Breyholtz, Ø., Nygaard, G., Godhavn, J.-M., Vefring, E. H., 2009a. Evaluating control designs for co-ordinating pump rates and choke valve during managed pressure drilling operations. In proceedings of the 18 th IEEE International Conference on Control Applications (CCA), St. Petersburg, Russia, 8-10 july. 446

6 Breyholtz, Ø., Nygaard, G., M. Nikolaou (2009b). Deep Water Drilling: Full Pressure Profile Control in Open Hole Section Using Model Predictive Control, IPTC Breyholtz, Ø., Nygaard, G., M. Nikolaou (2009c). Advanced Automatic Control for Dual-Gradient Drilling SPE Annual Meeting paper Breyholtz, Ø., Nygaard, G., M. Nikolaou (2010). Managed Pressure Drilling: A multi-level control approach, Intelligent Energy Brown, J. D., Urvant, V. V., Thorogood, J. L., Rolland, N. L., Deployment of a Riserless Mud-Recovery System Offshore Sakhalin Island. SPE presented at the SPE/IADC Drilling Conference, Amsterdam, The Netherlan, February. Cohen, J., Deskins, G., Use of Lightweight Solid Additives to Reduce the Weight of Drilling Fluid in the Riser. SPE presented at the IADC/SPE Drilling Conference, Miami, Fl, USA, February. Forrest, N., Bailey, T., Subsea Equipment for Deep Water Drilling Using Dual Gradient Mud System. SPE presented at the SPE/IADC Drilling Conference, Amsterdam, The Netherlan, 27 February- 1 March. Fossli, B., Sangesland, S., Controlled Mud-CapDrilling for Subsea Applications: Well-Control Challenges in Deep Waters. SPE Drilling & Completion 21(2): Fredricks, P., Reitsma, D., Runggai, T., Huon, N., Zaeper, R., Backhaus, O., Hernandez, M., Successful Implementation of First Closed Loop, Multiservice Control System for Automated Pressure Managment in a Shallow Gas Well Offshore Myanmar. SPE presented at the IADC/SPE Drilling Conference, Orlando, Fl, USA, 4-6 March. Godhavn, J.-M., Control Requirements for High-End Automatic MPD Operations. SPE presented at the SPE/IADC Drilling Conference and Exhibition, Amsterdam, The Netherlan, March. Godhavn, J.-M., Lauvdal, T., Egeland, O.,1996. Hybrid Control in Sea Traffic Management Systems. Hybrid Systems III Verification and Control. Gonzalez, R., Deepwater Drill String Shut-Off Valve System and method for Controlling Mud Circulation. In US. Patent 6,263,981. Hannegan, D. M., Managed Pressure Drilling. SPE Distinguished Lecturer Series. Hernandez, M., MacNeill, D., Reeves, M., Kirkwood, A., Lemke, S., Ruszka, J., Zaeper, R., High-Speed Wired Drill string Telemetry Network Delivers Increased Safety, Efficiency, Reliability and Productivity to the Drilling Industry. SPE presented at the SPE Indian Oil and Gas Technical Conference and Exhibition, Mumbai, India, 4-6 March. Kaasa, G.-O, A simple Dynamic Model of Drilling for Contro l. StatoilHydro, Research Centre Porsgrunn, Tech. Report. Meum, P., Tøndel, P., Godhavn, J.-M., Optimization of Smart Well Production through Nonlinear Model Predictive Control. SPE presented at the 2008 SPE Intelligent Energy Conference and Exhibition, Amsterdam, The Netherlan, February. Nikolaou, M., Experience from other industries. Drilling Workshop, Rice Unicersity. Nikolaou, M., Misra, P., Tam, V. H. & Bailey, A. D. (2005). Complexity in semiconductor manufacturing, activity of antimicrobial agents, and drilling of hydrocarbon wells: Common themes and case studies. Computers & Chemical Engineering, 29(11-12), Nygaard, G. & Gravdal, J. (2007). WeMod for Matlab user's guide, Technical Report nr 2007/234, International Research Institue of Stavanger AS (IRIS), Bergen, Norway. Nygaard, G. H., Johannessen, E., Gravdal, J. E., Iversen, F., Automatic Coordinated Control of Pump Rates and Choke Valve for Compensating Pressure Fluctuations during Surge and Swab Operations. SPE presented at the IADC/SPE Managed Pressure Drilling and Underbalanced Operations Conference and Exhibition, Galveston, Tx, USA, March. Rehm, B., Schubert, J., Haghshenas, A., Paknejad, A. S., Hughes, J., Managed Pressure Drilling. Gulf Publishing Company, Houston, Texas, USA. Santos, H., Catak, E., First Field Applications of Microflux Control Show Very Positive Suprises. SPE presented at the IADC/SPE Managed Pressure Drilling and Underbalanced Operations Conference and Exhibition, Galveston, Tx, USA, March. Saputelli, L., Nikolaou, M., Economides, M. J., Self-learning reservoir management. SPE Reservoir Evaluation & Engineering 8(6): Saputelli, L., Nikolaou, M., Economides, M. J.,2006. Real- time reservoir management: A multiscale adaptive optimization and control approach. Computational Geosciences 10(1): Schumacher, J. P., Dowell, J. D., Ribbeck, L. R., Eggemeyer, J. C., Planning and Preparing for the First Subsea Field Test of a Full-Scale Dual-Gradient Drilling System. SPE Drilling & Completion 17(4): Thorogood, J. L., Aldred, W. D., Florence, A. F., Iversen, F., Drilling Automation: Technologies, Terminology, and Parallels with other Industries. SPE presented at the SPE/IADC Drilling Conference and Exhibition, Amsterdam, The Netherlan, March. White, F. M., Fluid Mechanics. 4 th Edition. Boston, USA, McGraw-Hill Int. Lopes, C. A., Bourgoyne, A. T., The Dual Density Riser Solution. SPE presented at the SPE/IADC Drilling Conference, Amsterdam, The Netherlan, 4-6 March. Maciejowski, J. M., Predictive Control with Constraints. Prentice- Hall Inc. NJ. Maciejowski, J. M., Jones, C. N., MPC Fault-Tolerant Flight Control Case Study: Flight IFAC Safeprocess Conference Washington DC, 9-11 june. Mayne, D. Q., Rawlings, J. B., Rao, C. V., Scokaert, P. O. M., Constrained model predictive control: Stability and optimality. Automatica 36: