Download fulltext - DiVA Portal
Transcript
Robust control solutions for stabilizing flow from the reservoir: S-Riser experiments Mahnaz Esmaeilpour Abardeh Chemical Engineering Submission date: June 2013 Supervisor: Sigurd Skogestad, IKP Co-supervisor: Ole Jørgen Nydal, EPT Esmaeil Jahanshahi, IKP Norwegian University of Science and Technology Department of Chemical Engineering Robustcontrolsolutionsforstabilizing flowfromthereservoir:S‐Riser experimentsandsimulations MahnazEsmaeilpourAbardeh June26,2013 Preface ThisthesisiswrittenasthefinalpartofmyMasterdegreeinChemicalEngineering attheNorwegianUniversityofScienceandTechnology(NTNU),classof2013. I would like to express my greatest gratitude to my highly knowledgeable supervisor, professor Sigurd Skogestad, for all his helps, his good guidance and encouragements. I am also grateful to my co‐supervisors, professor Ole Jorgen Nydal and PhD student Esmaeil Jahanshahi, who helped and supported me throughout my thesis. It has been a great opportunity and honor for me to be part of your team in generatingnewideasandIamconfidentwhatIhavelearnedthroughthisthesiswillbe surelyusedinpracticeinmyprofessionalcareer. DeclarationofCompliance I, Mahnaz Esmaeilpour Abardeh, hereby declare that this is an independent work according to the exam regulations of the Norwegian University of Science and Technology(NTNU). Dateandsignature: i Abstract One of the best suggested solutions for prevention of severe‐slugging flow conditionsatoffshoreoilfieldsistheactivecontroloftheproductionchokevalve.This thesis is a study of robust control solutions for stabilizing multiphase flow inside the risersystems;throughS‐riserexperimentsandOLGAsimulations.“Nonlinearity”asthe importantcharacteristicofsluggingsystemposessomechallengesforcontrol.Focusof thisthesisisononlinetuningrulesthattakeintoaccountnonlinearityoftheslugging system.Themainobjectivehasbeentoincreasethestabilityofrisersystemsathigher levelsofvalveopeningswithmoreproductionrates. Similarresearchhasbeendonepreviously,butisrepeatedinthisthesisusingnew systematic tuning methods. Three different tuning methods have been applied in this thesis. One is Shams’s set‐point overshoot method developed by Shamsozzhoha (ShamsuzzohaandSkogestad2010).TheotherisIMC‐(InternalModelControl)based tuningmethodwithrespecttotheidentifiedmodelofthesystemfromclosed‐loopstep test.ThelasttuningmethodissimplePItuningruleswithgainschedulingforthewhole operating range of the system considering the nonlinearity of the static gain. The two latter methods have been developed very recently by Jahanshahi and Skogestad (JahanshahiandSkogestad2013). Twoseriesofexperimentshavebeencarriedoutusingamedium‐scaletwo‐phase flow S‐riser loop. A single loop control scheme with riser‐base pressure as the measurementwasused.Therobustnessofdifferenttuningmethodswascomparedby slowlydecreasingtheset‐pointoftheclosed‐loopsystem,whichwastheinletpressure, until instability was reached. The choke valve opening was increasing gradually by decreasing the set‐point. A control with a robust tuning method can maintain system stability in a large range of conditions. The choke valve was then replaced with a quicker valve after the first set of experiments. The same experiments were repeated andtheeffectofcontrolvalvedynamicswasinvestigatedthereafter. The experiments were simulated in OLGA and the same control tests were performed.TheOLGAcasewasconstructedbasedonthefirstseriesoftestswithvalve 1andthedesignedcontrollerswithdifferenttuningstrategieswereapplied.Resultsof theexperimentsverifiedthoseofthesimulations. The tuning method with the highest robustness was thus the one which could stabilizethesystematthelargestchokevalveopening(thelowestinletpressure).The besttuningmethod,withrespecttorobustnessisthesimplePItuningruleswithgain scheduling for the whole operating range of the system. With this method, it was possibletostabilizetheexperimentalrisersystemuptoachokevalveopeningof37% from an open‐loop stability of 16 %. It was also able to stabilize the simulated riser systemuntilachokevalveopeningof75%fromanopen‐loopstabilityof26%. Top side measurements were in general difficult to use in anti‐slug control. Measurement of the topside density using a conductance probe installation was not successful.Therefore,nocascadeanti‐slugcontrolschemescouldbetested. ii iii Contents Preface.....................................................................................................................................................i Abstract.................................................................................................................................................ii 1 Introduction................................................................................................................................1 1.1 2 3 Scopeofthethesis...........................................................................................................................2 Background.................................................................................................................................3 2.1 Multiphasetransport.....................................................................................................................3 2.2 Slugflow..............................................................................................................................................5 2.3 Riserscontainingmultiphaseflow...........................................................................................6 2.4 Riserslugging....................................................................................................................................7 2.5 Anti‐slugoperations.....................................................................................................................11 2.5.1 Choking.....................................................................................................................................11 2.5.2 Gaslift........................................................................................................................................11 2.5.3 Slugcatchers...........................................................................................................................12 2.5.4 Activecontrol.........................................................................................................................12 2.6 Modelingofrisersystems..........................................................................................................13 2.7 Bifurcationdiagrams....................................................................................................................13 2.8 PIDandPIcontrollers..................................................................................................................14 2.9 TuningofPIDandPIcontrollers.............................................................................................15 2.9.1 Method1:Shams’sset‐pointovershootmethodforclosed‐loopsystems..15 2.9.2 Method2:TuningbasedonIMCdesign......................................................................18 2.9.3 Method3:SimpleonlinePItuningmethodwithgainscheduling...................22 Experimentalwork................................................................................................................26 3.1 SetupDescription..........................................................................................................................27 3.2 Equipment........................................................................................................................................30 3.2.1 Mainwaterstoragetank....................................................................................................30 3.2.2 Airreservoirtank.................................................................................................................31 iv 4 3.2.3 Airbuffertank........................................................................................................................32 3.2.4 Overflowtank.........................................................................................................................33 3.2.5 Pressuretransmitters.........................................................................................................33 3.2.6 Smallseparator......................................................................................................................34 3.2.7 Centrifugalwaterpump.....................................................................................................34 3.2.8 Airflowmeter........................................................................................................................35 3.2.9 Waterflowmeter..................................................................................................................35 3.2.10 Chokevalves.......................................................................................................................37 3.2.11 Conductanceprobe(C)..................................................................................................38 3.2.12 LabVIEW..............................................................................................................................39 Simulationofexperimentalcases....................................................................................41 4.1 OLGA®,multiphasesimulationtool.......................................................................................41 4.2 Constructionofthecase.............................................................................................................41 4.2.1 Flowpathgeometry.............................................................................................................42 4.2.2 Fluidproperties.....................................................................................................................43 4.2.3 Boundaryandinitialconditions.....................................................................................43 4.2.4 Numericalsetting.................................................................................................................44 4.3 5 Resultsanddiscussion.........................................................................................................46 5.1 Experimentalresults....................................................................................................................46 5.1.1 Seriesofexperimentswithvalve1(slowchokevalve)........................................47 5.1.2 Seriesofexperimentswithvalve2(fastchokevalve).........................................57 5.1.3 CascadeControlusingtoppressurecombinedwithdensity............................68 5.2 ComparisonofSlowvalveandFastvalve...........................................................................70 5.3 SimulatedresultsfromOLGAmodel.....................................................................................74 5.3.1 Open‐loopsimulations.......................................................................................................74 5.3.2 Controlbytrialanderror..................................................................................................75 5.3.3 Tuningthecontroller..........................................................................................................79 5.4 ImplementingPIDcontrollerinOLGA.................................................................................44 Comparisonofexperimentalandsimulatedresults......................................................95 5.4.1 Open‐loopbifurcationdiagrams....................................................................................95 5.4.2 ComparisonofcontrolresultsfromIMC‐basedtuningmethod......................96 5.4.3 ComparisonofcontrolresultsfromSimpleonlinetuningmethod................96 5.4.4 Comparisonoftuningmethods......................................................................................97 v 6 7 8 Discussionandfurtherworks........................................................................................100 6.1 Tuningmethods..........................................................................................................................100 6.2 Controlstructures......................................................................................................................101 6.3 Discussableissuesrelatedtoexperimentalactivities................................................102 6.3.1 Oscillationsinflowrates................................................................................................102 6.3.2 Waterflowbackintothebuffertank........................................................................102 6.3.3 Leakageinsteelconnection..........................................................................................102 Conclusion..............................................................................................................................104 7.1 Stabilizingcontrolexperimentsusingbottompressure...........................................104 7.2 TestingonlinetuningrulesonS‐riserexperiments....................................................104 7.3 Controlusingtoppressurecombinedwithdensity....................................................105 7.4 Investigatingeffectofcontrolvalvedynamics..............................................................105 7.5 ControlsimulationsusingOLGA..........................................................................................106 References..............................................................................................................................107 A.LowpassfilterinLabVIEW..............................................................................................109 B. SimulatedresultstogetthebeststeptestsforShams’smethod.....................110 C. SomeexamplesofMATLABscripts..............................................................................111 vi vii 1 Introduction MultiphasepipelinesareacommonfeatureofoffshoreproductionintheNorthSea. Theyconnectsubseawellstothetopsideprocessingfacilitiesortheplatforms.Inmany points of transportation, these pipelines get the shape of L‐shaped or S‐shaped risers. The stability of multiphase flow inside these pipeline‐riser systems is of great importance and many efforts have been spent on this issue so far. In low reservoir pressuresorlowflowrateconditionstheliquidphasestendtoaccumulateinlowpoints and form liquid slugs. This leads to the pipeline or riser blockage and can be more dangerous when the length of slugs is comparable to the length of the riser. This phenomenon is called Severe slugging (also Terrain slugging or Riser slugging) and is characterizedbylargeoscillatoryvariationsinpressureandflowrates(Storkaas2005). These large variations lead to a poor separation, unwanted flaring and even a plant shutdownintheworstcase. Reducingopeningofthetopsidechokevalvehasbeenatraditionalwaytosuppress severe slugging. However, this increases the valve back pressure and therefore decreasestheproductionratefromthewell. Activefeedbackcontrolofthetopsidechokevalvecanmakeitpossibletostabilize theflowattheconditionswherenormallyseveresluggingispredicted.Thisreducesthe need for additional topside equipment and allows a higher rate of oil recovery. The control system is called anti‐slug control and its main objective is to keep the multiphaseflowasstableaspossiblebymanipulatingthetopsidechokevalveusingthe parameterssuchaspressureordensityasthecontrolvariables. In the way of developing new technologies for stabilizing control of severe slugginginrisersystemsmanyresearcheshavebeendoneattheNorwegianUniversity ofScience andTechnology.Thework hasbeenguided bySkogestad(Skogestad2003; Storkaas 2005; Shamsuzzoha and Skogestad 2010; Jahanshahi and Skogestad 2011; Skogestad and Grimholt 2011; Jahanshahi and Skogestad 2013) and performed at the department of Chemical Engineering. Storkaas (Storkaas 2005), Sivertsen (Sivertsen 2008), Jahanshahi (Jahanshahi and Skogestad 2011) and numerous master students haveworkedonmodelingandcontrollingofrisersystems. 1 Companies like ABB (Havre, Stornes et al. 2000), Statoil and Total have all researched prevention of slugging and built installations at offshore locations. Statoil completed in 2001 their first slug control installation at the Heidrun oil platform. Siemens is also involved in slugging research and funds a PhD program, which this thesisisconnectto. Intheanti‐slugcontrolsystem,itisveryimportantthatthecontrollersarefine tuned. Otherwise, the control system is not robust in practice and the closed‐loop systembecomesunstableafteraplantchange.Thesluggingsystemishighlynonlinear sincethegainchangesat differentoperatingpoints.Forsuchasystemthecontrollers needtoberetunedateachoperatingpoint. 1.1 Scopeofthethesis Inthisthesisthreedifferenttuningmethodswillbetestedwithexperimentsand simulations to find the most robust solution for anti‐slug control system. High robustnesswillbeobtainedifthesystemcanmaintainstabilityatlargedeviationsfrom openloopconditions.Thismeanslargechokevalveopenings.Thetuningmethodsare systematicandhavebeendevelopedveryrecently(ShamsuzzohaandSkogestad2010; JahanshahiandSkogestad2013). TheexperimentsofthisthesiswillbecarriedoutatthedepartmentofEnergyand ProcessEngineering.Twoseriesofexperimentswillberunusingamedium‐scaletwo‐ phase flow S‐riser loop. The difference between the two series is the type of choke valve.Theaimistoinvestigatetheeffectofcontrolvalvedynamicsonperformanceof the control system in addition to robustness of the tuning methods. Possibility of differentcontrolstructureswillbealsoinvestigated. The experiments will be simulated in multiphase flow simulator, OLGA, and the same control tests will be performed. Finally the simulated and experimental results willbecompared. 2 2 Background 2.1 Multiphasetransport Whenitcomestooffshoreproductionofoilandgas,longtransportofmultiphase flowhasrecentlybecomeofgreatattention.Manypipelinesandrisersarecarryingthe combination of natural gas, condensate, oil and water from the North Sea to shore. Previously, large production platforms equipped with process facilities were built on the sea floor with the aim of separating gas, oil and water. Today this can be too expensive and multiphase transportation can save billions of dollars for the oil companiesinstead. Design and operation of multiphase transportation systems raise many new challenges. These challenges could be either related to the flow, fluid or the pipe integrity. Pressure drop/ boosting, Slugging, liquid emulsion, temperature change, scaling, hydrate and wax formation can be examples of them. Overcoming these challengesandhavingasafeanduninterruptedmultiphaseflowreferstotheterm“flow assurance”. This term was first used by Petrobras in the early 1990s and it originally referred to only thermal hydraulics and production chemistry issues encountered duringoilandgasproduction(Fabre,Peressonetal.1990). One important issue in flow assurance is stabilizing the multiphase flow inside the pipeline‐riser systems. From a control engineering point of view, this can be referred as control of the disturbances in the multiphase flow as the feed to the separationprocess.Avoidingvariationsintheflowenteringtheprocessingunit,atthe outlet of the multiphase pipelines is the issue of interest for control (Bratland 2010). The ability of predicting the flow patterns and reserving a stable flow is of great importance, which is the objective of the thesis. Figures 2.1 and 2.2, adapted from Bratland (Bratland 2010), describe possible flow patterns inside the horizontal and verticalpipelines. 3 Figure2.1:Gas‐liquidflowregimesinhorizontalpipes. Figure2.2:Gas‐liquidflowregimesinverticalpipes.Slugflowisthepointofinterestin thethesis. Changingmultiphaseflowbetweendifferentflowregimescanbedescribedbya typical flow regime map shown in figure 2.3, adapted from Taitel (Taitel 1986). The boundariesbetween stableand unstableregionsare clearlyshownin theflowregime map. With applying feedback control these boundaries can be moved and thereby the stableregioncanbeincreased. 4 or multiph hase flow (Taitel ( 198 86). Stabilitty boundarries are Figure 2.3: Stabillity map fo clearlyshownintthemap. 2.2 Slu ugflow w A Among th he flow asssurance concerns, c managemeent of slu ugging in system Fard et al.. 2005). deliverrability hass received m much interrest in receent years (Godhavn, ( Slugflo owisoneo oftheflowp patternsch haracterizeedbyaltern natingslugsofgasandliquid flowingginthepip pes.Inthis typeofflo owregime, elongated bubblesoffgasseparratedby “slugs” ofliquid,ttravelfrom moneendo ofthepipettotheotheerend.Itcaanbeeitherrdueto differen ntvelocitieesofgasan ndliquidphasewhich hisreferreedashydro odynamicsslugging or pipeeline geom metry which h is referreed as terraain induced d slugging. The latterr one is commo on in riserss and its main m reason n is the gravity. A schematic map m of slug flow is shown infigure2 2.4,adapted dfrom(Yan nandChe2 2011).The masterunfavorableeeffectof slug flo ow is its instability i that has a a negative impact on n the operration of offshore o producction facilitties. The periodic p oscillations o of liquid and a gas ph hases due to t their inhomo ogeneous distribution d n cause oscillatory pressures p and decreeases the level l of producction as laarge as 50 0%. The average of these osccillations iss lower th han the equilibrium prod duction an nd this giives the production p n losses. More M overr these oscillattions can damage th he pipe an nd the separation process. p Fo or these reasons, r 5 suppressingtheslugflowisofdominantimportance.Ahomogeneoussteadyflowwith very small bubbles of gas well distributed in the continuous liquid phase is most desired.Insuchdesiredsituation,thepressureremainsconstantovertime. Figure 2.4: Schematic map of slug flow in a vertical pipe in a slug unit (Yan and Che 2011) 2.3 Riserscontainingmultiphaseflow Risers are a special type of pipeline developed for vertical transportation of materialsfromseafloortoproductionanddrillingfacilitiesonthewater'ssurface.They canbeintypesofrigidrisers,flexiblerisersandhybridrisersthatisacombinationof the rigid and flexible. Risers can have many different configurations. However in this thesisalltheS‐shapedtypesarethepointofinterestregardlessoftheirdifferences.The functionalsuitabilityandlongtermintegrityoftherisersystemaffectstheselectionof riserconfiguration(Bai2001).Figure2.5showsprevalentriserconfigurations. 6 Figure2.5:Commonriserconfigurationsappliedintheoilandgasindustry(Bai2001) 2.4 Riserslugging Riser slugging (also called severe slugging/ terrain induced slugging) is the toughest type of slugging happening in a pipeline‐riser system where a downward inclinedpipelineisconnectingintoanupwardriser.Storkaas(Storkaas2005)explains the cyclic behavior of riser slugging illustrated schematically in figure 2.6. It can be broken down into four steps. Step 1: Slug formation: gravity causes the liquid to accumulateinthelowpointandaprerequisiteforseveresluggingtooccuristhatthe gas and liquid velocity is low enough to allow for this accumulation. Step 2: Slug production:Theliquidblocksthegasflow,andacontinuousliquidslugisformedinthe riser.Aslongasthehydrostaticheadoftheliquidintheriserincreasesfasterthanthe pressuredropovertheriser,theslugwillcontinuetogrow.Step3:Blowout:Whenthe pressure drop over the riser overcomes the hydrostatic head of the liquid in the slug, theslugwillbepushedoutofthesystemandthegaswillstartpenetratingtheliquidin the riser. Since this is accompanied with a pressure drop, the gas will expand and furtherincreasethevelocitiesintheriser.Step4:Liquidfallback:Afterthemajorityof theliquidandthegashaslefttheriser,thevelocityofthegasisnolongerhighenough to pull the liquid upwards. The liquid will start flowing back down the riser and the accumulationofliquidstartsagain. 7 Figure 2.6:Graph hicalillustrrationofa slugcycle (Yanand Che2011)).Slugform mationis shown in part 1. Slug produ uction is sh hown in paart 2, Blow wout in parrt 3 and liq quid fall backin npart4. Severeslugggingcausesperiods ofnoliquiidorgasproduction intotheseeparator followeed by very y high liquiid and gass rates, wh hen the liq quid slug iss being produced. Lengthofliquidsslugscanb beseveralttimestheleengthofth heriser.Th hisphenom menonis highlyu undesirablle.Thelarggeliquidprroductionm mightcauseeoverflow andshutd downof thesep parator.Flu uctuations ingasprod ductionmigghtcauseo operationalproblemssduring flaring,,andthehighpressurefluctuattionsmighttreduceth heproductioncapacity yofthe field(Jaansen,Sho ohametal. 1996).Itccanreduceeoperatinggcapacityfforseparattionand compreession unitts. The red duced capaacity is cau used by th he need off larger op perating margin nstohandleethelargerrdisturban nces.Largerrdisturban ncesrequirrealargerb back‐off fromth heoptimaloperationp point,andtthusreducingthethroughput(SStorkaas20 005). occurintw wodifferen ntmodesoffIandII.IIntypeIoffsevere Severeslugggingcano sluggin ng the liquiid fully blo ock the ben nd while in n type II th here is a partial p blocckage at bend and a gas paasses throu ugh. The type t I is characteriz c zed by larrge oscillattions in pressurreandacceeleratedblo owout.Infactthepreessureosciillationsreflectstaticheadof 8 theriser.TherearesmallpressureoscillationsintheseveresluggingoftypeIIandthe sluglengthisshorterthantheheightoftheriser.Butflowoscillationscanbelarge.Type IIsluggingisnotusuallycriticalforastableoperation.Figures2.7and2.8,adaptedfrom Malekzadeh (Malekzadeh, Henkes et al. 2012), illustrate SS1 and SS2 respectively. Figure 2.7 is based on a measured cycle of the riser P for SS1 corresponding to U SL 0.20 ms 1 and U SGO 1.00 ms 1 .Figure2.8isbasedontheexperimentalcycleforthe riser P ofSS2correspondingto U SL 0.10 ms 1 and U SGO 2.00 ms 1 . Figure 2.7: Stages for SS1 (a) a graphical illustartion (b) marked on a cycle of an experimental riser P trace ( U SL 0.20 m s 1 and U SGO 1.00ms 1 ) (Malekzadeh, Henkes et al.2012) 9 Figure 2.8: Stages for SS2 (a) a graphical illustartion (b) marked on a cycle of an experimentalriser P trace( U SL 0.10 ms 1 and U SGO 2.00 ms 1 )(Malekzadeh,Henkesetal. 2012) 10 2.5 Anti‐slugoperations Asthefieldsbecomemorematurethemoreadvancedtechnologyisdemanded. Thereasonisthattheenergyofreservoirdecreasesduetoitsaging.Thisleadstolower pressureandtemperaturesinreservoir.Thelowerpressureofreservoircauseslimited drivingforcetotheflowandtherebylowerphasevelocitiesinresultandfinallymore probable riser slugging formation. Low temperatures also increase the probability of solidformation.Changingthedesignofpipe‐risersystemtoavoidsluggingcannotbe economicallyfeasible.Themostcommonmethodsforavoidingsluggingarepresented below. 2.5.1 Choking Schmidt et al. (1979) first suggested choking (decreasing the opening Z) of the valveattherisertopasaneliminationwayofsevereslugging.Thetheorybehindthis suggestionisthatthesteadyflowisgainediftheaccelerationofthegasabovetheriser is stabilized before reaching the choke valve (Jansen, Shoham et al. 1996). This increasesthebackpressureandthevelocityatthechokethereupon.Themechanismis explainedasapositiveperturbationintheliquidholdupinapipeline‐risersystemwith astableflowwillincreaseweightandwillcausetheliquidto“falldown”.Theresultof this is an increased pressure drop over the riser. The increased pressure drop will increasethegasflowandpushtheliquidbackuptheriser,resultinginmoreliquidat thetop of the riser than prior totheperturbation. With a valve opening larger than a certaincriticalvalue(Zcrit)toomuchliquidwillleavethesystem,resultinginanegative deviation in the liquid holdup that is larger than the original positive perturbation. Thus,wehaveanunstablesituationwheretheoscillationsgrow,resultinginslugflow. ForavalveopeninglessthanthecriticalvalueZcrit,theresultingdecreaseintheliquid holdupissmallerthantheoriginalperturbation,andwehaveastablesystemthatwill returntoitsoriginal,non‐sluggingstate(Storkaas2005). 2.5.2 Gaslift Gaslifthasbeensuggestedasanothermethodofeliminatingsevereslugging.In thismethodthehydrostaticheadoftheriserisreducedwithgasinjectionandthusthe pipelinepressurewillbereduced.Theinjectedgasliftstheliquidtowardsuptheriser. If sufficient gas is injected the liquid will be continuously liftedand a steady flow will occur.Thedrawbackofgasliftisthelargegasvolumesneededtoobtainasatisfactory stabilityoftheflowintheriserandthisistooexpensive(Storkaas2005). 11 2.5.3 Slugcatchers Oneotherwaytoaccommodatesluggingiscommontoinstallalargeseparator asaslugcatcherattheexitofthepipeline.Theslugcatcheristhefirstelementinthe processing facility and determining its proper size is vital to the optimal operation of theentirefacility.Thefundamentalpurposeofslugcatcheristoremovefreegasfrom the liquid phase and to deliver a relatively even supply of liquid to the rest of the production facility. An advantage of this set‐up is that inspection and maintenance on the slug catcher can be done without interrupting the normal operation. There are mainlytwotypesofslugcatchers,thevesselandthemultiple‐pipetypesandtheuseof each type depends on the type of flow stream. Multiple‐pipe separators have been widelyappliedingas‐condensateprocessingfacilities(Miyoshi,Dotyetal.1988). Installingslugcatchershasseveraldrawbacks;itputsalowerboundontheoperating pressureofthepipe,whichagainlimitstheflowfromthereservoir.Italsoincreasesthe mechanical wear of the pipeline due to large oscillations in pressure. The capital and maintenancecostsofaslugcatcherarerelativelylarge(Olsen2006). 2.5.4 Activecontrol Risersluggingcanbepreventedusingstabilizingfeedbackcontrol.Anapproach basedonfeedbackcontrolwasfirstproposedbyShmidt(Shmidtet.al.1978).Theidea ofpaperwastosuppressterrainsluggingbyusingthetop‐sidechokevalveandasimple feedback loop, measuring pressure at the inlet and upstream the riser or the top pressure before the choke valve as inputs. With feedback control, the stability of the flowregimescanbechangedtoenhanceoperation.Infacttheboundariescanbemoved via feedback control, thereby stabilizing a desirable flow regime where riser slugging “naturally”occurs(Storkaas2005).Anti‐slugcontrolcanmove theboundariesinflow regime map resulting in increased stable region. It sounds to be one of the best solutions for prevention of severe‐slugging. Several models have been suggested by researchers to describe the system dynamics and several controllers have been designed.Themodelsaremeanttoaidtuningofcontrollerswhichusetheproduction chokevalveastheactuatorandtrytostabilizethesystemwithamoreproductionrate in a higher valve opening. The objective could be defined as obtaining the most robustness for the system against large inflow disturbances. “Nonlinearity” as the important characteristic of slugging system provides some challenges for control. However, a good control system using a model that is most consistent with the plant couldhavegoodresultsinachievingdesiredstableflowregimes. 12 2.6 Modelingofrisersystems Themainobjectivesofmodelingofproductionflowinpipelinesandrisersareto predict the pressure drop, the phase distributions, the potential for unsteady phase delivery(slugging) and the thermal characteristics of the system (Pickering,Hewitt et al. 2001). The reliability of these simplified models is however questionable. The analysisandmodelingofmultiphaseflowsreliesheavilyonempiricalcorrelationsand thepredictionsforthemodelsareonlyasreliableastheempiricaldataonwhichthey arebased.Thereforeitcanbequestionedwhetherthemodelswouldbevalidifapplied torealsystems.Theyaretestedbytheuseofsmalldiametersexperimentalrisersand maybemorethangoodenoughforsuchsystems,buttheystillmaybeinvalidforusein largersystems(Pickering,Hewittetal.2001). The tuning methods used in this work are provided via linear and nonlinear multiphase flow models based on the mass balances over the different sections of the pipeline‐risersystem.Thesimplifiedfour‐statemechanisticmodelmadebyJahanshahi andskogestad(JahanshahiandSkogestad2011)usessimplerelationshipstocalculate thephasedistributionsoverthedifferentsectionsofsystem.Themodelhasbeenthen linearizedaroundanunstableoperatingpointandafourth‐orderlinearmodelwithtwo unstable poles, two stable poles and two zeros is produced. Since a model with two unstable poles is enough for control design, the model order is reduced by using balancedmodeltruncationviasquarerootmethod.Thisidentifiedmodelofthesystem is then used for an IMC (Internal Model Control) design and finding new IMC‐based tuningrules.(JahanshahiandSkogestad2013).Moreover,asimplemodelforthestatic nonlinearity of the system is proposed by Jahanshahi and based on this static model, simplePItuningrulesconsideringnonlinearityofthesystemaregiven(Jahanshahiand Skogestad2013).Thesetuningruleshavebeenusedinthesimulationsandexperiments ofthisthesisandaclearcomparisonoftheresultshavebeenpresented. 2.7 Bifurcationdiagrams Bifurcation diagrams have been used in this thesis in order to plot the values of pressureversusthevaluesofvalveopeningforthesluggingsystemeitherinopen‐loop positionorinclosed‐looppositionwithdifferentcontrollers.Bifurcationdiagramsare thesimplestwaytoillustratethestabilityofthesystem.Inthestableregionstheplot consistsofaunitcurveshowingtheexactvalueofthepressure(insimulations)orthe average of very small pressure oscillations (in experiments) while in the unstable regions the plot consists of three curves, one for steady state conditions and the two others showing the maximum and minimum of oscillations at each value of valve openingovertheworkrangeofchokevalve. 13 2.8 PIDandPIcontrollers PI(proportional‐integral)andPID(Proportional‐integral‐derivative)controlare oftheearliercontrolstrategies.ThePIDcontrollerincludestheproportionalaction(P), integralaction(I),andderivativeaction(D).Thecontrollerusestheerrorsignal e ( t ) to generatetheproportional,integral,andderivativeactions.Amathematicaldescription ofthePIDcontrolleris: 1 u (t ) K p [ e ( t ) Ti de(t ) e ( ) d ( ) T ] d 0 dt t Equation2.1 Where u ( t ) istheinputsignaltotheplantmodel.Theerrorsignal e ( t ) isdefined as e (t ) r (t ) y (t ) and r ( t ) isthereferenceinputsignal(Fabre,Peressonetal.1990). AfteraLaplaceTransformthecontrollercanbeshownas: c K c (1 1 s) Is d Equation2.2 Where Kc , I and d are the respective tuning parameters for the P, I and D actions. PI and PID controllers are the most widely used controllers in the industry. However,theyneedtobetunedappropriatelyforrobustnessagainstplantchangesand large inflow disturbances. (Jahanshahi and Skogestad 2013) Thus finding the most appropriateamountsof Kc , I and d couldbeextremelyrequired.Atypicalstructureof aPIDcontrolsystemisshowninFigure2.9. Figure2.9:AtypicalPIDcontrolstructure 14 2.9 TuningofPIDandPIcontrollers Many tuning methods for different systems have been introduced so far by researchers andengineers.Dependingon thecharacteristics ofthe system(plant), for instancenonlinearityandstability,differentlevelsofrobustnessisachievedbydifferent tuningmethods.Threedifferenttuningmethodshavebeenappliedinthisthesis.Twoof them are quite new and have been recently developed (Jahanshahi and Skogestad 2013).Theyarespecifiedforthesluggingsystem.Infact,thisthesisisaverificationof thesenewmethods. 2.9.1 Method1:Shams’sset‐pointovershootmethodfor closed‐loopsystems Some systems like slugging system are originally unstable in open‐loop. For these systems model from closed‐loop response with P‐controller can be used to find the appropriate tuning parameters. A method called “Shams’s set‐point overshoot method” was first constructed by Shamsuzzoha et al. (Shamsuzzoha and Skogestad 2010).Skogestadetal.(Skogestadand Grimholt 2011)developedthismethodfurther intoatwo‐stepclosed‐loopprocedure.Astepbystepdescriptionofthetwostepclosed‐ loopShams’smethodispresentedbelow. Theclosed‐loopsystemwithP‐controllershouldbeatsteady‐stateinitially,that is,beforetheset‐pointchangeisapplied.Then,aset‐pointchange, y s ,isapplied.The step change and the P controller gain ( Kc0 ) should be adjusted in a way that the overshoot (D) is approximately 30 %. Figure 2.10 shows a graphical illustration and equation2.4findstheovershoot. 15 w P‐only y controller (Skogesttad and Figure 2.10: clossed‐loop seet‐point reesponse with olt2011) Grimho onfromtheegraphicallsteprespo onse: Extracttinformatio Timetofirrstpeak: t p Maximumoutputchaange: y p Relativestteadystateoutputchaange: Alternativeely, y caanbeestim matedfromequation2 2.6,usingth heoutputch hange y atfirstund dershoot( yu ): y 0.45( y p yu ) Overshoott: D y p y y Equattion2.4 Steadystatteoffset: B Equatiion2.3 y s y y 16 Equattion2.5 Theparameter(A): A 1.152 D 2 1.607 D 1 Equation2.6 Theparameter(r): r 2A B Equation2.7 Thefirstorderplusdelaymodelparameters: Steadystategain: k 1 Kc0 B Equation2.8 Delay: t p (0.309 0.209 e 0.61r ) Equation2.9 Timeconstant: 1 r Equation2.10 Nowafirstorderplusdelaymodelisfoundandwithrespecttothismodel,the tuningparametersare: 1 1 Kc . k c I min 1,4 c Equation2.11 Equation2.12 InthepaperbySkogestad(Skogestad2003),itwasrecommendedtouse c asagoodcompromisebetweenperformanceandrobustness. 17 2.9.2 Method2:TuningbasedonIMCdesign The Internal Model Control (IMC) method was developed by Morari. et.al. (Morari and Zafiriou 1989) The method supposes a model, states desirable control objectives,and,fromthese,proceedsinadirectmannertoobtainboththeappropriate controllerstructureandparameters.Fortheobjectivesandsimplemodelscommonto chemical process control, the IMC design procedure leads naturally to PID‐type controllers,occasionallyaugmentedbyafirst‐orderlag.(Rivera,Morarietal.1986) Consider the block diagram for the IMC structure (See figure 2.11). Here, g is model oftheplantthatingeneralhassomemismatchwiththeplant. g c isinverseof minimumphasepartof g andf(s)isalow‐passfilterforrobustnessoftheclosed‐loop system. Thegoalofcontrolsystemdesignisfastandaccurateset‐pointtracking: y y s t , d Equation2.13 Efficientdisturbancerejection: y ys d t , d Equation2.14 andinsensitivitytomodelingerror. Figure2.11:Theinternalmodelcontrol(IMC)structure 18 JahanshahiiandSkogeestaddonotusethis configurattionfortheeunstable system; instead dtheyuseaanequivaleentasshow wninfiguree2.14,wherre: gc f C cf 1 gg on2.15 Equatio Figuree2.12:Clossed‐loopsy ystemwithIMCcontro oller(Jahan nshahiandSkogestad2013) They prop pose onlinee identificaation of lin near modell by a clossed‐loop sttep test. TheydesignanIM MC(InternaalModelControl)bassedonthe identified model.Theen,they usetheeresulting IMCcontro ollertoobttaintuningparameterrsforPIDaandPIconttrollers. A summ mary of th heir work, which thiis thesis has h been done d based d on that, will be presenttedbelow. 2.9.2.1 1 Mo odelIdenttification use the sttep test To identify y process model ( g ), Jahansh hahi and Skogestad S informaation in a a closed‐loo op stable system to o do online model identificatio on. The suggesttedmodelh hastwoun nstablepoleesandisintheformo of: b s b0 g ( s ) 2 1 s a1s a0 on2.16 Equatio meters, b1 , b0 , a1 and a 0 needto obeestimattedbyinfo ormationexxtracted Fourparam fromcllosed‐loopstepresponse.Jahansshahiusesasystematticmannertofindtherelated fourpaarameters.Inhismeth hodtheloo opisclosed dbyaprop portionalco ontrollerw withgain K C 0 , to o get the closed‐loop c p stable sysstem. For closed‐loop c p transfer function frrom the set‐point to the output o onee similar to o the modeel used by Yuwana. et.al. e (Yuwaana and Seborgg1982)isco onsidered: K 2 1 z s Gcl ( s) 2 2 s 2 s 1 19 on2.17 Equatio Thefourparameters( K2 , z , and )areestimatedbyusingsixdata( y p , yu , y , ys , t p , and t ) observed from the closed‐loop response (see figure 2.10). Then, they use a systematic procedure to back‐calculate the parameters of the open‐loop unstablemodelinequation2.16(JahanshahiandSkogestad2013). 2.9.2.2 IMCdesignforunstablesystems TodesigntheIMCcontroller(C),theidentifiedmodel( g )isusedastheplant model. g ( s ) ks b1s b0 s a1s a0 s 1 s 2 2 Equation2.18 g c ( s ) 1 / k s 1 s 2 s Equation2.19 Theyalsodesignthefilter f ( s ) forrobustnessofthesystemasexplainedby Morari.et.al.(MorariandZafiriou1989).Thefilterisinthefollowingform: f ( s ) 2 s 2 1 s 1 n s 1 Equation2.20 λisanadjustablefiltertime‐constant.Thecoefficients 1 and 2 arecalculated bysolvingthefollowingsystemoflinearequations: 1 2 3 1 1 1 1 2 2 13 1 2 1 2 2 Equation2.21 Filteronlyactstothederivativeaction. FinallytheresultingIMCcontrollerisfoundasthefollowing: 1 2 k 3 2 s 1s 1 C ( s) s s 20 Equation2.22 2.9.2.3 PIDandPItuningbasedonIMCcontroller Jahanshahi writes the IMC controller of equation 2.22 in form of a PIDF controller and propose the tuning parameters based on that. PIDF is a PID controller whichalow‐passfilterhasbeenappliedonitsderivativeaction. K K s K PID ( s) K p i d s f s 1 Equation2.23 Wherethetuningparametersare: f 1/ Equation2.24 Ki f k 3 Equation2.25 K p K i 1 K i f Equation2.26 K d K i 2 K p f Equation2.27 AnimportantpointtobeconsideredintuningofPI/PIDcontrollersbasedonIMC design is choosing an appropriate . It must be chosen in a way that the required followingconditionsaresatisfied: Kp 0 Kd 0 21 Equation2.28 Equation2.29 APIcontrollerhasbeenalsoobtainedbyreducingtheorderofIMCcontrollerto1. K PI ( s ) K c (1 1 ) Is Equation2.30 Andthesuggestedtuningrulesare: Kc 2 k 3 Equation2.31 I 2 2.9.3 Equation2.32 Method3:SimpleonlinePItuningmethodwith gainscheduling One main part of the thesis is tuning the controller by a new method called “SimpleonlinePItuningrules”proposedbyJahanshahiandSkogestad(Jahanshahiand Skogestad 2013). One advantage of this method is that Nonlinearity of the slugging system has been considered when providing the tuning rules. Gain of the slugging system changes drastically for different operating conditions and as the source of nonlinearity,makescontrolofthesystemdifficult.Themethodconsiststwoparts: First,asimpleMATLABstaticmodelforthestaticnonlineargainisidentifiedat eachoperatingpoint(valveopening). Then,theidentifiedmodelateachoperatingpointisusedandsimplePItuning rulesbasedonsinglesteptestbutwithgaincorrectiontocounteractnonlinearityofthe systemareproposedasfunctionsofvalveopening. In this method of tuning, Jahanshahi and Skogestad have used gain‐scheduling withmultiplecontrollersbasedonmultipleidentifiedmodels.TheMATLABmodeland theobtainedPItuningrulesforeachcontrollerwillbeexplainedbelow. 22 2.9.3.1 SimpleMATLABstaticmodel ThesimplemodelforanL‐shapedriserconsideringstaticnonlinearitywasmade by Jahanshahi (Jahanshahi and Skogestad 2013). The model is based on the mass balancesanditcalculatesthephasedistributionsoverthedifferentsections.Thismodel neededtobemodifiedforanS‐shapedrisertobeusedinthethesis. A good assumption of valve equation is very important in using the simple model.Thereasonisthattheslugginggainofthesystemasafunctionofvalveopening, is derived based on this equation. Jahanshahi assumes the valve equation as the following: w K pc f ( z) p Equation2.33 Wherewistheinletmassflowratetotheriser, K pc isthevalveconstantand f ( z ) isthecharacteristicsofthevalvewhichisdefinedasthefollowingforthelinear valveusedinexperiments: Equation2.34 f ( z) z andasfollowsfortheOLGAvalvemodelinsimulations: f ( z) z.cd 1 z 2 .cd 2 Equation2.35 p isthepressuredropoverthevalveandasitisclearinthevalveequation,it’s afunctionofvalveopeningthatcanbewritteninthefollowingform: 2 p 1 w K pc . f ( z ) Equation2.36 Thenthesimplemodelfortheinletpressureis: Pin p Pfo 23 Equation2.37 Pfo istheinletpressureatfullyopenpositionofthevalveandhasbeencalculated fromthebelowequation: Pfo P * p * Equation2.38 P* isalargeenoughinletpressuretoovercometheriserslugging: P * L . g . Lr Ps Pv ,min Equation2.39 Here L isthedensityofliquidwhichiswaterinoursystem. g isthegravityand Lr is the length of riser. Ps is the separator pressure in downstream and Pv ,min is the minimum pressured drop over the valve and has been considered zero in the simulations. p * is the pressure drop over the valve at the critical valve opening of the system(bifurcationpoint). Then based on the above equations, the static gain of the slugging system is derivedasafunctionofvalveopeningbydifferentiating Pin withrespecttoZ.Finallythe simplemodelforthestaticgainofthesystemis: k ( z) 2.9.3.2 Pin z Equation2.40 SimplePItuningrulesbasedonidentifiedMATLABmodel Jahanshahi and Skogestad (Jahanshahi and Skogestad 2013) then perform a closed‐loop step test with a P‐only controller at the initial valve position of Z 0 . The parameter ( ) is then calculated by using data ( y p , yu , y , t p , and t ) observed fromtheclosed‐loopresponse(seefigure2.10)andthestaticmodelgiveninequation 2.40. y yu ln y p y 2 t y y K c 0 k ( z0 ) p y Equation2.41 4t p 24 Where Kc0 istheproportionalgainusedforthesteptest.ThesuggestedPItuning parametersasfunctionsofvalveopeningaregivenasthefollowing: K c ( z0 ) Tosc k ( z0 ) z 0 / z * Equation2.42 I z 0 3Tosc ( z 0 / z * ) Equation2.43 Tosc istheperiodofsluggingoscillationswhenthesystemisinopen‐loopposition and z * isthecriticalvalveopeningoftheopen‐loopsystem(wheresluggingstarts). 25 3 Experimentalwork ControlofSevereSluggingandcreatingastableflowregimebyapplyingcontrol using new online tuning methods has been verified in this thesis. Air‐water Sever slugging control experiments in S‐shaped riser has been one of the main parts of this thesisinadditiontomodelingandsimulations.Aseriesoftestshavebeenconductedat a medium scale setup located in NTNU multiphase flow laboratory at department of Energy and Process Engineering (See figure 3.1). It has been tried to evaluate the applicability of three tuning methods explained previously in different conditions. Experimentsinthisissueandcomparingthemwithsimulatedresultsarealsovaluable inthewayofapprovingpredictionofsimulations. Theexperimentalworkincludetryingtwodifferentchokevalveswithdifferent dynamics as the actuator and running series of control experiments for each valve separately. Series of control experiments have been in the following order: First the open‐loop experiments have been run in order to make the open‐loop bifurcation diagramofthesystem.ThenaP‐onlycontrollerhasbeenusedtoclosetheloopandthe set‐point step change test has been run with the aim of finding appropriate tuning parameters. Finally, after calculating different tuning rules based on the data of step change test, closed‐loop experiments were run and the closed‐loop responses of differentcontrollerstunedwithdifferentmethodswereevaluated.Buffertankpressure (riserinletpressureintherealsystems)hasbeenselectedasthecontrolvariable(CV) inseriesofcontrolexperiments. Moreover cascade control experiment using topside pressure combined with outflowdensityasthecontrolvariableshasbeentried. 26 Figure 3.1: Medium scale experimental setup of multiphase flow laboratory located at departmentofEnergyandProcessEngineeringofNTNU 3.1 SetupDescription Thethree‐dimensionaloverviewofthemultiphaseflowrigusedtoperformthe seriesofexperimentsinthisthesisisshowninfigure3.2.Theflowloopwasconsisting ofwaterandcompressedairsupply. 27 ut,NTNU(Lilleby200 03) Figurre3.2:MulttiphaseTesstRigLayou Figure 3.3 shows a schematic overview of the exp perimental setup witth more oragewateerandpresssurized details..Thewholesystemissplacedatttwolevelss.Largesto air tan nks (T1 an nd T2) and d water pu ump (P1) were placed at baseement. Flow lines continu uedtolab‐llevelandaallflowmetters,contro olvalves,h horizontalttestsection nandS‐ riserw wereplaced datthislev vel.Theflo owlineofttestwithin nnerdiameeterof50m mmwas connecctedtoamixer/inletssectioncon ntainingtheair/wateersupplyan ndthemulltiphase flow w was forced up the S‐rriser. The air a buffer tank (T3) was installled upstreeam the mixing point to increase th he air volum me and em mulate a long pipeline. The air volume shouldbelargeen noughtofo orcetheliqu uiduptheriserandcausesluggiingtooccu ur. A Asoneoftthemostim mportanteq quipment, chokevalv ve(V)was mountedaattopof the S‐riser. It was used as the t control actuator for controlling the in nlet pressu ure/ top pressurreandoutlletflowden nsityastheecontrolvaariables.Itwasalsopossibletoaadjustit manuallly while running thee system in n open‐loop p position. Pressure transmitters (PT1 andPT T2)andtheeconductan nceprobeaasthedenssitymeter (C)werein nstalledat various placesiinthesetup p,andwereeusedtoco onstructanumberofdifferentccontrolstru uctures. A After the S‐riser, S air and waterr were enttered into an overflow w tank (T4 4), then movedintoasmaallseparato or(T5)thro oughalarggeflexiblep pipemadeo ofhoses,an ndwere separatted there. The waterr is then reeturned fro om the tesst section back b to thee water largesttoragetank kinthebassement.Theeairisventedoutwitthoutfurth hertreatmeent. Thedimen nsionsoftheexperimeentalsetup pareillustratedinfigu ure3.4.Theelength scaleissgiveninm meters. 28 Figure3.3:MediumscaleTestRigLayoutwithmoredetails,NTNU Figure3.4:ConfigurationoftheS‐shapedrisertestsection(Lilleby2003) 29 3.2 Equipment Inthissectionpropertiesandpurposeofthemainequipmentaregiven.Allthe pipes,bendsandotherconnectionsaremadeofacid‐proofsteel,AISI316L.Thisisthe casefortheentirepipinguptothetestsections.Thevalvesaremadeoftreatedbrass, andarequiteresistanttocorrosion. 3.2.1 Mainwaterstoragetank Water is filled in a separator (T1). It is a 3 m 3 acid proof tank placed in the basement.Fromtheseparator,waterispumpedthroughtheinfrastructure,intothetest sectionandreturnedtotheseparatoragain. Figure3.5:Mainwaterstoragetanklocatedinbasement 30 3.2.2 Airreservoirtank Theairsupply(T2)isconnectedtothecentralhigh‐pressuresupply.Thissupply is a pressure vessel made by Nessco and gives a pressure of 6‐7 bars, which is then reduced through a pressure reduction valve to the operational pressure of (usually) approximately3bars. Figure3.6:Airreservoirtanklocatedinbasement 31 3.2.3 Airbuffertank Theairbuffertank(T3)withavolumeof200litersandthetype“DN50flange” has been made by the company “Laguna”. It is installed before the mixing point. To make slugging possible, a large pipe volume for pressure buildup is necessary. The buffertankisusedtoemulatethislargepipevolume.Themaximumpressurethebuffer tankcanwithstandislimited.Forsafety,thetankhasbeenequippedwithasafetyvalve, toensurethatthepressurenotwillexceed3Bars. Figure3.7:Airbuffertank 32 3.2.4 Overflowtank An overflow system is made to achieve pressure dependent liquid flow. It is a ventedsteeltank(T4)filledwithwater.Flexiblepipesconnectthetanktotheseparator. Abypassflowwillflowintothetankandbacktotheseparatorandmaintainaconstant liquidlevelinsidethetank.Thepressureattheoverflowtankwillbeconstantequalto the hydrostatic pressure of the liquid column from the tank. This will simulate a constant reservoirpressure and make the inflow to the test section dependent onthe inletpressure.Thesupply pipesfortheplasticoverflowtank aresmall,so itwillonly workproperlyiftheflowthroughitisverylow. Figure3.8:Overflowtankattopofriser 3.2.5 Pressuretransmitters Pressure transducers (PT1 and PT2) made by Siemens were installed on the buffer tank and riser to measure the buffer pressure and top pressure respectively. Theyhaveaworkingrangeof0‐4bars. 33 3.2.6 Smallseparator The flow from the overflow tank (T5) is moved into a small separator located downthehosespipe.Apictureoftheseparatorisshownunderneathinfigure3.9.The airfromtheriserisreleasedfromthetopoutlet.Thebottomoutletisusedforthewater recycleandreturnsthewatertothewaterstoragetank. Figure3.9:Smallseparator 3.2.7 Centrifugalwaterpump A large centrifugal water pump (P1) of the type DN100 flange made by Wilo Norge AS was used topush the water into the system. In order to prevent water flow oscillationsthecentrifugalwaterpumpwasruninaveryhighlevelofpower(80%of themaximum).Howevertogetthedesiredflowrateofwaterwhichwasnothigh(0.39 kg/sec)thewatercontrolvalvewasopeninsmallvalues,instead. 34 Figure3.10:Centrifugalwaterpump 3.2.8 Airflowmeter The vortexflow meterof type DN40 wafer manufactured by JF Industrisensorer wasusedtomeasuretheairflowrate(FIT1.01).Thenumberthatitgavewasintheunit of Kg/hour and needed to be converted into the desired unit (kg/sec). It was located upstreamtheairbuffertank.Theworkingrangeoftheairflowmeterwas5‐2180kg/h. 3.2.9 Waterflowmeter The Electro‐magnetic water flow meter of type 1/2'' union, manufactured by JF Industrisensorer was located upstream of the mixing point (FIT2.01). It has a working rangeof0.19‐6.4 m 3 /h. 35 Figure3.11:Airflowmeter Figure3.12:Waterflowmeter 36 3.2.10 Chokevalves Two different choke valves (V) have been used in this thesis and the series of experimentshavebeenrunwithboth.Firstaslowvalvewasusedastheactuatortorun the control experiments and then it was replaced with a fast valve. The effect of their dynamics was then investigated. They are angle seat valves located on the top of the riserupstreamoftheseparator.Thechokevalveisoperatedbypressurizedair(4bars) supplied from the pressurized air system in the laboratory, through the valve positioner. The specifications of the old slow valve were not available, while the specificationsofthefastvalveareasfollows: Manufacturer:ASCO Diameter:2inch Material:StainlessSteel Operation:NC(NormallyClosed) PilotPressure:4‐10bar MaximumWorkingPressure:6bar OperatorDiameter:90mm Signal:4‐20mAmp OpeningTime:2sec ClosingTime:2.5sec Figure3.13:Chokevalves;left:Fastvalve,Right:Slowvalveanditspositioner 37 3.2.11 Conductanceprobe(C) In the second series of experiments with the fast valve a cascade control structurewasusedwithoutflowdensityandthetoppressureasthecontrolvariables. Conductance probe was applied to measure the density of the outflow from the riser. The probe has been calibrated by Kazemihatami (Kazemihatami 2012) very recently. Theoutputoftheprobewasintheformofvoltage.Thecalibrationcurvepresentedby Kazemihatamiwasusedtofindtherelationbetweenvoltageandholdup.Equation3.1 showsthisrelation.HmeansholdupandVmeansvoltage. H 0.9857V Equation3.1 Thedensityofmixedflowisfoundfromtheequation3.2: m water . H air . 1 H Equation3.2 Afterinsertingtherelatedvaluesintheaboveequation,thedensityofmixedflow isfoundasafunctionofvoltage: m 984.513V 1.204 Equation3.3 Figure3.14:Theconductanceprobe 38 3.2.12 LabVIEW The Laboratory Virtual Instrumentation Engineering Workbench (LabVIEW) softwaredevelopedbyNationalInstrumentswasusedforinstrumentationcontroland data logging. The user interface is illustrated in figure 3.15. The pressures, flow rates and valve position could be monitored directly from the interface. In addition it was possible to run the loop manually by manipulating choke valve opening, or automatically by setting tuning parameters for PID/PI/P controllers. Some modificationswereappliedincaseofcontrol.Twomodesofcontrolwereimplemented in the program; a single mode and a cascade mode. The single mode used buffer pressureascontrolvariableandthecascademodewasusingtoppressureandoutflow densityascontrolvariables.Aschematicviewofcontrolmodesarepresentedinfigure 3.16. Figure3.15:LabVIEWuserinterface 39 Figure3.16:ImplementedcontrolmodesinLabVIEW 40 4 Simulationofexperimentalcases 4.1 OLGA®,multiphasesimulationtool OLGA® (OiL and GAs simulator) is a commercial multiphase flow simulator widelyusedintheoilandgasindustry.Itsolvesmanynumericalequationstosimulate theflowbyconsideringthesystemdynamicsandoffersheatandmasstransfermodels. TheexperimentalcasewasconstructedinOLGA.Thedesignedcontrollerswith differenttuningstrategieswereusedandtheresultswerecompared.Inordertofitthe OLGAmodelwiththeMATLABmodelsandexperimentssomeoftheparameterswere manipulatedwithinlimitedranges.OLGA®version7.1wasusedforthesimulations. In this chapter the case construction with implementing the S‐shaped riser geometry, fluid properties, numerical settings and boundary conditions is explained stepwise. 4.2 Constructionofthecase Establishment of a good case with appropriate particular items such as fluid properties,numericalsettings,initialandboundaryconditionsandflowpathgeometry, wastheinitialstepforsimulationprocess.The“S‐risersimple”casemadebyJahanshahi (Jahanshahi and Skogestad 2011) was basically used for the open‐loop simulations. Some improvements and modifications were applied after the file was received. For open‐loop simulations the modifications were in terms of numeric and for theclosed‐ loopsimulationstheywererelatedtoimplementingthePIDcontrollerintothecase.In termsofnumericsomeIntegrationparametersweremanipulatedinPropertieswindow oftheprogram. 41 4.2.1 Flowpathgeometry The “S‐riser simple case” with a geometry based on the experimental set‐up at the Department of Energy and Process Engineering was used. The reason to use such geometry is that the simulation results are to be compared with the experimental resultsinthethesis.Theexactgeometryispresentedintable4.1. The X‐Y coordinates have been calculated with respect to table 4.1and the resultinggeometryhasanoverviewofthefigure4.1. Accordingtotheexperimentalsetupinmultiphaseflowlaboratory,thesources ofairandwaterareplacedinthebeginningandtheendofthebuffertankrespectively. Table4.1:ThegeometryoftheS‐riserexperimentalset‐up Pipe L[m] D[m] [ ] out in 1 8.125 0.20 ‐45.0 2 3.000 0.05 ‐10 3 6.050 0.05 ‐4.0 4 1.200 0.05 ‐1.8 5 1.106 0.05 ‐1.8 ‐61.8 6 4.110 0.05 61.8 7 0.709 0.05 61.8 ‐32.0 8 2.160 0.05 ‐32 9 1.716 0.05 ‐32 79.0 10 1.820 0.05 79.0 11 1.150 0.05 90 42 S‐risergeometry 7 6 5 y[m] 4 3 2 1 0 ‐1 ‐5 0 5 x[m] 10 Figure4.1:GeometryofS‐riserinOLGA 15 4.2.2 Fluidproperties AllfluidpropertieshadbeenwritteninPVTfileby(JahanshahiandNilsen2012). Itisatableofphasecompositionsatdifferenttemperaturesandpressuresandismade by a program called PVT‐Sim. By specifying temperature and pressure limits and the compositions of the fluids involved, the program calculates the values for the phase compositions.Heattransferandtemperaturechangewerenotimportantinsimulations due to experimental condition. Water was assumed as an incompressible flow. Heat transfer and temperature related properties such as enthalpy or entropy were filled withdummynumbers. 4.2.3 Boundaryandinitialconditions Thetypesoftheairandwatersourcesasinletnodsweredefinedasinletmass flow. The flow rates were fixed for all simulations. The volume fractions were establishedto1forbothnodessinceonlywaterorairwasinjectingthroughthenode. The outlet nod type was selected to pressure type and it has been set to atmospheric pressure. 43 4.2.4 Numericalsetting Thenumericalsettingspecificationssuchassimulationtimeandtimestepwere adjusted in different numbers from case to case. This is due to the diversity of phase velocityindifferentcases. 4.3 ImplementingPIDcontrollerinOLGA InordertoimplementaPIDcontrollerinOLGAfirstapositivecheckvalvewas placedright afterthe water source in pipe2, section 1 of the case. The reason wasto make sure that the flow will move only in the defined direction. Then a pressure transmitterwaslocatedinpipe2,section2thatistheinletoftheriser,rightafterthe buffertank.Itwasaimedtomeasurethebufferpressureandsendthepressuresignal into the PID controller. The PID controller was used in a way that it received the measurementsignalfromthepressuretransmitterandsenttheoutputsignalintothe chokevalvelocatedattopoftheriser(Pipe8,section3).Chokevalvescanbesimulated byselectingtheHydrovalveforthevalvemodelinOLGA. Figure 4.2: OLGA case with PID controller. The controller receives the measurement signal from thepressure transmitter and sends the output signal into the choke valve locatedattopoftheriser. When applying a PID controller in OLGA several specifications need to be established by user, depending on the desired conditions and results. The more importantspecificationsthathavebeenmanipulatedmanytimesduringsimulationsare the PID parameters and the time varying specifications. When it comes to PID 44 parametersinpropertywindowofthesimulator,AMPLIFICATIONreferstothegainof thecontroller;BIASisthedesiredinitialoutputvalue(itwasusedasthedesiredvalve openinginoursimulations);DERIVATIVECONSTisthetimeconstantforthederivative action and INTEGRALCONST is the time constant for the integral action. As the time varying specifications the MODE was set to AUTOMATIC and the SET‐POINT values werechangedfromonesimulationtoanother. 45 5 Resultsanddiscussion The purpose of this chapter is to present the results from experiments and simulationsandaclearcomparisonofthem.Theexperimentalresultsfromtwoseries ofexperimentsusingaslowandafastchokevalvewillbepresentedinsection5.1.The effort of cascade control experiment using top pressure combined with density as measurements and the faced issues has been also mentioned there. Section 5.2 evaluatestheeffectofcontrolvalvedynamicsthroughcomparingresultsofslowvalve withthoseoffastvalve.Thesimulatedresultswillbeexplainedinsection5.3.Insection 5.4 the experimental results are compared with simulated results. In section 5.5 the threedifferentusedtuningmethodshavebeencomparedandthebesttuningmethod hasbeeninvestigated. 5.1 Experimentalresults The operating procedures and the results from experimental activities done at NTNUmultiphaseflowlaboratoryarediscussedinthissection. Foreachseriesofexperimentswithvalve1(slowvalve)orvalve2(fastvalve) the open‐loop system with basic conditions would be explained first. Then the procedureofimplementingclosed‐loopsteptestandcalculatingthetuningparameters byusingdifferenttuningmethodswillbediscussed.Theresultsoftuningintheformof tuningrulesareexplainedthereafter.Finallytheclosed‐loopresponsesusingcalculated tuningparameterswillbepresentedasthemainresultsoftheexperimentalwork. 46 5.1.1 Seriesofexperimentswithvalve1(slowchoke valve) Theexperimentalworkinthisthesisstartedwithusingslowchokevalveasthe actuator.Thegoalwas torepeatthe sameseriesoftestswithaslow andafastchoke valveandthenevaluatetheeffectofcontroldynamicsonthefinalresults. 5.1.1.1 Open‐loopexperiments Thestartingpointintheexperimentswasrunningtheloopinmanualmode.The testswererunindifferentvalveopeningswithfixedliquidandgasflowrateswhileno controllerwasimplementedinthesystem.Itwasaimedtopresentthesystembehavior innaturalconditionswithoutcontrol.Theinflowconditionsandtherelatedbifurcation diagramarepresentedbelow. 5.1.1.1.1 Inflowconditions The applied fixed flow rates have been wl 0.3927 [kg / sec ] for water and wg 0.0024 [ kg / sec ] for air (See figure 5.1.) These flow rates correspond to U sl 0.2 [m / sec] and Usg 1 [m / sec] astheliquidandgassuperficialvelocities.Thewaterflow rate could be set in lab view by adjusting the pump frequency and the control valve, whiletheairflowrateneededtobesetwithamanualvalveinthepathoftheflow.The reasonwasthatthecontrolvalvefortheairwasbroken.Themanualvalvewasfarfrom thescreenandthismadeitdifficulttoobtaintheexactflowrate. The water flow rate was not also easy to set. Large variations in the flow rate were eliminated by running the pump with a high frequency and opening the control valve in a small value. The more opening the choke valve, the more slugging the flow regime and the more unstable the flow rates were resulted. In the following series of experiments,aconstantflowrateofairandwaterwasused.Asaresult,thewaterand air flow rates needed to be readjusted when the valve opening in open‐loop was changed.However,whenusingacontrollerinclosed‐loopmode,itwasconsiderednot to be reasonable to readjust the inflow conditions. Figure 5.1 compares variations of flowratesandpressureintwodifferentvalveopenings. 47 ‐3 Basisconditionwith20%valveopening AirFlowrate[kg/sec] x10 2.4 2.2 2 0 100 200 300 400 WaterFlowrate[kg/sec] 0.38 0.36 0.34 0.32 0 100 200 300 400 200 198 196 194 192 190 0 100 200 time[sec] 300 2.6 x10 400 Basisconditionwith100%valveopening 2.4 2.2 2 0 50 100 150 200 250 300 350 50 100 150 200 250 300 350 50 100 150 200 time[sec] 250 300 350 0.5 0.45 0.4 0.35 0 BufferPressure[kPa] BufferPressure[kPa] WaterFlowrate[kg/sec] AirFlowrate[kg/sec] ‐3 2.6 200 180 160 140 120 100 0 Figure5.1:Illustrationofbasisopen‐loopconditionsincaseofflowratesandpressure. The left series of plots are illustrating the system with valve opening Z=0.2 that is related to the stable region while the right side plots present the system with valve openingZ=1thatisrelatedtotheunstableregion.Largeoscillationsareclearsignsof instabilityatZ=1. 5.1.1.1.2 Bifurcationdiagram The experiments were started with the valve opening of Z=0.2. Then the valve was open stepwise until it was fully open. The results of buffer pressure were logged and the related bifurcation diagram was plotted, presented in Figure 5.2. The critical stabilitypoint(thebifurcationpoint)isthemaximumchokevalveopeningthesystem canhavewhilebeingstable.Inthepresentedbifurcationdiagram,thetoplinetracksthe maximum values of pressure at each operating point, the bottom line presents the minimumvaluesofpressureandthemiddlelineshowstheaveragevaluesofthebuffer pressureatdifferentvalveopenings.Asclearinthefigurethecriticalstabilitypointwas foundtobeatapproximately26%chokevalveopening(Z=0.26). 48 OpenloopBifurcationDiagram‐Valve1 210 200 InletPressure[Kpa] 190 180 170 160 150 140 130 120 110 0.2 0.3 0.4 0.5 0.6 Z 0.7 0.8 0.9 1 Figure5.2:Open‐loopbifurcationdiagramfromtheslowchokevalveexperiments.The bifurcationpointoccursatvalveopeningofZ=0.26.Thetopandbottomlineillustrate themaximumandminimumvaluesofoscillationsforinletpressurerespectivelyateach operatingpoint.Themiddlelineshowstheaveragevaluesofpressure. 5.1.1.2 Closed‐loopsteptest Inordertoapplyeachoftuningmethodstogetanappropriatecontrollerforthe sluggingsystemaclosed‐loopsteptestisrequiredwithastepchangeinset‐point(the bufferpressure).Todothisitwastriedtocontrolthesystembytrialanderror.AP‐only controllerwasselectedandastheinitialguessforthegain,abigvalueof100wastried. Thereasonwasthattheset‐pointvaluewasasmallnumber(pressureinbars)andthe gainhadtobeselectedinawaythatitcouldchangetheoutput(Z)inalargerangeafter a small change in set‐point. Increasing the gain resulted in a more stable flow with smaller pressure variations or smaller amplitude of slugs. Finally a high value of K c 0 220 wasselectedtoperformthesteptest.Set‐pointwasmanipulatedtogetthe averagevalveopeninghigherthan0.26andtheobtainedvalueof0.29wassatisfying.It was aimed to do the test in a region that is unstable in open‐loop position. After the system was stabilized, four step tests were implemented and data were logged. The relatedspecificationsarepresentedintable5.1andtherelateddiagramsareshownin figure5.3. 49 Table5.1:Closed‐loopsteptestspecificationsrunwithslowchokevalve I Kc0 Initialset‐point Finalset‐point 1.52 1.72 1.73 1.54 1.54 1.73 1.49 1.70 Test_1 Test_2 220 Test_3 Test_4 Test‐1 1.65 1.6 Setpoint Data 1.55 1.5 0 100 200 1.7 1.65 1.6 Setpoint Data 1.55 1.5 0 200 400 time[sec] 1.6 1.4 Setpoint Data 200 600 1.7 1.6 Setpoint Data 1.5 0 600 400 Test‐4 1.8 1.8 0 300 Test‐3 1.75 Inletpressure[bar] Inletpressure[bar] 1.7 Test‐2 2 Inletpressure[bar] Inletpressure[bar] 1.75 200 400 600 time[sec] 800 Figure 5.3: Presentation of different tests of set‐point step change for a closed‐loop feedback experiment with a P_only controller using inlet (buffer) pressure as control variable.Test‐4showsthebestcharacteristicsincaseofdesiredovershootandsteady stategainrequiredfortuningthecontroller. Afterevaluatingdatafromsteptestsitseemedthatthelastone(test_4)has bettercharacteristicscomparedtotheotherswithrespecttothepointthataunitstep test was going to be used for all tuning methods. It was decided to use test_4 in the tuning of controller by different methods. Some important considerations in selecting thebeststeptestwere: 1. ForthesteptesttobeusedinShams’smethodtherecommended0.3overshoot wasdesired. 50 2. The steady state gain of the system must be smaller than one ( y 1 ) to be ys usedinIMC‐basedtuningmethod. Since the response was noisy, a low‐pass filter in MATLAB from the type of Simple infiniteimpulseresponsefilterwasusedtoreducethenoiseeffect.Asmoothingfactor of 0.001 was used to smooth the signal as well as required ( 1 means no filtering).Figure5.4illustratesthestepresponseusedinthetuningmethods. 175 InletPressure[Kpa] 170 165 160 155 Setpoint Data Filtered 150 145 0 100 200 300 400 500 time(sec) 600 700 800 900 Figure 5.4:Set‐pointstep changefor aclosed‐loopfeedback experiment withaP_only controller using inlet (buffer) pressure as control variable. A low pass filter with a smoothingfactorof 0.001 wasusedtoremovethenoiseeffectfromtheresponse. 5.1.1.3 Tuningthecontroller The tuning methods explained in section 2.9 have been used to tune the controllerusingbuffer(inlet)pressureasthecontrolvariableandslowchokevalveas the actuator. The tuning procedure and the related results are explained in the following. 51 5.1.1.3.1 TuningbyShams’sclosed‐loopmethod The first method to be used for tuning of the controller was Shams’s method developed by Shamsuzzoha (Shamsuzzoha and Skogestad 2010). In order to tune by Shams’smethod,explainedinsection2.9.1theinformationfromthesteptestexplained inprevioussection(Seefigure5.4)wereused.Then,theovershootwascalculatedand the appropriate tuning parameters were found. Table 5.2 shows the resulted tuning parameters by Shams’s method. Kc0 is the initial gain used in the step test, Kc is the calculated proportional gain, and I is the integral tuning parameter. The system has beenconsideredasafirstorderplusdelaymodel. Table5.2:TuningparametersfromSham’smethodforthesluggingsystem Kc0 Z ave Overshoot Offset Kc I 220 0.29 0.3846 0.6501 121.5189 224.3679 Itwastriedtocontrolthesystembytherelatedtuningparametersseenintable 5.2.Yet,thementionedtuningparameterscouldn’twork;meaningthatthePIcontroller withtheseparameterswasnotabletostabilizethesystemandseveresluggingwasnot eliminated.WemaysaythattheSham’stuningmethodisnotasuitableapproachfor thesluggingsystem. 5.1.1.3.2 TuningbasedonIMCdesign Next method applied in tuning of controller in experiments was the IMC‐based tuning described in section 2.9.2. To do this, it was tried to identify the closed‐loop stable system with respect to the data from step test and according to the method proposedbyJahanshahi(JahanshahiandSkogestad2013)explainedinsection2.9.2.1. Theidentifiedmodelofclosed‐loopsystemwasintheformof: Gcl ( s) 11.74 S + 0.606 96.38S 2 10.88S 1 Equation5.1 Theidentifiedclosed‐looptransferfunctionisshownbytheblacklineinfigure5.5. 52 Closed‐loopstepresponse 175 Inletpressure[kPa] 170 165 160 155 set‐point Exp.data Filtered identified 150 145 250 300 350 400 450 500 time(sec) 550 600 650 700 Figure5.5:Presentationofidentifiedclosed‐loopstepresponse.Thedashedblackline showstheidentifiedclosed‐looptransferfunctionobtainedfromIMCdesign. Then, the open‐loop unstable system has been back calculated by using the procedure proposed by Jahanshahi (Jahanshahi and Skogestad 2013). The open‐loop unstablesystemhastheformof: -0.0005538 S - 2.858e-05 P ( s) 2 S 0.008984S 0.004088 Equation5.2 ThentheIMCcontroller(C)isdesignedbyusingthemethodexplainedinsection 2.9.2.2. The time constant of the closed‐loop system is an important manipulated parameterandhasbeenselectedas 20 .Thisnumberwasobtainedbytrialanderror andexperiencingdifferentresults.ThedesignedIMCcontrolleris: C ( s) 287.0673( S 0.02146S 0.0007862) S(S+0.05161) 2 Equation5.3 The IMC controller is a second order transfer function which can be written in form of a PIDF controller. PIDF is a PID controller which a low‐pass filter has been appliedonitsderivativeaction.ItwillbementionedasPIDcontroller. 53 APIcontrollerwasalsoobtainedbyreducingtheorderofIMCcontrollerto1. Therelatedtuningparametershavebeenobtainedandareshownintable5.3. Table5.3:IMC‐basedPIDandPItuningparameter Kc0 Kc I D F PID 220 34.6387 7.92 141.2113 19.3773 PI 220 287.0673 65.6371 _ _ Theapproachofimplementingthelowpassfilterintheexperimentsisdescribed inappendixA. To find the control results all related tuning parameters were implemented in LabVIEW and the loop was run in the stable region with an average valve opening of Z=25%.Thenitwastriedtodecreasetheset‐pointvalueinastepwisemanner.Ateach stepitwaswaiteduntilthesteadystatewasreachedandthenanewstepofreduction was done. Figures 5.6and 5.7 describe the results of control using the IMC‐based PID andPIcontrollersrespectively.Theexperimentalsluggingsystemcouldbestabilizedup to Z= 40% with IMC‐based PID controller and up to Z= 38.4% with IMC‐based PI controllereventhoughthecontrollershavebeendesignedatvalveopeningofZ=28%. 54 InletPressure[Kpa] IMCbasedPIDController 190 Setpoint Measurement 180 170 160 150 140 0 200 400 600 800 1000 1200 X:1363 Y:40.2 50 40 Z[%] 1400 30 20 10 0 0 200 400 600 800 time[sec] 1000 1200 1400 Figure5.6:ResultofcontrolusingtheIMC‐basedPIDcontroller.Thecontrollerhasbeen abletomovethebifurcationpointfromZ=26%uptoZ=40.2%. InletPressure[Kpa] IMCbasedPIController 190 Setpoint Measurement 180 170 160 150 140 0 200 400 600 800 1000 1200 50 Z[%] 40 X:1266 Y:38.39 30 20 10 0 0 200 400 600 800 time[sec] 1000 1200 1400 Figure5.7:ResultofcontrolusingtheIMC‐basedPIcontroller.Thecontrollerhasbeen abletomovethebifurcationpointfromZ=26%uptoZ=38.39%. 55 5.1.1.4 Inconclusiveeffortsandtherelatedpracticalissues Whenworkingwiththefirstvalve,someeffortswereinconclusiveandnoresults wereproduced.Belowsomeexplanationsaregiven. 5.1.1.4.1 TuningthecontrollerbySimpleonlinemethodbasedonidentified MATLABmodelofthesystem AsthelastmethodoftuningitwastriedtousesimplePItuningrulesdescribed in section 2.9.3. The method has been proposed by Jahanshahi (Jahanshahi and Skogestad2013)andisbasedontheidentifiedMATLABstaticmodelofthesystem.To implement this method, first the simple static MATLAB model of the system which tuningrulesarebasedonneededtobemodifiedandfittotheexperimentalsteadystate model. For a reasonable result, it was required to have an accurate model of the experiments. Though, right in that time the lab technician replaced the current valve withthefastvalvesincehewasgoingtovacationandthiscouldn’tbedoneforalong time.Thereforethistuningmethodwastriedonlybythesecondvalve. 5.1.1.4.2 Applyingtimedelayinthecontroller Oneimportantissueregardingtheslowvalveteststhatneedstobementionedis about applying time delay. It was aimed to check the robustness of control system by implementing delay on measurement. In order to do this, an algorithm was implementedinLabVIEWbyoneofthelabtechnicians.Itwasadigitaldelaylinewhich delayed the samples of the measured data by a desired given time. The desired delay timecouldbesetfromthefrontpanel.Yetthedelaysettingcouldn’tworkinadesired way, meaning that even for very small values of delay the system switched to severe sluggingandthecontrolwasimpossible.Itwasclearthatforsuchalongpipelineriser systemsmallvaluesofdelayintherangeofmillisecondscouldn’tcrashthecontroland thereasonofinconveniencymaybefromLabVIEW.Itmightbebecauseofmistakesin the algorithm or in the connections inside LabVIEW. Since the system was inmedium scaleandnooneelseexceptforthelabtechnicianswasabletodomodificationsinthe systemorLabVIEWandalsoduetotimeissuesitwasdecidedtoignoreimplementing timedelayaftercounselingwithmysupervisor.Itwasanextraworktobedoneinthe thesiswhilethenextrequiredexperimentswerenotstartedyetatthattime. 56 5.1.2 Seriesofexperimentswithvalve2(fastchoke valve) Thenextseriesofexperimentalworkinthisthesiswasrepeatingthefirstseries of tests with a new fast valve as the actuator. A new method of tuning has been used hereinadditiontothetuningmethodsofprevioussection. 5.1.2.1 Open‐loopexperiments Theloopwasrunin manualmodewithfixedflowratesof wl 0.3927 [kg / sec ] forwaterand wg 0.0024 [kg / sec ] forair.Theseflowratesarethesamevaluesusedfor the slow valve. The related inflow conditions have been fully described in section 5.1.1.1. The tests were run in different valve openings with fixed liquid and gas flow rates without applying control. The system behavior in natural conditions was then presentedwiththerelatedbifurcationdiagramasseeninfigure5.8. 5.1.2.1.1 Bifurcationdiagram The starting point was the valve opening of Z=0.1. Then the valve was open stepwiseuntilitwasfullyopen.Theresultsofbuffer(inlet)pressurewereloggedand therelatedbifurcationdiagramwasplotted.Thecriticalstabilitypoint(thebifurcation point)isthemaximumchokevalveopeningthesystemcanhavewhilebeingstableand islocatedatZ=0.16forthesystemwithvalve2.Inthepresentedbifurcationdiagram, thetoplinetracksthemaximumvaluesofpressureateachoperatingpoint,thebottom line presents the minimum values of pressure and the middle line shows the average values of the buffer pressure at different valve openings. Small pressure oscillations before the bifurcation point are due to hydrodynamic slugs and are not the signs of instabilities. 57 OpenloopBifurcationDiagram‐Valve2 220 InletPressure[Kpa] 200 180 160 140 120 0.1 0.2 0.3 0.4 0.5 Z 0.6 0.7 0.8 0.9 Figure 5.8:Open‐loopbifurcation diagramfromthefastchokevalveexperiments.The bifurcationpointoccursatvalveopeningofZ=0.16.Thetopandbottomlineillustrate the maximum and minimum values of inlet pressure respectively at each operating point.Themiddlelineshowstheaveragevaluesofpressure. 5.1.2.2 Closed‐loopsteptest Justliketheexperimentserieswithvalve1,thefirststeptotunethecontroller withanytuningmethodwasaclosed‐loopsteptestwithastepchangeinset‐point(the buffer pressure). The loop was closed with a P‐only controller with a gain value of K c 0 250 .Thestepchangewasdoneinaregionthatisunstableinopen‐loopposition. The average of valve opening was Z 0.18 . The related plot is shown in figure 5.10. Since the response was noisy, a low‐pass filter in MATLAB from the type of Simple infiniteimpulseresponsefilterwasusedtoreducethenoiseeffect.Asmoothingfactor of 0.25 wasusedtosmooththesignalaswellasrequired( 1 meansnofiltering). 58 166 164 InletPressure[Kpa] 162 160 158 156 154 152 Setpoint Measurement Filtered 150 148 0 100 200 300 time(sec) 400 500 600 Figure 5.9:Set‐pointstep changefor aclosed‐loopfeedback experiment withaP_only controller using inlet (buffer) pressure as control variable. A low pass filter with a smoothingfactorof 0.25 wasusedtoremovethenoiseeffectfromtheresponse. 5.1.2.3 Tuningthecontroller Three different methods explained in section 2.9 have been used to tune the controller using buffer (inlet) pressure as the control variable and fast choke valve as theactuator.Thetuningprocedureandtherelatedresultswillbepresentedbelow. 5.1.2.3.1 TuningbyShams’sclosed‐loopmethod Shams’s tuning method developed by Shamsuzzoha (Shamsuzzoha and Skogestad 2010) was used as the first tuning method. Table 5.4 shows the resulted tuningparametersbyShams’smethod.Thesystemhasbeenconsideredasafirstorder plus delay model. The information from the closed‐loop step test (See figure 5.9) was usedtofindthetuningparameters. Table5.4:TuningparametersfromSham’smethodforthesluggingsystem Kc0 Z ave Overshoot Offset Kc I 250 0.18 1.5738 1.3823 331.0775 246.7640 59 Asexpected,accordingto resultsof valve1,thePIcontroller withthesetuning parameterscouldn’tstabilizethesystem,meaningthatShams’smethodisnotasuitable methodtotunethesluggingsystemcontroller. 5.1.2.3.2 TuningbasedonIMCdesign IMC‐based tuning method described in section 2.9.2 was applied as the next methodtotunethesystemwithfastvalve.Datafromsteptest(Seefigure5.9)wereused andThemodelofclosed‐loopsystemwasidentifiedasexplainedinsection2.9.2.1. Gcl ( s ) 9.076 S + 0.7406 64.76S 2 4.635S 1 Equation5.4 Theidentifiedclosed‐looptransferfunctionisshownbytheblacklineinfigure5.10. Closed‐loopstepresponse 166 164 InletPressure[Kpa] 162 160 158 156 154 152 set‐point Measurement Filtered identified 150 148 0 100 200 300 time(sec) 400 500 600 Figure5.10:Presentationofidentifiedclosed‐loopstepresponse.Thedashedblackline showstheidentifiedclosed‐looptransferfunctionobtainedfromIMCdesign. Theopen‐loopunstablesystemwasthencalculatedastheformofequation5.5 byusingtheprocedureproposedbyJahanshahi(JahanshahiandSkogestad2013). 60 -0.0005606 S - 4.574e-05 P ( s ) 2 S 0.06858S 0.004006 Equation5.5 TheIMCcontroller(C)wasobtainedthenastheequation5.6(Seesection2.9.2.2). Avalueof 24.5 wasusedforthetimeconstantoftheclosed‐loopsystem.Thisvalue wasmanipulatedbytrialanderroruntilasatisfyinggain,phaseanddelaymarginwas obtainedforthecontroller. C 340.7491( S 0.005194S 0.000356) S ( S 0.0816) 2 Equation5.6 TheIMCcontrollerasasecondordertransferfunctionwasthenwritteninform ofaPIDcontrollerwithalow‐passfilterappliedonitsderivativeaction(Wemaysaya PIDFcontroller). APIcontrollerwasalsoobtainedbyreducingtheorderofIMCcontrollerto1. Therelatedtuningrulesareshownintable5.5. Table5.5:IMC‐basedPIDandPItuningparameters Kc0 Kc I D F PIDF 250 3.4736 2.3368 1189.9378 12.2552 PI 250 340.7491 229.2276 _ _ Thefunction“PIDAdvancedVI”fromLabVIEWwasusedtoimplementthelow‐ passfilterintheexperiments(seeappendixA). ThePIDtuningparameterswereimplementedinLabVIEW.Firstthesystemwas runinopen‐loopmannerwithamanualvalveopeningofZ=0.2anddatawerelogged. Then the loop was closed with a set‐point P=170 kPa that results in an average valve openingofZ=0.16.Aftercoupleofminutesitwastriedtodecreasetheset‐pointvaluein a stepwisemanner. At each step it was waited until the steady state was reached and then a new step of reduction was applied. The same was done with PI tuning parameters.Figures5.11and5.12describetheresultsofcontrolusingthe IMC‐based PID and PI controllers respectively. The experimental slugging system could be stabilizeduptoZ=0.30withIMC‐basedPIDcontrolleranduptoZ=0.29withIMC‐based PIcontroller. 61 InletPressure[Kpa] IMCbasedPIDController 220 Setpoint Measurement 200 180 160 140 120 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Z[%] 100 50 0 0 X:1810 Y:30.19 200 400 600 800 1000 1200 1400 1600 1800 2000 time[sec] Figure 5.11: Result of control using the IMC‐based PID controller. The controller has beenabletomovethebifurcationpointfromZ=16%uptoZ=30.19%(aboutthedouble value). InletPressure[Kpa] IMCbasedPIController Setpoint Measurement 200 180 160 140 0 200 400 600 800 1000 1200 1400 1600 1800 1400 X:1526 Y:29.35 1600 1800 Z[%] 100 50 0 0 200 400 600 800 1000 time[sec] 1200 Figure5.12:ResultofcontrolusingtheIMC‐basedPIcontroller.Thecontrollerhasbeen abletomovethebifurcationpointfromZ=16%uptoZ=29.35%. 62 5.1.2.3.3 SimpleonlinePItuningbasedonMATLABmodelwithgainscheduling SimplePItuningrulesdescribedinsection2.9.3wasusedasthelastmethodof tuningthecontroller.SincethemethodisbasedonthesimplestaticMATLABmodelof the system, the MATLAB model needed to be identified and fit to the experimental steadystatemodel. For a reasonable result, it was first required to have an accurate model of the experiments.TofindthemodeltheloopwasclosedwithaPIcontrollerandwasrunin the region after stability point in open‐loop bifurcation diagram. This was done such thatset‐pointwassettoavaluelowerthanthecorrespondingvalueofthebifurcation point, then it was waited until steady state was reached and data were logged. The average of valve opening was found from logged data and the obtained point was located on the steady state experimental model. By repeating this for some other set‐ pointvaluesthesteadystatelineofexperimentalmodelwasfound.Itisshowninfigure 5.13withtheblackmidline. Next step was to modify the MATLAB static model and fit that with the experimentalmodel.Asdescribedinsection2.9.3.1theMATLABmodelisderivedbased onthevalveequationandisafunctionofvalveopening.Thereforeagoodassumption ofvalveequationisveryimportantinusingthesimplemodel.Thevalveequationisas theform: w K pc f ( z) p Equation5.7 Thevalveislinearanditscharacteristicisdefinedas: f ( z) Z Equation5.8 Thesimplemodelfortheinletpressureisasdefinedinsection2.9.3.1andthe staticgainofthesystembecomesinformof: k ( z) 2w2 . z 3.K pc 2 Equation5.9 Since the tuning parameters are found based on this MATLAB model, a good matchbetweenthismodelandtheexperimentalmodelisveryimportantmeaningthat thevaluesofinletpressureandthestaticgainobtainedbythemodelneededtobetrue values. The parameters Lr (length of riser), P _ Vmin (minimum Pressure drop over the valve) and K pc (the valve constant) were manipulated many times until the desired match with the experimental model was reached. Below is a discussion of these parameters. 63 Lengthofriser InMATLABmodellengthofriserisdirectlyusedtocalculatethestaticpressure oftheriserwhenitisfilledwithliquidandthereafterthisstaticpressureisusedtofind theinletpressureatanylevelofvalveopening.Thereforemanipulatingofthatcouldbe veryhelpfulinproducingdesiredresults.Theexactlengthofriserintheexperimental setuphasbeen6.433m.Though,itwaschangedto6.7minmodeltoprovidethebest results. Minimumpressuredropoverthevalve Thisparameterisusedinseveralcalculationsinthemodel.Themostimportant oneisthevalueofinletpressureinthefullyopenpositionofthevalvethatuses P _ Vmin directly (See section 2.9.3.1). Level of the curve in the inlet pressure plot of the model was quite affected by inlet pressure at fully open position of the valve. A value of P _ Vmin 3kPa wasusedtogetthebestfitnessofthemodels. Valveconstant The valve constant K pc has a major effect on the slope of the curve in the inlet pressureplotofthemodel.Avalueof K pc 1.6 103 wasusedintheMATLABmodel. Figure 5.13 compares simple static MATLAB model to the experimental model. As clear in the figure there is a good match between the two models. The MATLAB modelisattachedinAppendixC.5.Theblackmidlineinthefigurepresentsthesteady statevaluesoftheinletpressureandtheredmidlineisthevaluesofinletpressurefrom theMATLABmodel.Thetopandbottomblacklinesshowthemaximumandminimum valuesofpressureoscillationsateachoperatingpointintheopen‐loopsystem. 64 220 SimplestaticMATLABmodel Experimental InletPressure[kPa] 200 180 160 140 120 0.1 0.2 0.3 0.4 0.5 Z 0.6 0.7 0.8 0.9 Figure 5.13: Simple static MATLAB model compared to the experimental model. The blackmidlineinthefigurepresentsthesteadystatevaluesoftheinletpressureandthe redmidlineisthevaluesofinletpressurefromtheMATLABmodel.Thetopandbottom black lines show the maximum and minimum values of pressure oscillations at each operatingpointintheopen‐loopsystem. In order to find the appropriate tuning parameters based on the identified MATLABmodelaclosed‐looptestwithstepchangeofset‐pointisrequired(Seesection 2.9.3.2).Datafromthesteptestexplainedinsection5.1.2.2wasused(Seefigure5.9)and theparameter wasfoundfromtheequation2.41as =0.038.Theperiodofslugging oscillationsinopen‐loopexperimentshasbeen Tosc =90Sec.Whenrunningthemodelat aspecialoperatingpointtheparameters K c ( z ) and I ( z ) werefoundforthespecified operatingpointasfunctionsofvalveopening(Z)bytheequations2.42and2.43. ByrunningthemodelwithaloopforawiderangeofZvalues,itwaspossibleto find multiple tuning parameters each for a controller at a specified operating point. Thengain‐schedulingwithmultiplecontrollerswasusedtostabilizethesystem.Todo this in the experiments, five PI controllers were used with the related found tuning parameters. Table 5.6 shows the resulted controller for each operating point of valve opening. 65 In order to perform the gain scheduling between the controllers the correspondingvalueofinletpressuretothatspecificoperatingpointofvalveopening wasdeterminedfromthemodelandthenthispressurevalueastheset‐pointtogether withtherelatedtuningvalueswereenteredinLabVIEW.Theclosed‐loopwasrunandit was waited until the system was in steady state. Then the next pressure value (set‐ point) corresponding to the next valve opening was tried and the new tuning values wereenteredinLabVIEWveryfast(Iwasworkingasthecontrolwoman!).Thisaction wasrepeateduntiltheinstabilitywasappeared. Figure 5.14 shows the results of control with gain scheduling tuned by simple onlinetuningmethod.Thecontrollerscouldstabilizetheflowuptoavalveopeningof Z=0.35. Changing bifurcation point from Z=16 % into Z=35 % could be a very good result. 66 Table5.6:PItuningvaluesandthecorrespondingoperatingpointsfromsimpleonline tuning method based on MATLAB model. These five controllers were connected and performedgainschedulingwithmultiplecontrollersforthenonlinearsluggingsystem. Kc I Set‐point Valveopening 360.6481 320.625 166 0.19 511.9335 354.375 165 0.21 816.9311 405 161 0.24 2000.8383 523.125 157 0.31 4080.2676 641.25 156 0.38 InletPressure[Kpa] Multiplecontrollerswithgainscheduling Setpoint Measurement 200 180 160 140 0 200 400 600 800 1000 1200 1400 1600 1800 Z[%] 100 X:1604 Y:34.63 50 0 0 200 400 600 800 1000 time[sec] 1200 1400 1600 1800 Figure5.14:ResultsofgainschedulingusingfourPIcontrollers.Whenthesystemwas switched into the fifth controller the instability was appeared; meaning that the maximum level of stability was reached with four controllers tuned by simple online tuningrules.ThecontrollershavebeenabletomovethebifurcationpointfromZ=16% uptoZ=35%. 67 5.1.3 CasscadeControlu usingtoppressu urecom mbinedw with density y Oneoftheetasksinth hethesish hasbeenru unningthe closed‐loo opusinga cascade controll structuree with seleecting top pressure aand density of outflo ow as the control variables.Theinttentionsweeretuning thiscontro olstructureebytrialanderrorandthen analyzeethecontro ollabilitych haracteristticsincomp parisonwitththesinglleloopstru ucture. 5.1.3.1 1 Theetestpra acticalisssue–testtincompllete The outflow density was used as the control variab ble of innerr loop and the top pressurrewasseleectedastheecontrolvaariablefor theouterlloop.Havin ngarightm measure of denssity could be very im mportant in i controlling sluggin ng system with the cascade structu ure.Figure5.15illustrratesascheematicoverrviewofcaascadestructureinLaabVIEW. The deevice used d for meassuring thee outflow density was w the co onductancee probe explain nedinsection3.2.11. Figure 5.15:Asch hematicvieewofcascaadecontrollstructure inLabVIEW W.Theoutterloop receivees signal frrom top prressure sen nsor as thee measurem ment and produces the t set‐ point ssignal for the inner loop. Thee inner loop uses th he density y signal from the conducctanceprob beinstalled dintheoutlletoftheriiser. 68 Howeveritseemedthat theconductanceprobeisnota goodmeasuringdevice forthedensity.Aftersomeunsuccessfultriestocontrolthesystembytuningtheloops withtrialanderror,itwasdecidedtoperformasteptestinopen‐loopsituationofthe system and evaluate the open‐loop step response of the conductance probe. It was aimed to check the applicability of probe as an appropriate sensor to measure the density.Todothis,theloopwasruninmanualmodeatthestableregionwithavalve opening of Z=7%. Data from density meter (Conductance probe) and top pressure sensor was logged. After some minutes the valve was changed to Z=12% while it was tried to keep the system in the stable region. Figure 5.16 presents the open‐loop step responseoftheprobeasthedensitymeter. Openloopresponse Z 0.1 0.08 Pressure 0.06 0 200 100 200 300 400 500 600 100 200 300 400 500 600 100 200 300 400 time(sec) 500 600 100 Density 0 0 20 15 0 Figure 5.16: Open‐loop step response of conductance probe (density meter) in the stableregion.Thefirstplot(Z)showsthestepchangeinvalveopening,thesecondplot presentsthestepresponseof toppressure inKiloPascalandthethirdplotillustrates thestepresponseoftheoutflowdensityinkg/sec. As seen in the figure, the density signal does not show a clear response to the step change and is very noisy. This signal couldn’t be a suitable measurement for the control targets. In order to have anefficient cascade control withdensityas the inner loopcontrolvariable,moreaccuratesignalsofdensityarerequired. Trying each loop separately toevaluate their response independently, could be considered as an alternative work. But this was not practical since the backup of the LabVIEW file was lost and the compiled file couldn’t be manipulated or modified. Makinganewfilewasnotpossibleduetotimeissues. 69 5.2 ComparisonofSlowvalveandFastvalve In this section it has been tried to compare the dynamics of the applied control valves(slowvalveandfastvalve)byinvestigatingtheirrelatedresults.Forthistarget theopen‐loopandclosed‐loopresultsofthetwovalveswerecompared.Ourcriterionto evaluate control loop is the stability. For a fast stability the dynamic response of the valveisimportant.Itmeansasmalldeadtimeforthevalve. The criterion for evaluating the stability of the slugging control loop has been usually the level of valve opening (Z). However, this criterion couldn’t be useful for comparing control valves with each other. The reason gets back to the valve inherent characteristicsthatwillbeexplainedinthefollowing. Therelationbetweentheflowrateandthelevelofvalveopeningisaninherent characteristicofthevalvethathasbeendefinedasthevalveequation: qmix K pc f ( z ) P mix Equation5.10 Here qmix isthevolumetricflowrate, K pc isthevalveconstant, P isthepressure dropoverthevalveand mix isthemixeddensityofoutflow. Bothvalveshavebeenconsideredlinearwith f ( z ) z .Butinrealityvalve2(fast valve)couldbenonlineartosomeextent,meaningthatitproducesthesameflowrate as the slow valve even with lower levels of valve openings. Figure 5.17 describes this concept more clearly by illustrating the characteristic curves for the two valves. If we specifyalevelofflowrateandtrytofindthecorrespondinglevelsofvalveopeningfor each of the valves, it will be seen that the slow valve may give higher level of valve openingforthesameflowrate. The main desired result that can be affected by the valve dynamics is the minimum inlet pressure the system could obtain. For the open‐loop system, this is defined as the minimum inlet pressure at fully open position of the valve and for the closed‐loop system it will be defined as the minimum set‐point the controller can stabilizethesystem.Figure5.18comparestheopen‐loopbehaviorofthesystemforthe slowvalvewiththatoffastvalve.Asclearinthefigure,theslowvalvegiveslowerinlet pressuresatmostoperatingpointsofvalveopeningincludingthefullyopenpositionof thevalve(Z=1). 70 Figure 5.17: Characteristic curves for slow (linear) and fast (quick opening) valves. There is a lower level of valve opening for the fast valve at a specific flow rate (for instance 50), meaning that the fast valve can produce the same flow rate as the slow valveevenatlowerlevelsofvalveopening. OpenloopBifurcationDiagrams 220 Slowvalve Fastvalve InletPressure[Kpa] 200 180 160 140 120 100 0.2 0.4 0.6 Valveopening(Z) 0.8 1 Figure5.18:Comparisonofinletpressurebetweentheslowvalveandthefastvalveat their different operating points for the open‐loop system. At a certain level of valve opening,theslowvalvegivesalowerinletpressure. 71 Based on the previous descriptions, it was decided to compare the minimum achievable set‐points in the closed‐loop responses. Figure 5.19 present the control results with IMC‐based PI and IMC‐based PID controllers. The valve opening is also presented,justincase,andisnotapointofinteresttocomparetheresults. Fromthefiguresitcanbesaidthattheslowvalvehashadabetterperformance comparedtothefastvalve.Thismeansthattheslowvalvehasbeenalreadyfastenough forourcontroltargetsandtherehasbeennoneedtovalve2(fastercontrolvalve).In other words the stability of the slugging system is more affected by the tuning parametersforthecontrollerinsteadofcontrolvalvedynamics. 72 IMCbasedPIController InletPressure[Kpa] 190 SetpointfromSLOWvalve MeasurementfromSLOWvalve SetpointfromFASTvalve MeasurementfromFASTvalve 180 170 160 150 140 0 50 Z[%] 40 200 400 600 800 1000 1200 1400 ValveopeningfromSLOWvalve ValveopeningfromFASTvalve 30 20 10 0 0 200 400 600 800 time[sec] 1200 1400 IMCbasedPIDController 190 InletPressure[Kpa] 1000 SetpointfromSLOWvalve MeasurementfromSLOWvalve SetpointfromFASTvalve MeasurementfromFASTvalve 180 170 160 150 140 0 500 1000 50 Z[%] 40 1500 ValveopeningfromSLOWvalve ValveopeningfromFASTvalve 30 20 10 0 0 500 time[sec] 1000 1500 Figure5.19:ComparisonofIMC‐basedcontrolresultsfromslowvalvewiththoseoffast valve.Withtheslowvalveithasbeenabletodecreaseset‐pointinawiderrangetoa lowerlevel.Withthefastvalve,theopen‐loopsystemswitchedtosluggingatP=170 kpa while with the slow valve the instability started at P=180 kpa in the open‐loop system. These are the initial points, respectively, to start control. The minimum achievableset‐pointhasbeenP=154kpafortheslowvalveandP=158kpaforthefast valve.Thismeansthattheslowvalvehasshownabetterperformancefortheslugging systemandhasbeenalreadyfastenoughaswell. 73 5.3 SimulatedresultsfromOLGAmodel InthischaptersimulationoftheexperimentalcasesinOLGA®arepresented.The simulationshavebeenmatchedwiththeexperimentalmodelsfromvalve1(Slowvalve). The open‐loop simulations are discussed in section 5.3.1. In section 5.3.2 results of control simulations by trial and error would be explained. Section 5.3.3 deals with finding the appropriate tuning rules based on the methods explained in section 2.9. Results of control by applying the calculated tuning parameters are also discussed in thissection. 5.3.1 Open‐loopsimulations The first step before implementing the controller is running simulations for different valve openings with fixed liquid and gas flow rates. The values of Z and the relatedflowregimetypesarepresentedbytable5.7. Table5.7:Differentvaluesofvalveopening(Z)usedinopen‐loopsimulations Z 0.20 0.25 0.26 0.27 0.28 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Flowregimestability Stable Stable Stable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Figure 5.20 describes the open‐loop bifurcation diagram from simulations. The diagram shows the maximum, minimum, average and steady state values of buffer pressureversusthevalveopenings.Theappliedfixedflowrateshavebeen wl 0.3927 [ kg / sec ] for water and wg 0.0024 [ kg / sec ] for air (The same as experiments). These flow rates correspond to U sl 0.2 [m / sec] and Usg 1 [m / sec] as the liquid and gas superficialvelocities.Thecriticalstabilitypoint(thebifurcationpoint)isthemaximum chokevalveopeningthesystemcanhavewhilebeingstable.In abifurcationdiagram, the critical stability point is where the maximum and minimum pressures approach a finite value. In the presented bifurcation diagram, the red line shows the steady state 74 valuesofthebufferpressureatdifferentvalveopeningsandtheaveragevaluesofthe pressure are on the mid black line that is higher than the steady state line. The coefficientofdischargewaschangedto cd=0.34inordertomanipulatetheplacementof the critical valve opening (the bifurcation point) based on the experimental result of valve 1 and also fit the steady state OLGA values with the steady state values from experimentsandmodels.ThediagramcomparingsteadystatevaluesfromOLGAwith that of the model will be presented in section 5.3.3. As clear in the figure the critical stabilitypointwasfoundtobeatapproximatelychokevalveopeningofZ=26%. OpenloopBifurcationDiagram‐OLGAmodel 220 Steadystate InletPressure[Kpa] 200 180 160 140 120 100 0.2 0.3 0.4 0.5 0.6 Z 0.7 0.8 0.9 1 Figure 5.20: Open‐loop bifurcation diagram from OLGA simulations. The bifurcation point occurs at valve opening of Z=0.26. The top and bottom line illustrate the maximum and minimum values of oscillations for inlet pressure respectively at each operatingpoint.Themid‐blacklineistheshowstheaveragevaluesofpressure. 5.3.2 Controlbytrialanderror As the first work after implementing PID controller in OLGA, it was tried to stabilizetheflowbytrialanderror.TwotypesofcontrollerincludingP‐onlycontroller andPIcontrollerweretriedtobetunedbytryingmanydifferentvaluesastherelated tuningparameters. 75 5.3.2.1 P‐onlycontroller As the first try a P‐only controller was used to stabilize the system. P‐only controller has been designed by inserting D 0 and I 1010 .A point in unstable region with Z 0.3 was selected and different values of gain parameter were tried to check which gain can create stability with the highest level of valve opening (Z). For each gain value it was tried to find the minimum amount of buffer (inlet) pressure as Set‐point or in other words the maximum level of valve opening as manipulated variablebystepwisereductionoftheSet‐point.Table5.8showsdifferentvaluesofgain thathavebeentriedandthecorrespondingminimumvalueofSet‐pointandmaximum valueofZ. Table5.8:differenttriedP‐onlycontrollers initialvalve opening Kc Minimum Set‐pointvalue (P) Maximum Manipulatedvariable (Z) 0.3 0.01* 0.05 0.1 0.5 1 2 5 10 _ 141 138 135.6 136 _ _ _ _ 0.3686 0.4497 0.6454 0.6305 _ _ _ ThebestcontrollerthatgivesstabilitywiththehighestlevelofZandthelowest level of achievable set‐point is the one with K c 0.5 . With this controller, the bifurcationpointwasmovedfromZ=26%intoZ=65%.Figure5.21showstheresultof controlbyP_Onlycontrollerfor K c 0.5 .Forthecontrollerwith Kc 0.01 ,specifiedby thestarinthetable,thesimulatorcouldconvergeinsomevaluesofSet‐point.However, theresultwasnotgoodandthereweremanyoscillationsinpressureandvalveopening. ItwasalmostimpossibletomakeareductionintheSet‐point. For the controller with K c 1 , control was difficult and the Set‐point reduction was challenging. Figure 5.22 shows the result of control by for K c 1 . The steps of reduction had to be selected very small and the simulator could not converge with a largerstepthanitisobservedinthefigure.Forthevaluesfilledwithdashthesimulator couldnotconvergeforanyvaluesofSet‐point,meaningthatitwasimpossibletocontrol thesystemwiththegainvalueshigherthan1.AP‐onlycontrollerwith 0.05 K c 1 can stabilizethesystem. 76 InletPressure(Controlledvariable) Set‐point Measurement P[Kpa] 150 140 130 0 500 1500 2000 Valveposition(Manipulatedvariable) 1 Z 1000 X:2179 Y:0.6436 0.5 0 0 500 1000 time[sec] 1500 2000 Figure 5.21: Simulation result of control by P‐Only controller for K c 0.5 with OLGA. Thishasbeenthebestresultfromtrialanderrorduetothelowestachievableset‐point orinotherwordsthehighestlevelofvalveopening. InletPressure(Controlledvariable) 150 P[Kpa] Set‐point Measurement 140 130 0 500 1500 2500 X:2370 Y:0.6306 0.5 0 0 2000 Valveposition(Manipulatedvariable) 1 Z 1000 500 1000 1500 time[sec] 2000 2500 Figure5.22:SimulationresultofcontrolbyP‐Onlycontrollerfor K c 1 77 5.3.2.2 PIcontroller PIcontrollerwasusedtostabilizethesysteminthesecondseriesofsimulations bytrialanderror.Thecontrollerwasdesignedbyinserting D 0 andtryingdifferent valuesfor Kc and I .Theaimasthepreviouspartwastotunethecontrollertocreate stable flow with the highest production rate (the highest level of valve opening (Z)). StepwisereductionoftheSet‐pointwasimplementedastheone forP‐onlycontroller. Table 5.9 shows different values of tuning parameters that have been tried and the correspondingminimumvalueofSet‐pointandmaximumvalueof Z . Table5.9:SimulationResultsofdifferenttriedtuningparametersforPIcontroller K c I 80 0.01 0.05 0.1 Min. Max. Min. Max. P Z P Z _ _ 143.2 0.368 0.5 1 Min. Max. Min. Max. Min. Max. P Z P Z P Z 140 0.416 136.5 0.621 _ _ 130 _ _ 141.9 0.379 140 0.434 136.5 0.627 _ _ 180 _ _ 141.7 0.388 139.6 0.457 136.2 0.639 _ _ 300 _ _ 141.5 0.395 139.6 0.460 136.2 0.644 _ _ 800 _ _ 141.3 0.397 139.4 0.463 136.2 0.646 _ _ As it is observed in the table, the best tuning parameters are K c 0.5 and I 800 .Highervaluesof800werealsotriedfortheparameter I andnodifference wasmadeinresult.ResultofcontrolbyPIcontrollerwiththebesttuningparametersis presentedinfigure5.23.Increasingtheparameter I decreasedthesystemoscillations verywellandeveneliminateditinsomecases.However,itcausedalongertimetobe requiredfortheoutputtotracktheSet‐pointineachstepofSet‐pointreduction.This important effect of applying integral time constant could be verified by comparing figures5.23and5.22.Asitisobserved,alessoscillatorysystemwithlongersimulation timeistheresultofPIcontrollercomparedwithP‐onlycontroller. The same as the one for P‐only controller happened for the PI controllers with thegainvaluesof Kc 0.01 and K c 1 .Thesimulatorcouldnotconvergeforanyvalues of Set‐point, meaning that it was impossible to control the system with the tuning parametersfilledwithdash. 78 InletPressure(Controlledvariable) P[Kpa] 180 Set‐point Measurement 160 140 120 0 500 1500 2000 3000 X:2951 Y:0.6415 0.5 0 0 2500 Valveposition(Manipulatedvariable) 1 Z 1000 500 1000 1500 time[sec] 2000 2500 3000 Figure5.23:simulationresultsofcontrolbyPIcontrollerfor K c 0.5 and I 800 5.3.3 Tuningthecontroller Threetuningmethodsexplainedinsection2.9wereusedinsimulations.Theaim hasbeentocomparethetuningmethodsbasedonsimulationsaswellasexperiments. 5.3.3.1 TuningusingShams’sclosed‐loopmethod InordertotunethecontrollerwithShams’smethodaclosed‐loopsteptestwas required. As explained in section 2.9.1 a P‐only control is required to determine the optimal tuning parameters. The P‐only controllers described in section 5.3.2.1 were usedtoachieveastepresponseclosetotherecommend0.3overshoot.Firstthesystem was set with the choke opening at Z=30 % where it is unstable in open‐loop position andstableinclosed‐loopposition.Itwasatsteady‐stateinitiallyandthenastepchange was applied in set‐point. Different values of step change were tried to get a step response close to the recommend 0.3 overshoot. Then the system was set with the chokeopeningatZ=40%andthesametrieswereimplemented.Valueoftheresulting overshoot was highly depended on the initial gain and the amount of step change. In some cases with the same initial gain several tests with different amounts of step changewereruninordertogetthedesired0.3overshoot.Allsimulationsruntogetthe 79 desired overshoot at different basis conditions of the controller are presented in AppendixB. When the desired overshoot was achieved the Shams’s method for closed‐loop systemsexplainedinsection2.9.1wasusedtofindtheappropriatetuningparameters. Table5.10showstheresultingtuningparametersbyShams’smethodatdifferentinitial positions of choke valve. Kc0 is the initial gain used in the tuning simulation, Kc is the calculatedproportionalgain,and I istheintegraltuningparameter. Table5.10:TuningparametersfromSIMCmethodforthesluggingsystem Initialvalve position Kc0 Overshoot Offset Kc I 0.3 0.1 0.3085 0.0787 0.0614 34.5702 0.4 0.15 0.3210 2.1132 0.0904 3.1150 UsingPIcontrollerswiththeparametersfoundintable5.10,thesystembecame unstableatachokevalveopeningofapproximateZ=38.84%withthecontrollertuned attheinitialpositionof30%andatachokevalveopeningofapproximateZ=39.45% withthecontrollertunedattheinitialpositionof40%. Figures5.24and5.25showthesteptestusinginitialchokevalveopeningof30% asthebasisinflowconditionandtheresultofcontrolbyShams’smethodrespectively. Figures5.26and5.27arepresentedfortheinitialchokevalveopeningof40%. Twoinitialpointswereusedfortuningtoimprovetheresults.Howeverasitis clear from the figures, no notable change is observed in the results of control. Decreasingset‐pointevenforaverysmallvaluemorethanthefinalvalueshowninthe figurescausedsystemtobecomeunstable.Severesluggingoccurredandthesimulator couldnotconverge. Asseenintheresults,thesecondcontrollertunedattheinitialpointofZ=40% hasn’t been able to stabilize the system for any further valve openings. It hasn’t been ableeventoachievethepointthathasbeentunedfor.Thismaynotbestrangesincethe Shams’smethodhasbeendesignedforthestablesystemswhilethesluggingsystemis unstable. 80 Inletpressure(controlledvariable) P[Kpa] 151 150 149 0 Setpoint Measurement 100 200 300 400 500 600 500 600 Valveposition(manipulatedvariable) Z 0.3 0.25 0.2 0.15 0 100 200 300 time(sec) 400 Figure 5.24: Set‐point step change using initial choke valve opening of 30% and the initialgainof K c 0 0.1 .AnovershootofD=0.3,astherecommendedvaluebyShamshas beenachieved. Inletpressure(controlledvariable) Setpoint Measurement P[Kpa] 150 145 140 0 500 1000 1500 2000 2500 3000 3500 Valveposition(manipulatedvariable) 0.5 0.4 Z X:3452 Y:0.3884 0.3 0.2 0 500 1000 1500 2000 time(sec) 2500 3000 3500 Figure 5.25: Simulationresult ofcontrolby Shams’sclosed‐loopmethodfortheinitial choke valve position of 30%. The values of Z=0.389 and P=142 kpa have been the maximumachievedvalveopeningandtheminimumachievedset‐point,respectively. 81 Inletpressure(controlledvariable) P[Kpa] 146 Setpoint Measurement 144 142 0 100 200 300 400 500 600 500 600 Valveposition(manipulatedvariable) Z 0.6 0.4 0.2 0 0 100 200 300 time(sec) 400 Figure5.26:Set‐pointchangeusinginitialchokevalveopeningof40%andtheinitial gainof K c 0 0.15 .AnovershootofD=0.32,neartotherecommendedvaluebyShams hasbeenachieved. P[Kpa] 148 Inletpressure(controlledvariable) Setpoint Measuremet 146 144 142 0 500 1000 1500 2000 Valveposition(manipulatedvariable) Z 0.6 0.4 X:1952 Y:0.3945 0.2 0 0 500 1000 time(sec) 1500 2000 Figure5.27:SimulationresultofcontrolbyShams’smethodfortheinitialchokevalve position of 40%. The values of Z=0.395 and P=142 kpa have been the maximum achievedvalveopeningandtheminimumachievedset‐point,respectively. 82 5.3.3.2 TuningbasedonIMCdesign Next method used in tuning of controller in simulations was the IMC‐based tuningdescribedinsection2.9.2.Asexplainedbefore,theopen‐loopsystemswitchesto sluggingflowatvalveopeningofZ=26%anditisunstableatZ=30%or40%.Tuningby this method was done for two different operating points of the system; Z=30% and Z=40%.Bothsimulationsaswellastheirresultsarepresentedinthissection. 5.3.3.2.1 IMC‐basedtuningatZ=30%astheinitialvalveposition TheloopwasclosedbyaP‐onlycontrollerwithaninitialgain K c 0 0.1 andset‐ pointwaschangedby2kPa,atZ=30%.Thenwithrespecttothedatafromsteptestand according to the method proposed by Jahanshahi (Jahanshahi and Skogestad 2013) explainedinsection2.9.2.1,closed‐loopstablesystemwasidentifiedasthefollowing: Gcl ( s) 8.105 S + 0.919 17.73S 2 3.765S 1 Equation5.11 Figure 5.28 illustrates the implemented step change and the identified closed‐ looptransferfunctionshownbytheblackline. Then, the open‐loop unstable system has been back calculated by using the procedureproposedbyJahanshahi.Theopen‐loopunstablesystemhastheformof: P ( s) -4.572 s - 0.5184 2 s 0.2448s 0.00457 Equation5.12 Then the IMC controller (C) is then designed by using the method explained in section2.9.2.2.Thetimeconstantoftheclosed‐loopsystemhasbeenselectedas 10 . This number was obtained by trial and error and experiencing different results. The designedIMCcontrolleris: C ( s) 0.11916( S 2 0.04668S 0.001835) S (S+0.1134) 83 Equation5.13 Closed‐loopstepresponsefromOLGAsimulations 151.5 Inletpressure[kPa] 151 150.5 150 149.5 149 148.5 Setpoint OLGAmeasurement Identifiedmodel 148 147.5 180 200 220 240 time(sec) 260 280 300 Figure 5.28: Closed‐loop response of step change at initial valve opening Z=0.3. The dashedblacklineshowsthetransferfunctionoftheIMC‐basedidentifiedmodel. TheIMCcontrollerisasecondordertransferfunctionandcanbewritteninform ofaPIDFcontroller.PIDFisaPIDcontrollerwhichalow‐passfilterhasbeenappliedon itsderivativeaction.ItwillbementionedasPIDcontroller. APIcontrollerhasbeenalsoobtainedbyreducingtheorderofIMCcontrollerto one. The related PID and PI tuning parameters have been calculated as described in section2.9.2.3andareshownintable5.11. Table 5.11: IMC‐based PID and PI tuning parameters tuned at the initial choke valve positionof30% Kc0 Kc I D F PIDF 0.1 0.03204 16.6113 23.9802 8.8191 PI 0.1 0.11916 61.7797 _ _ Implementing low pass filter was not possible in OLGA. Therefore, despite the fact that the filter time constant was an important part of tuning parameters, it was neglectedinsimulationsandaPIDcontrollerwasusinginstead. Figures5.29and5.30describetheresultsofcontrolusingtheIMC‐basedPIDand PIcontrollersrespectively.ThecontrollersweretunedforvalveopeningofZ=30%.But, theycanstabilizethesystemuptoverylargervalveopenings.ThePIDcontrollercould stabilizetheflowwithamaximumof50.27%valveopeningandthePIcontrollercould stabilize the system up to valve opening of Z=47%. The PID controller has shown a better performance compared to the PI. A lower set‐point as well as a higher level of valve opening has been achievedwith PID controller. Inaddition, the outputfrom the PIDcontrollerislessoscillatory. 84 Inletpressure(controlledvariable) P[Kpa] 150 Setpoint Measurement 145 140 0 500 1500 2000 0.5 0 0 2500 Valveposition(manipulatedvariable) 1 Z 1000 X:2440 Y:0.5027 500 1000 1500 time(sec) 2000 2500 Figure5.29:SimulationresultofcontrolusingtheIMC‐basedPIDcontrollertunedatthe initialchokevalvepositionof30%. Inletpressure(controlledvariable) P[Kpa] 150 Setpoint Measurement 145 140 0 500 1000 1500 2000 2500 Valveposition(manipulatedvariable) Z 0.6 X:2417 Y:0.4707 0.4 0.2 0 500 1000 1500 time(sec) 2000 2500 Figure5.30:SimulationresultofcontrolusingtheIMC‐basedPIcontrollertunedatthe initialchokevalvepositionof30%. 85 5.3.3.2.2 IMC‐basedtuningatZ=40%astheinitialvalveposition ThesamesimulationastheoneexplainedinprevioussectionwasrunatZ=40%. TheloophasbeenclosedbyaP‐onlycontrollerwithaninitialgain K c 0 0.15 andset‐ pointhasbeenchangedby 1kPa,atZ=40%.Thesameprocedureandcalculationsas describedinprevioussectionwasusedtofindIMC‐basedPIDandPItuningparameters. Theclosed‐loopstablesystemwasidentifiedasthefollowing: Gcl ( s) Equation5.14 7.011 S + 0.805 23.64S 2 2.27S 1 The implemented step change and the identified closed‐loop transfer function areillustratedinfigure5.31. Closedloopstepresponse InletPressure[kPa] 143 142.5 142 141.5 141 250 Setpoint OLGAMeasurement Identifiedmodel 300 350 400 450 500 time(sec) 550 600 650 700 Figure 5.31: Closed‐loop response of step change at initial valve opening Z=0.4. The dashedblacklineshowsthetransferfunctionoftheIMC‐basedidentifiedmodel. Theopen‐loopunstablesystemhastheformof: P ( s) -2.966 S - 0.3405 S 0.2006S 0.00825 2 86 Equation5.15 ThedesignedIMCcontrolleris: C ( s) Equation5.16 0.16877( S 2 0.04345S 0.001998) S (S+0.1148) And finally the related PID and PI tuning parameters have been calculated as shownintable5.12. Table 5.12: IMC‐based PID and PI tuning parameters tuned at the initial choke valve positionof40% Kc0 Kc I D F PIDF 0.15 0.038293 13.0406 29.6774 8.7097 PI 0.15 0.16877 57.4753 _ _ Just like the previous part, the filter time constant was neglected due to impossibilityofapplyinglow‐passfilterinOLGAandaPIDcontrollerwasusedinstead. Figures5.32and5.33describetheresultsofcontrolusingtheIMC‐basedPIDand PI controllers respectively, tuned for Z=40%. The controllers were tuned for valve openingofZ=40%.ThePIDcontrollercouldstabilizethesystemuptoZ=54.61%.Infact withthiscontroller,thebifurcationpointhasbeenmovedfromZ=26%intoZ=54.61%. ThePIcontrollercouldstabilizethesystemuptoZ=51%. As well as the result for the initial point of Z=30%, the PID controller shows a betterperformancewithlessoscillationsinoutputandahigherlevelofvalveopening hasbeenachieved. 87 Inletpressure(controlledvariable) 146 Setpoint Measurement P[Kpa] 144 142 140 138 136 0 200 600 800 1000 1200 1400 0.5 0 0 1600 1800 Valveposition(manipulatedvariable) 1 Z 400 X:1742 Y:0.5461 200 400 600 800 1000 time(sec) 1200 1400 1600 1800 Figure5.32:SimulationresultofcontrolusingtheIMC‐basedPIDcontrollertunedatthe initialchokevalvepositionofZ=40%. 146 Inletpressure(controlledvariable) Setpoint Measurement P[Kpa] 144 142 140 138 136 0 500 1000 1500 2000 Valveposition(manipulatedvariable) Z 0.6 X:1929 Y:0.5101 0.4 0.2 0 0 500 1000 time(sec) 1500 2000 Figure5.33:SimulationresultofcontrolusingtheIMC‐basedPIcontrollertunedatthe initialchokevalvepositionofZ=40%. 88 5.3.3.3 Tuningusingsimpleonlinemethodwithgainscheduling Simple PI tuning rules based on identified MATLAB static model of nonlinear partofthesystemwasusedasthelasttuningmethodinthesimulations.Themethod has been proposed by Jahanshahi (Jahanshahi and Skogestad 2013) and described in section2.9.3. 5.3.3.3.1 ModifyingMATLABmodel AsthefirststepinimplementingthismethodthesimplestaticMATLABmodelof thesystemwhichtuningrulesarebasedon,neededtobemodifiedtobesimilartothe OLGAcaseusedinthesimulationsofthethesis.Asexplainedbefore,theOLGAcasewas thepipeline‐S‐shapedrisersetuplocatedatmultiphaselaboratoryofNTNU. Asdescribedinsection2.9.3,thesimplemodelisbasedonthevalveequation: w Kpc f ( z) p Equation5.17 ForthevalveusedinOLGAsimulationsthevalvecharacteristicisdefinedas: f ( z) z.cd 1 z 2 .cd 2 Equation5.18 Here cd is the discharge coefficient of the valve and had an important role in fittingtheMATLABmodeltothesimulations. K pc wasconsideredas: Kpc 2 A Equation5.19 Aisthecrosssectionalareaofthepipeandfinallythemodelwasfoundasthefollowing forthesimulations: k ( z) 2w2 . z 3.cd 2 .K pc 2 Equation5.20 Themodelisafunctionofvalveopeningandthereforthevalueofinletpressure andthestaticgainachievedataspecifiedoperatingpoint(valveopening)wasdifferent fromtheoneinanotheroperatingpoint.Sincethetuningparametersarefoundbased onthismodel,itisveryimportantthatthemodeltoberealistic,meaningthatthevalues ofinletpressureandthestaticgainobtainedbythemodelneededtobetruevalues.In ordertomakeagoodmatchbetweenthemodelandtheOLGAcasethegeometrywas changed tosuit the experimentalsetup. However it soon became clear thatthe model needed to be manipulated to achieve the desired results. As the manipulated 89 parameters, length of riser and the discharge coefficient of the valve were quite effective.Adescriptionregardingthisissuewillbepresentedbelow. Lengthofriserasthefirstmanipulatedparameter InMATLABmodellengthofriserisdirectlyusedtocalculatethestaticpressure oftheriserwhenitisfilledwithliquidandthereafterthestaticpressureoftheriseris usedtofindtheinletpressureatanylevelofvalveopening.Thereforemanipulatingthat could be very helpful in producing desired results. The exact length of riser that was usedinsimulationsis7.7054m.Though,itwaschangedto5.15minmodeltoprovide thebestresults. Dischargecoefficientofchokevalve(cd)asthesecondmanipulatedparameter Thecoefficientofdischargeinthevalveequationisaconstantwhichdependson thepressuredropoverthevalve.InordertofitthesimplestaticMATLABmodeltothe OLGA case this parameter was manipulated. Decreasing the value of cd caused the modeltohaveabettermatchwiththesimulations.TheparametercdusedinOLGAcase was0.34whileavalueof0.31wasimplementedinMATLABmode. Thesimplestwaytocheckifthemodeliscorrectiscomparingthevaluesofinlet pressure and static gain from the MATLAB model by the same values from OLGA simulations.Todothis,thesteadystatevaluesofinletpressurefromOLGAsimulations wereused.Thesimulatorgivesthesteadystatevaluesastheinitialvalueofanyvariable including inlet pressure in the simulations. Therefor the initial value of the inlet pressure at each open‐loop simulation for a specified valve opening was used to be compared with those of obtained from the model. Figure 5.34 compares simple static modeltotheOLGAcase.Asclearinthefigurethereisquiteagoodmatchbetweenthe modelandOLGA.TheMATLABmodelisattachedinAppendixC.7. 90 220 SimplestaticMATLABmodel OLGAcase InletPressure[kPa] 200 180 160 140 120 100 0.2 0.3 0.4 0.5 0.6 Z 0.7 0.8 0.9 1 Figure 5.34: Simple static MATLAB model compared to the OLGA model. The blue midline inthefigurepresents thesteadystatevaluesoftheinletpressurefromOLGA simulationsandtheredmidlineisthevaluesofinletpressurefromtheMATLABmodel. The top and bottom blue lines show the maximum and minimum values of pressure oscillationsateachoperatingpointintheopen‐loopsystem. 5.3.3.3.2 CalculatingTuningParametersbasedonMATLABmodel In order to find tuning parameters based on the identified MATLAB model a closed‐looptestwithstepchangeofset‐pointwasrequired.Thesteptestwasdonebya P‐only controller as it was proposed by Jahanshahi (Jahanshahi and Skogestad 2013). The same step tests applied in section 5.3.3.2 were used here too. Two different step tests,onewiththegainvalueof Kc 0 0.1 attheinitialvalvepositionof Z 0 0.3 andthe otherwiththegainvalueof K c 0 0.15 attheinitialvalvepositionof Z 0 0.4 wereused tofindtwosetsoftuningparameters.Themethodofhowtofindthetuningparameters hasbeendescribedinsection2.9.3.2. 5.3.3.3.3 ResultsoftuningusinginitialvalvepositionofZ0=0.3 With respect to the information extracted from the step test, the parameter hasbeenfoundfromtheequation2.41as =0.2848.Theperiodofsluggingoscillations in open‐loop simulations have been Tosc = 140 Sec. The model has been run for each operatingpointseparately,meaningthattheparameterZhasbeenchangedaftereach 91 running of the MATLAB model. The parameters K c ( z ) and I ( z ) have been found as functionsofvalveopening(Z)bytheequations2.42and2.43andarepresentedintable 5.13. Table5.13:PItuningvaluesinOLGAsimulationswithinitialchokevalvepositionof30% Kc I 0.0499 484.6154 Set‐point (Inletpressure) [kpa] 148.5 0.0774 549.2308 145.5 0.3244 0.1142 613.8462 143 0.3616 0.1622 678.4615 141 0.4033 0.2229 743.0769 140 0.4307 0.2985 807.6923 138.5 0.4846 0.3908 872.3077 138 0.5084 0.5650 969.2308 137 0.5673 0.7477 1050 136.5 0.6061 0.9691 1130.8 136 0.6539 1.2338 1211.5 135.5 0.7145 1.5465 1292.3 135.3 0.7562 Valveopening 0.3000 Then gain‐scheduling with multiple controllers based on multiple identified modelswasusedtostabilizethesystem.Todothisinthesimulations,12PIcontrollers wereimplementedinOLGAwiththerelatedfoundtuningparameters.Thecontrollers couldstabilizetheflowupto75.5%ofvalveopening.Changingbifurcationpointfrom Z=26%intoZ=75.5%couldbeaverygoodresult.Figure5.35illustratestheresultof control using gain scheduling between PI controllers tuned for the initial choke valve positionofZ=30%. 92 Inletpressure(controlledvariable) P[Kpa] 150 Setpoint OLGAMeasurement 145 140 135 0 500 1000 1500 2000 2500 3000 3500 4000 Valveposition(manipulatedvariable) Z 0.8 X:3881 Y:0.7554 0.6 0.4 0.2 0 500 1000 1500 2000 2500 time(sec) 3000 3500 4000 Figure 5.35: Simulation result of control using gain scheduling between PI controllers tunedfortheinitialchokevalvepositionofZ=30%. 5.3.3.3.4 ResultsoftuningusinginitialvalvepositionofZ0=0.4 Everythinghasbeendoneinthesamewayasexplainedinprevioussectionfor Z0=0.3 except for the step test that has been run for Z0=0.4. With respect to the information extracted from the step test, the parameter has been found as: = 0.8183.Inordertodogain‐schedulingwithmultiplecontrollersinthesimulations,eight PI controllers were implemented in OLGA with the related found tuning parameters. The controllers could stabilize the flow up to Z=66.34 % of valve opening. Table 5.14 and Figure 5.36 describe the result of tuning and control using control using gain schedulingbetweeneightPIcontrollerstunedfortheinitialchokevalvepositionofZ= 40%. 93 Table5.14PItuningvaluesinOLGAsimulationwithinitialchokevalvepositionofZ=40% Kc I Set‐point Valveopening 0.3927 646.1538 141 0.4024 0.5482 710.7692 140 0.4323 0.7434 775.3846 138.5 0.4871 0.9838 840 138 0.5111 1.2751 904.6154 137 0.5713 1.8207 1001.5 136.5 0.6111 2.2661 1066.2 136.2 0.6398 2.7843 1130.8 136 0.6633 Inletpressure(controlledvariable) 144 Setpoint OLGAMeasurement P[Kpa] 142 140 138 136 134 0 500 1500 2000 X:2476 Y:0.6634 0.5 0 0 2500 Valveposition(manipulatedvariable) 1 Z 1000 500 1000 1500 time(sec) 2000 2500 Figure 5.36: Simulation result of control using gain scheduling between PI controllers tunedfortheinitialchokevalvepositionofZ=40%. 94 5.4 Comparisonofexperimentalandsimulated results The simulations in this thesis have been matched with the experimental models fromvalve1(Slowvalve).Thereforeincaseofnumericalcomparison,itisreasonableto comparesimulatedresultswithexperimentalresultsfromvalve1.Butgenerallyincase ofcomparisonofdifferenttuningmethodsandfindingthebesttuningapproachforthe sluggingsystemthesimulatedresultsdoagreewiththeexperimentalresultsfromthe bothvalves.Inthissectioneachtuningmethodwouldbediscussedseparatelyandthe result of control from simulations and experiments will be compared. Finally a comparisonofalltuningmethods,usedinthethesis,basedonthesimulationsandboth valvesexperimentswillbepresented. 5.4.1 Open‐loopbifurcationdiagrams A comparison of the simulated open‐loop results from the OLGA case and the experimental results from valve1 is shown in figure 5.37. It can be seen in the figure thatthebifurcationpoint isfairlythe sameforthebothmodels.Itoccursat thesame valveopeningofZ=0.26forbothmodelsbutatahigherpressurefortheexperiments. Modelsareslightlydeviatedfromeachother.FortheOLGAsimulationsthemaximumof inletpressureoscillationsarelocatedathighervalues. OpenloopBifurcationDiagrams 220 OLGAsimulations Experiments InletPressure[Kpa] 200 180 160 140 120 100 0.2 0.3 0.4 0.5 0.6 0.7 Valveopening[Z] 0.8 0.9 1 Figure 5.37: simulated results from the OLGA case compared with the experimental resultsfromvalve1.Thebifurcationpointisfairlythesameforbothmodels. 95 5.4.2 ComparisonofcontrolresultsfromIMC‐based tuningmethod A comparison of simulated and experimental closed‐loop responses from controllerstunedwithIMC‐basedmethodispresentedintable5.15.MaxZshowsthe maximumvalveopeningachievedwiththatcontrollerandMinPpresentstheminimum value of set‐point that is inlet pressure in kilo Pascal. The numbers are the rounded values.ThecontrollershavebeentunedattheinitialvalvepositionofZ=0.30. Table5.15: Comparisonofsimulatedandexperimentalresultsfromcontrollerstunedwith IMC‐basedmethod.ZisthelevelofvalveopeningandPistheinletpressureinKPa. Stabilitybefore control Stabilityaftercontrol withIMC‐basedPI controller MaxZ MinP Stabilityaftercontrol withIMC‐basedPID controller MaxZ MinP MaxZ MinP Experiments 0.26 177.8 0.38 154.5 0.40 154 Simulations 0.26 153 0.46 139.5 0.50 138.5 Althoughthesimulatedclosed‐loopresultsshowahigherlevelofvalveopenings, still the amount of set‐point reduction is larger for the experiments. Both models confirm that the IMC‐based tuning method is a fine approach for the slugging system. Moreover they agree that the IMC‐based PID controller has a better performance comparedtothePI. 5.4.3 ComparisonofcontrolresultsfromSimpleonline tuningmethod SimpleonlinetuningmethodbasedonMATLABmodelwasnottriedintheseries of experiments with valve1 (See section 5.1.1.4 for more explanations). Instead it was triedwithvalve2.Thereforetheresultscan’tbenumericallycomparedsincetheOLGA simulationsarebasedontheexperimentswithvalve1.However,theexperimentaland simulatedresultsdoagreeonconfirmationofthismethodasthebestmethodoftuning withthehighestlevelofstabilityforthesluggingsystem.Thiswillbeseenmoreclearly inthenextsection. 96 5.4.4 Comparisonoftuningmethods An overview of all experimental and simulated results from the applied tuning methods is presented in table 5.16 in numeric form. The maximum valve opening achieved as well as the minimum obtained set‐point for each closed‐loop test or simulationisillustrated. Itcanbesaidthatthebesttuningmethodforthesluggingsystemisthesimple onlinePItuningruleswithgainschedulingforthewholeoperatingrangeofthesystem based on MATLAB model. Tuning based on IMC design also works very well for the slugging system. These tuning methods are able to move the critical stability point significantlyandconsiderablyincreasetheproductionrateasaresult. Itwasalsotriedtomakeaclearcomparisonbetweentheappliedtuningmethods by using figures. To do this the open‐loop and the closed‐loop bifurcation diagrams wereplottedforthesimulationsandeachseriesofexperiments.Figure5.38compares theresultsofstabilizingcontrolsimulationsbydifferenttuningmethods.Figures5.39 and5.40dothesamefortheresultsofcontrolexperiments. Thebifurcationpointasasignofstabilitylevelisshownbeforeandaftercontrol witheachtuningmethod.Therightmostbifurcationpointisrelatedtothebesttuning methodthatprovidedthemoststabilityineachseries. ItshouldbenotedthatShams’smethoddidn’tworkintheexperiments.Thisis notsurprisingsinceithasbeendevelopedforthesystemsthatarestableinopen‐loop while the slugging system is highly unstable. Simple online tuning method based on MATLAB model was not tried in the series of experiments with valve1 (See section 5.1.1.4formoreexplanations). 97 Table5.16:Comparisonofsimulatedandexperimentalresultsfromalltuningmethods. Z Max is the maximum level of valve opening and P Min is the minimum set‐point achievedthatistheriserinletpressureinKiloPascal. Open‐ Shams’s IMC‐ loop set‐point basedPI stability overshoot tuning limit method method Set1of 177.8 ‐ 154.5 PMin Experiments withslow 0.26 ‐ 0.38 ZMax valve Set2of 170.8 ‐ 158.5 PMin Experiments withfast 0.16 ‐ 0.29 ZMax valve OLGA simulations IMC‐ simplePI basedPID tuning tuning withgain method scheduling 154 0.40 NOT Performed 157.5 156 0.30 0.35 PMin 153 142 139.5 138.5 135.5 ZMax 0.26 0.38 0.46 0.50 0.75 220 Comparisonofdifferenttuningmethodsfromsimulations InletPressure[Kpa] 200 180 Openloop Shams 160 IMCbased Trialanderror Simpleonline 140 120 100 0.2 0.3 0.4 0.5 0.6 0.7 Valveopening[Z] 0.8 0.9 1 Figure 5.38: Comparison of stabilizing control results from different tuning methods appliedinthesimulations.Itcanbesaidthatsimpleonlinemethodwithgainscheduling is the most stabilizing and the IMC‐based designed method is the second best as systematicmannerstotunethecontrollers. 98 Comparisonofclosedloopandopenloopstabilityfromexperimentswithvalve1 210 200 InletPressure[Kpa] 190 180 170 160 Openloop IMCbased 150 140 130 120 110 0.2 0.3 0.4 0.5 0.6 0.7 Valveopening[Z] 0.8 0.9 1 Figure5.39:ComparisonbetweenthestabilizingcontrolresultsfromIMC‐basedtuning methodandtheopen‐loopsystemfortheexperimentswithvalve1. Comparisonofdifferenttuningmethodsfromexperimentswithvalve2 220 InletPressure[Kpa] 200 180 Openloop IMCbased Simpleonline 160 140 120 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Valveopening[Z] 0.8 0.9 Figure 5.40: Comparison of stabilizing control results from different tuning methods applied in the experiments with valve 2. Simple online method with gain scheduling showsthebestperformance. 99 6 Discussionandfurtherworks 6.1 Tuningmethods The mainobjective inthisthesiswastoverifythe veryrecentlydevelopedtuning methods (Jahanshahi and Skogestad 2013) by medium scale experiments and OLGA simulationsandidentifythemostrobusttuningmethodforthesluggingsystem.From theresultsitcanbeseenthatthehighestlevelofstabilityisrelatedtothecontrollers tunedbysimpleonlinemethodbasedonMATLABmodel.Itshouldbenotedthatinthis thesis wherever simple online method has been applied gain scheduling between multiple controllers has been also performed. Since the slugging system is nonlinear andthegainofsystemchangesdrasticallywithchanginglevelofvalveopening,tuning the controllers at each operating point and then connecting them via gain scheduling has a huge effect on the control performance. It can be said that applying gain schedulingbetweentheIMC‐basedcontrollersmayalsoleadtoahigherlevelofstability compared to applying a single IMC‐based PID/PI controller. This can be tried in the future. Generally, both IMC‐based method and simple online method are very useful systematicapproachestotunethecontrollersforthesluggingsystem.Previously,trial anderrorwasmostlyusedfortuning the controllersin thesluggingsystem.It canbe saidthatthetuningrulesusedinthisthesisarefromthefirstsystematicrulesforanti‐ slugcontrolandgiveveryclearfineresults. Infutureworkstimedelaycanbeaddedtothemeasurementsintheexperimentsin ordertohaveabetterinvestigationofthesystemrobustness.Actuallythiswastriedin thisthesisasaninconclusiveeffort(Seesection5.1.1.4.2). One important point needs to be mentionedin relation with tuning based on IMC design.TheIMC‐basedPIDtuningrulesincludeafiltertimeconstantthatmeansanIMC filter must be implemented on the derivative action of the PID controller. This was impossibleinOLGAandthereforehadtobeneglected.Althoughthesimulationresults do agree with the experimental results it can’t be denied that neglecting filter action deviatesthesimulatedresultsfromthereality.Thismaybepossibleinfutureversions ofthesimulator. 100 Aboutsimpleonlinemethod,theMATLABmodelisdiscussable.Fromtheresultsit canbeobservedthattheMATLABmodeldidfittothesimulatedresultsfromOLGAand alsotheexperimentalresults.However,themanipulationsdonetofitthemodeltothe simulations and experiments may lead to the inaccuracy of the results. Also, in the MATLABmodelthevalveisassumedtohavealinearcharacteristic;howeverthismay notbethecaseforthevalveintheexperiment(Seesection5.2). Shams’stuningmethodhasbeendesignedforthestablesystemswhiletheslugging system is unstable. Therefore it may not be far from the expectation that it couldn’t work for the slugging system. This method didn’t work in any of the experimental series. Though it worked in simulations but didn’t give very good results. This even small stability found with this method in OLGA simulations may deviate from the reality. This deviation may be due to inappropriate assumptions or inaccurate initial and boundary conditions in OLGA model. The overall result can be that this method can’t be a suitable one to tune the controllers in the slugging system. Instead, the recentlydevelopedIMC‐basedandsimpleonlinemethodsperformmuchbetter. 6.2 Controlstructures The control structure used in the series of experiments and simulations was a SISOcontrolwithbufferpressureasthecontrolvariable.Thismeasurementistheriser inletpressureinrealsubseasystemsandmaynotbeveryeasytomeasure.Howeverit has been proved previously that it’s the best control variable for the active control of severe slugging (Jahanshahi, Skogestad et al. 2012) (Meland 2011) . In the article by Jahanshahi (Jahanshahi, Skogestad et al. 2012) one pressure measurement from the pipelinecombinedwithchokeflowratehasbeensuggestedasthebestmeasurements for a multivariable structure. At the beginning of the thesis it was decided to try a similar structure with top pressure combined with riser outflow density as the measurements. But this didn’t become practical during the thesis due to the inconvenientdensitysensor(Seesection5.1.3).Thenewtuningmethodsappliedinthis thesiscanbetriedbyothermeasurementsandcontrolstructuresinthefuture.Asthe first step an accurate density sensor shall be used to give correct measurements of densities.Thenitcanbeusedinthenewcontrolstructures. 101 6.3 Discussableissuesrelatedtoexperimental activities 6.3.1 Oscillationsinflowrates Inordertohaveafixed U sl and U sg ineachtestitwasimportanttohaveconstant andconsistentflowrates.Theairandwaterflowrateshadmanyoscillationsanditwas very difficult to set the exact required flow rates. Specially, for the case of air this problem was more challenging. The reason was that the control valve for the air was brokenandtheairflowratehadtobesetwithamanualvalvefarfromthescreen.The manualvalvemadeabigchangeinairflowrateevenwhenitwastriedtoopenorclose itverylittle.Itwasnecessarytogoandcomemanytimestomakeaflowratecloseto the desired value. For water flow rate the centrifugal water pump was the reason of oscillations.However,itwastriedtodealwiththisissuethroughrunningthepumpina veryhighlevelofpower(80%ofthemaximum)andopeningthewatercontrolvalvein smallvalues,instead. 6.3.2 Waterflowbackintothebuffertank Whenthebufferpressurebecamelowerthanthepressureinsidepipeline,water didflowbackintothebuffertank.Thisreducedthevolumeofbuffertankandcaused the buffer pressure deviates from the real values. This discrepancy could distract the controllerperformanceandthereforeitwasveryimportanttoremembertodrainthe buffer tank between the experiments. Installing an automatic sensor to quickly sense thewaterinthebuffertankcouldbeveryhelpfultoovercomethisissue. 6.3.3 Leakageinsteelconnection Thelaboratoryfacilitywasinawaythatasinglepipelineneededtobeconnected toanyoftherisers(SteelS‐riser,HoseL‐riserorHorizontalpipeline).Ontheotherhand severalpeoplewereworkinginthelabandonthedifferentsetupsduringthesemester. Thiscausedthepipeline‐risersconnectionsneededtobechangedseveraltimesaweek. Thiswasnotaveryeasyjobandsometimestheconnectioncouldn’tbefittedquitewell even with trying many different sealing rubber O‐rings, screws and nuts. Therefore therewassomeflowleakagefromtheconnectionduringthework.Thiscouldaffectthe accuracyoftheexperimentssincetheflowmeterswerelocatedbeforethisconnection. 102 However, the flow meters themselves were not of the best quality and their numbers may be also inaccurate. One way to overcome this occasional leakage is to make a multipleconnectionbetweenthepipelineandallriserswiththemanualvalvesforeach connection.Thenthevalvescanbemanipulatedtochangeflowdirectionsinsteadofthe timeconsumingchangeoftheconnectionsbymechanicalwork. 103 7 Conclusion Thischapterisorganizedbasedonthetasksdefinedinthethesisdescription.These taskshavebeenfollowedandthedesiredresultshavebeenobtainedmostly. 7.1 Stabilizingcontrolexperimentsusingbottom pressure Stabilization control experiments using the medium scale S‐riser setup proved thattheseveresluggingphenomenacanbedelayedtoalargeextentbyactivecontrolof production choke valve and using the bottom pressure (buffer tank pressure) as the controlvariable.Twosetsofexperimentswithtwodifferentchokevalvesshowedthat the anti‐slug control structure using bottom pressure as measurement and a good tuningmethodaswell,thestabilityregioncouldbeextendedwidely. 7.2 TestingonlinetuningrulesonS‐riser experiments Threedifferenttuningmethodsforanti‐slugcontrolweretestedonlineandtheir robustness was compared with respect to the stability limits they provided (See table 5.16andalsofigures5.39and5.40). The Shams’s set‐point overshoot method (Shamsuzzoha and Skogestad 2010) failed to stabilize the system in both sets of experiments. This was not far from the expectation,sinceShams’smethodhasbeendevelopedforthestablesystemswhilethe sluggingsystemishighlyunstable. For implementing the IMC (Internal Model Control) based tuning method (JahanshahiandSkogestad2013)themodelofthesystemwasidentifiedfromaclosed‐ 104 loop step test. The identified model was used for an IMC design, and then PID and PI tunings were obtained from the resulted IMC controller. The IMC‐based PID tuning rulescouldincreasethestabilitylimitfrom26%to40%ofchokevalveopeninginthe first set of experiments using the slow valve and from 16 % to 30 % of choke valve openinginthesecondsetofexperimentsusingthefastvalve. ThesimplePItuningruleswithgainschedulingforthewholeoperatingrangeof the system, was used as the last tuning method and proved to be the best tuning approach for the slugging system. To implement this method, a MATLAB model was modified and fitted to the steady state model of experiments. Then based on this MATLABmodelandalsoasingleclosed‐loopsteptest,thesimplePItuningruleswere found.Thistuningmethodcouldincreasethestabilitylimitfrom16%to35%ofchoke valveopeninginthesecondsetofexperimentsusingthefastvalve. 7.3 Controlusingtoppressurecombinedwith density Measurement of the topside density using a conductance probe installation was not successful. The open‐loop step test proved that the probe is not applicable as an appropriatesensortomeasuretheflowdensity.Theprobesignalcouldn’tshowaclear response to the step change and therefore was not a suitable measurement for the control targets (See figure 5.16). In order to have an efficient cascade control with densityastheinnerloopcontrolvariable,moreaccuratesignalsofdensityarerequired. Thereforenocascadeanti‐slugcontrolschemescouldbetested. 7.4 Investigatingeffectofcontrolvalvedynamics Thecriteriontoevaluatethesluggingcontrolloopisthe stabilityandsincethe valves’inherentcharacteristicsaredifferent,thelevelofvalveopeningcan’tbeusedto comparethevalves’performanceinthecontrolloop.Instead,theminimumachievable set‐points in the closed‐loop responses and also the achieved range of set‐point reductionwereusedtocomparethevalvebehaviors.Fromtheclosedloopresponses,it was proved that the slow valve has a better performance compared to the fast valve. Thismeansthattheslowvalvehasbeenalreadyfastenoughforourcontroltargetsand therehasbeennoneedtovalve2(fastercontrolvalve).Inotherwordsthestabilityof thesluggingsystemismoreaffectedbythetuningparametersforthecontrollerinstead ofcontrolvalvedynamics.Figure5.19comparestheclosedloopresponseofIMC‐based 105 controller for the two valves. With the slow valve, the IMC‐based controller has been abletodecreaseset‐pointinawiderrange,downtoalowerlevel. 7.5 ControlsimulationsusingOLGA TheOLGAmodelwasdevelopedbasedonthefirstseriesofexperimentswithvalve 1 and the implemented PID controller was fine‐tuned using the different tuning strategies.Resultsoftheexperimentsverifiedthoseofthesimulations. Inopen‐loopconditionthere wasagoodmatchbetweentheOLGA modelandthe experimentalmodelofvalve1(seefigure5.37). The same as the experimental results, the simulated ones proved that simple PI tuning rules with gain scheduling for the whole operating range of the system (Jahanshahi and Skogestad 2013) is the best tuning method providing the largest stability region for the sluggingsystem. The PI controller in the simulations, tuned by thismethod,couldincreasethestabilitylimituptothevalveopeningofZ=75%from theopenloopstabilityofZ=26%(seetable5.16orfigure5.38). From the simulation results it can be said that the IMC‐based tuning method (Jahanshahi and Skogestad 2013) is the second best systematic manner to tune the controllersforthesluggingsystem.ThePIDcontrollertunedbythismethod,increased thestabilitylimitfrom26%to50%ofchokevalveopening. TheShams’sset‐pointovershootmethod(ShamsuzzohaandSkogestad2010)was used to tune the PI controller in two initial points of Z=30% and Z=40% in the simulations.TheonetunedattheinitialpointofZ=30%couldsurprisinglystabilizethe systemuptothevalveopeningofZ=38%.However,theotheronetunedattheinitial point of Z=40% wasn’t able even to achieve the stability for the point that has been tunedfor. 106 8 References Bai,Y.(2001).Pipelinesandrisers,ElsevierScience. Bratland,D.O.(2010)."TheFlowAssuranceSite."from http://www.drbratland.com/PipeFlow2/chapter1.html. Fabre,J.,L.Peresson,etal.(1990)."Severeslugginginpipeline/risersystems."SPE ProductionEngineering5(3):299‐305. Godhavn,J.‐M.,M.P.Fard,etal.(2005)."Newslugcontrolstrategies,tuningrulesand experimentalresults."Journalofprocesscontrol15(5):547‐557. Havre,K.,K.O.Stornes,etal.(2000)."Tamingslugflowinpipelines."ABBreview4:55‐ 63. Jahanshahi,E.andS.Skogestad(2011).Simplifieddynamicalmodelsforcontrolof severeslugginginmultiphaserisers.WorldCongress. Jahanshahi,E.andS.Skogestad(2013).Closed‐loopmodelidentificationandPID/PI tuningforrobustanti‐slugcontrol.Mumbai,India,10thIFACInternational SymposiumonDynamicsandControlofProcessSystems. Jahanshahi,E.,S.Skogestad,etal.(2012).Controllabilityanalysisofseveresluggingin well‐pipeline‐risersystems.IFACWorkshop‐AutomaticControlinOshoreOil andGasProduction,Trondheim,Norway. Jansen,F.,O.Shoham,etal.(1996)."Theeliminationofsevereslugging—experiments andmodeling."Internationaljournalofmultiphaseflow22(6):1055‐1072. Kazemihatami,M.(2012).ExperimentsonLiquidFlushinginPipes.MasterProject. Trondheim,NorwegianUniversityofScienceandTechnology. Lilleby,K.(2003).User'sManualforMultiphaseFlowLoop.Trondheim,Norwegian universityofscienceandtechnology. Malekzadeh,R.,R.Henkes,etal.(2012)."SevereSlugginginaLongPipeline‐Riser System:ExperimentsandPredictions."Internationaljournalofmultiphaseflow. Meland,K.O.(2011).Stabilizationoftwo‐phaseflowinrisersfromreservoirs.Chemical Engineering.Trondheim,NorwegianUniversityofScienceandTechnology. Master. Miyoshi,M.,D.Doty,etal.(1988)."Slug‐catcherdesignfordynamicslugginginan offshoreproductionfacility."SPEProductionEngineering3(4):563‐573. Morari,M.andE.Zafiriou(1989).Robustprocesscontrol,Morari. Olsen,H.(2006).Anti‐slugcontrolandtopsidemeasurementsforpipeline‐risersystem, Master’sthesis,NorwegianUniversityofScienceandTechnology. Pickering,P.,G.Hewitt,etal.(2001).Thepredictionofflowsinproductionrisers‐truth &myth.IIRConference. 107 Rivera,D.E.,M.Morari,etal.(1986)."Internalmodelcontrol:PIDcontrollerdesign." Industrial&engineeringchemistryprocessdesignanddevelopment25(1):252‐ 265. Shamsuzzoha,M.andS.Skogestad(2010)."Thesetpointovershootmethod:Asimple andfastclosed‐loopapproachforPIDtuning."Journalofprocesscontrol20(10): 1220‐1234. Sivertsen,H.(2008).Stabilizationofdesiredflowregimes.DepartmentofChemical EngineeringNorwegianUniversityofScienceandTechnology.Master. Skogestad,S.(2003)."SimpleanalyticrulesformodelreductionandPIDcontroller tuning."Journalofprocesscontrol13(4):291‐309. Skogestad,S.andC.Grimholt(2011).TheSIMCmethodforsmoothPIDcontroller tuning.PIDControlintheThirdMillennium,Springer:147‐175. Storkaas,E.(2005).Controlsolutionstoavoidslugflowinpipeline‐risersystems. ChemicalEngineering.Trondheim,NorwegianUniversityofScienceand Technology.PhD. Storkaas,E.(2005).Stabilizingcontrolandcontrollability.Controlsolutionstoavoid slugflowinpipeline‐risersystems,NorwegianUniversityofScienceand Technology. Taitel,Y.(1986)."Stabilityofsevereslugging."Internationaljournalofmultiphaseflow 12(2):203‐217. Yan,K.andD.Che(2011)."Hydrodynamicandmasstransfercharacteristicsofslugflow inaverticalpipewithandwithoutdispersedsmallbubbles."International journalofmultiphaseflow37(4):299‐325. 108 A.LowpassfilterinLabVIEW In order to implement the low‐pass filter in the experiments the function “PID AdvancedVI”fromLabVIEWwasused.ThefunctionimplementsaPIDcontrollerusinga PID algorithm with advanced optional features. Figure 5.6, adapted from the National Instruments’website,showstheblockdiagramofrelatedthefunction. FigureA.1:PIDAdvanced(DBL) Inthepresentedfigurealphaspecifiesthederivativefiltertimeconstantandcan beavaluebetween0and1.ThedefaultisNaN,whichspecifiesthatnoderivativefilter isapplied.Therelationbetween F fromthemethodand fromLabVIEWisasfollows: f D 109 EquationA.1 B. Simulated results to get the best step tests for Shams’smethod In this appendix all simulations run to get the desired overshoot at different basisconditionsofthecontrollerarepresented.Thesimulationsarerelatedtotuningof thecontrollerbyShams’smethod(Seesection5.3.3.1).TableB.1presentstheinitialand finalvaluesofbuffer(inlet)pressureusedascontrolvariableinsimulationsbeforeand after step change and the resulting overshoot. The units are in kilo Pascal. K c 0 is the initialgainusedinthetuningsimulations.Thevaluesspecifiedbytheredcolorarethe bestresultsthosewereusedtofindtuningparameters. TableB.1:Resultingovershootstothedifferentsteptestsatdifferentinitialpositionsof chokevalve Initialset‐ Finalset‐point Kc0 Overshoot Bias pointvalue value 142 143 3.4058 0.5 142 144 1.7426 145 147 2.1049 142 143 0.5741 142 144 0.6318 142 145 1.2219 0.30 143 144 0.5402 0.1 143 145 0.5609 144 145 0.4894 149 144 0.3108 149 150 0.3308 150 151 0.3085 141 142 0.8839 145 142 0.4471 0.1 0.40 145 146 0.5028 148 149 0.3535 0.15 145 142 0.3210 110 C. SomeexamplesofMATLABscripts C.1 TuningbyShams’smethod clc clear all load Data Kc0 = -0.1; dy_s = 1; t_init = 200; dt = 0.1; t = Data(:,1); r = Data(:,2); y = Data(:,3); u = Data(:,4); %% figure(1) clf subplot(2,1,1) plot(t,r,'r',t,y,'k','Linewidth',1.5) xlim([0,1000]) title('Inlet pressure(controlled variable)') ylabel('P [Kpa]') legend('Setpoint','Data',2) grid on hold on subplot(2,1,2) plot(t,u,'k','Linewidth',1.5) xlim([0,1000]) xlabel('time(sec)') ylabel('Z') title('Valve position (manipulated variable)') grid on %% i_init = find(t==t_init); y_plant = y(i_init:end); t_plant = t(i_init:end); u_plant = u(i_init:end); y_init = y(i_init-200); u_init = u(i_init-200); yp = max(y_plant); dy_p = abs(yp - y_init); i_yp = find(y_plant==yp); t_p1 = mean(t_plant(i_yp)); yu = min(y_plant(i_yp:10*i_yp)); dy_u = abs(yu - y_init); i_yu = find(y_plant==yu); t_u = mean(t_plant(i_yu)); y_inf = y_plant(end); dy_inf = abs(y_inf - y_init); tp = t_p1 - t_init; 111 Overshoot = abs((dy_p-dy_inf)/dy_inf); D = Overshoot Offset = abs((dy_s-dy_inf)/dy_inf) B = Offset; A = 1.152*D^2 - 1.607*D +1; r = 2*A/B; K = 1/(Kc0 * B); Tetha = tp*(0.309 + 0.209*exp(-0.61*r)); tau1 = r*Tetha; Kc = tau1/(K*2*Tetha) tauI = min (tau1, 8*Tetha) Tau_c=Tetha C.2 ModelidentificationbasedonIMC‐design clc clear all close all load z30_148_150 Kc0 = -0.1; dy_s = 2; t_init = 200; dt = 0.1; t y r u = = = = z30_148_150(:,1); z30_148_150(:,2); z30_148_150(:,3); z30_148_150(:,4); %% figure(1) plot(t,r,'--r','LineWidth',2.25); hold on plot(t,y,'b','LineWidth',2.25); xlabel('time(sec)'); ylabel('Inlet pressure [kPa]'); xlim([170 300]); title('Closed-loop step response from OLGA simulations'); grid on hold on %% i_init = find(t==t_init); y_plant = y(i_init:end); t_plant = t(i_init:end); u_plant = u(i_init:end); y_init = y(i_init-100); u_init = u(i_init-100); yp1 = max(y_plant); dy_p1 = abs(yp1 - y_init); i_yp1 = find(y_plant==yp1); t_p1 = mean(t_plant(i_yp1)); yu = min(y_plant(i_yp1:10*i_yp1)); dy_u = abs(yu - y_init); i_yu = find(y_plant==yu); t_u = mean(t_plant(i_yu)); 112 yp2 = max(y_plant(i_yu:2*i_yu)); dy_p2 = abs(yp2 - y_init); y_inf = y_plant(end); dy_inf = abs(y_inf - y_init); D0 = (dy_p1 - dy_inf)/dy_inf; deltaT = t_u - t_p1; v1 = (dy_inf - dy_u)/(dy_p1 - dy_inf); z = -log(v1)/sqrt(pi^2+(log(v1))^2); Tau = (deltaT/pi)*sqrt(1-z^2); K = dy_inf/(dy_inf-dy_s); K2 = K/(K-1); alpha = (K+1)/(K-1); Tau1 = 2*z*Tau*(K-1)+sqrt(4*z^2*Tau^2*(K-1)^2+(K+1)*(K-1)*Tau^2); tp = t_p1 - t_init; Phi = atan((1-z^2)/z)-tp*sqrt(1-z^2)/Tau; E = sqrt(1-z^2)/Tau; D1 = D0/(exp(-z*(tp)/Tau)*sin(E*(tp)+Phi)); Tauz = z*Tau + sqrt(z^2*Tau^2-Tau^2*(1-D1^2*(1-z^2))); s=tf('s'); disp('The identified closed loop model:') G2 = K2*(1+Tauz*s)*exp(-0*s)/(Tau^2*s^2 + 2*z*Tau*s + 1) u = [zeros(1,round(t_init/dt)) dy_s*ones(1,round((3600-t_init)/dt)+1)]; t = 0:dt:3600; y1 = lsim(G2,u,t); plot(t,y1+y_init,'--k','LineWidth',2.25); legend('Setpoint','OLGA measurement','Identified model',3); %% %BACK CALCULATION OF THE OPEN LOOP UNSTABLE SYSTEM%% A0 A1 B0 B1 = = = = 1/Tau^2; 2*z/Tau; K2/Tau^2; K2*Tauz/Tau^2; Kp a0 b0 b1 a1 = = = = = dy_inf/(Kc0*abs(dy_s-dy_inf)); A0/(1+Kc0*Kp); -Kp*a0; -B1/Kc0; -A1-Kc0*b1; s = tf('s'); disp('Identified model:') Ge = (-b1*s-b0)/(s^2-a1*s+a0) gcl = feedback(Kc0*Ge,1); 113 C.3 DesignofInternalModelController %% Internal Model Controller (IMC) % Plant Information [Zero,Pole,Gain,Ts] = zpkdata(Ge,'v'); indRHPzero = (real(Zero)>0); zeros indRHPpole = (real(Pole)>0); poles RHPpoles = Pole(indRHPpole); NumRHPzeros = sum(indRHPzero); zeros NumRHPpoles = sum(indRHPpole); poles % indices of open RHP % indices of open RHP % RHP poles % number of open RHP % number of open RHP Tauc = 10; % Tuning parameter: time constant of the closed-loop system % for MP systems q_tilde = zpk(Pole,Zero,1/Gain); k = NumRHPpoles+1; % since Vm always contains an pole at origin for step input m = max(length(zero(q_tilde))-length(pole(q_tilde)),1); % make sure q=q_tilde*f is proper filterOrder = m+k-1; % 3. calculate filter as sum(a(k)s^k)/(tau*s+1)^filterOrder coefficients = ones(1,k); if NumRHPpoles>0 A = zeros(NumRHPpoles,NumRHPpoles); for ctRHPpole = 1:length(RHPpoles) A(ctRHPpole,:) = RHPpoles(ctRHPpole).^(1:NumRHPpoles); end b = (Tauc*RHPpoles+1).^filterOrder-coefficients(1); coefficients(2:end) = (real(A\b))'; end % computing f num = fliplr(coefficients); den = fliplr(poly(repmat(-Tauc,1,filterOrder))); f = tf(num,den); q = minreal(q_tilde*f); C = feedback(q,Ge,+1); disp('The IMC controller:') C = minreal(C) L1 = C*Ge; allmargin(L1) 114 C.4 FindingPID/PItuningrulesbasedonIMC‐design disp('IMC based PID tuning:') [Kc_PID,Ki_PID,Kd_PID,Tf_PID] = piddata(C) Ti_PID = Kc_PID/Ki_PID; Td_PID = Kd_PID/Kc_PID; disp(['Kp = ' disp(['Ti = ' disp(['Td = ' disp(['Tf = ' disp('FEED TO num2str(Kc_PID)]) num2str(Ti_PID)]) num2str(Td_PID)]) num2str(Tf_PID)]) OLGA AND CLOSE THE LOOP!') C2 = Kc_PID*( 1 + 1/(Ti_PID*s) + Td_PID*s/(Tf_PID*s+1)); L2 = C2*Ge; allmargin(L2) %% %Reduce to PI Controller C3 = balancmr(C,1); [Kc_PI,Ki_PI] = piddata(C3); Ti_PI = Kc_PI/Ki_PI; disp('IMC based PI tuning:') disp(['Kp = ' num2str(Kc_PI)]) disp(['Ti = ' num2str(Ti_PI)]) disp('FEED TO OLGA AND CLOSE THE LOOP!') C3 = Kc_PI*(1+1/(s*Ti_PI)); L3 = C3*Ge; allmargin(L3) C.5 Simplestaticmodelfittedtoexperiments %%%%Simple Static Model%%% clc clear all g = 9.81; Wg_in=0.0024; Wl_in=0.39298; W = Wg_in+Wl_in; R = 8314; M_g = 29; p_s=101325; p_vmin = 0; T=15+273.15; par.r2 = 0.025; par.A2 = pi*par.r2^2; rho_g= (p_s+p_vmin)*M_g/(R*T); %Gravity (m/s2) %Inlet mass flow rate of gas (Kg/sec) %Inlet mass flow rate of liquid(Kg/sec) %Inlet mass flow rate (Kg/sec) %Gas constant (J/(K.Kmol)) %Molecular weight of Gas (kg/kmol) %Separator pressure (pa) %minimum Pressure drop over the valve (Pa) %Riser temperature (K) %Radius of riser (m) %Cross section area of riser (m2) %Average gas density at the outlet(Kg/m3) 115 rho_l=1000; %Liquid density (Kg/m3) alpha_g = Wg_in/(Wg_in+Wl_in); %Average gas mass fraction alpha_l=(1-alpha_g)*rho_g/((1-alpha_g)*rho_g+alpha_g*rho_l); %liquid volume fraction rho = alpha_l*rho_l+(1-alpha_l)*rho_g; L_r = 5.15; z_star= 0.26; cd = 0.31; K_pc = sqrt(2)* par.A2; a = (1/rho)*((W/K_pc)^2); p_star= (rho_l*g*L_r)+p_s+ p_vmin; fz_star = z_star*cd/sqrt(1-z_star^2*cd^2); delta_p_star = a/fz_star^2; p_fo = p_star-delta_p_star; position of the valve(at z=1) (pa) z_t = 0.2:0.001:1; n = length(z_t); Pin = zeros(1,n); K_z_t = zeros(1,n); %inlet pressure at fully open %Different valve openings %Inlet Pressure %Static gain of the system (pa) for i = 1:n fz = z_t(i)*cd/sqrt(1-z_t(i)^2*cd^2); Pin(i) =(a/fz^2 + p_fo)/1000; K_z_t(i) = (-2*a/z_t(i)^3*cd^2)/1000; end figure(1) clf subplot(2,1,1); plot(z_t,Pin,'k','LineWidth',2); xlabel('Z'); ylabel('Inlet Pressure [kPa]'); hold on grid on subplot(2,1,2); plot(z_t,K_z_t,'k','LineWidth',2); xlabel('Z'); ylabel('K(z)'); hold on grid on %Average mixture density %Length of riser %Bifurcation point %Discharge coefficient of valve %Valve constant (m2) %Constant parameter 116 C.6 Onlinetuningbasedonsimplestaticmodelandaclosed loopsteptest clc clear all load z40_141_142 z0 = 0.4; Kc0= -0.15; dy_s = 1; t_init = 300; dt = 0.1; t y r u = = = = %Initial valve position in step-test %gain used for the step test z40_141_142(:,1); z40_141_142(:,2); z40_141_142(:,3); z40_141_142(:,4); i_init = find(t==t_init); y_plant = y(i_init:end); t_plant = t(i_init:end); u_plant = u(i_init:end); y_init = y(i_init-10); u_init = u(i_init-10); yp = max(y_plant); %Step change is positive dy_p = abs(yp - y_init); i_yp = find(y_plant==yp); t_p1 = mean(t_plant(i_yp)); yu = min(y_plant(i_yp:10*i_yp)); dy_u = abs(yu - y_init); i_yu = find(y_plant==yu); t_u = mean(t_plant(i_yu)); y_inf = y_plant(end); dy_inf = abs(y_inf - y_init); tp = t_p1 - t_init; deltat = t_u - t_p1; %% %%%%%%MODEL%%%%%% z = 0.3 g = 9.81; Wg_in=0.0024; Wl_in=0.39298; (Kg/sec) W = Wg_in+Wl_in; R = 8314; M_g = 29; %The operating point %Gravity (m/s2) %Inlet mass flow rate of gas (Kg/sec) %Inlet mass flow rate of liquid p_s=101325; p_vmin = 0; (Pa) T=15+273.15; par.r2 = 0.025; par.A2 = pi*par.r2^2; %Seperator pressure (pa) %minimum Pressure drop over the valve %Inlet mass flow rate (Kg/sec) %Gas constant (J/(K.Kmol)) %Molecular weight of Gas (kg/kmol) %Riser temprature (K) %Radius of riser (m) %Cross section area of riser (m2) 117 rho_g= (p_s+p_vmin)*M_g/(R*T) outlet(Kg/m3) rho_l=1000; %Average gas density at the %Liquid density (Kg/m3) alpha_g = Wg_in/(Wg_in+Wl_in) %Average gas mass fraction alpha_l=(1-alpha_g)*rho_g/((1-alpha_g)*rho_g+alpha_g*rho_l) volume fraction rho = alpha_l*rho_l+(1-alpha_l)*rho_g L_r = 5.30; z_star= 0.26; cd = 0.31; K_pc = sqrt(2)* par.A2 a = (1/rho)*((W/K_pc)^2) p_star= (rho_l*g*L_r)+p_s+ p_vmin fz_star = z_star*cd/sqrt(1-z_star^2*cd^2) delta_p_star = a/fz_star^2 p_fo = p_star-delta_p_star position of the valve(at z=1) (pa) fz = z*cd/sqrt(1-z^2*cd^2) K_z = -2*a/z^3*cd^2 K_z0 = -2*a/z0^3*cd^2 %liquid %Average mixture density %Length of riser %Bifurcation point %Discharge coefficient of valve %Valve constant (m2) %Constant parameter %inlet pressure at fully open %static gain of the system (pa) %% %%%PI tuning parameters%% T_osc= 140 %period of slugging oscillations in sec in the open loop system Betha=-log((dy_inf-dy_u)/(dy_p-dy_inf))/(2*deltat)+ Kc0*K_z0 *((dy_pdy_inf)/dy_inf)^2/(4*tp) Kc=Betha*T_osc/K_z*sqrt(z/z_star) tauI_z=3*T_osc*(z/z_star) disp('FEED TO OLGA AND FIND THE MAXIMUM STABILITY!') %% %%%%%%%%%%Plot%%%%%%% %MODEL z_t = 0.2:0.001:1; n = length(z_t); Pin = zeros(1,n); K_z_t = zeros(1,n); for i = 1:n fz = z_t(i)*cd/sqrt(1-z_t(i)^2*cd^2); Pin(i) =(a/fz^2 + p_fo)/1000; K_z_t(i) = (-2*a/z_t(i)^3*cd^2)/1000; end figure(1) clf plot(z_t,Pin,'r','LineWidth',2.5); xlabel('Z'); ylabel('Inlet Pressure [kPa]'); hold on grid on %% 118 %%%%%%OLGA MODEL load Openloop z_olga = Openloop (:,1); P_max = Openloop (:,2); P_min = Openloop (:,3); P_ss = Openloop (:,4); figure (1) plot (z_olga,P_ss,'b','LineWidth',2.5); hold on legend('Simple static MATLAB model','OLGA case',2) plot hold plot grid (z_olga,P_max,'b','LineWidth',2.5); on (z_olga,P_min,'b','LineWidth',2.5); on C.7 SimplestaticmodelfittedtotheOLGAsimulatedmodel clc clear all %% %%%%%%%STEP TEST INFORMATION%%%%%%%% % % % % % % % % % % % load z30_148_150 z0 = 0.3; Kc0= -0.1; dy_s = 2; t_init = 200; dt = 0.1; t y r u = = = = z30_148_150(:,1); z30_148_150(:,2); z30_148_150(:,3); z30_148_150(:,4); load z40_141_142 z0 = 0.4; Kc0= -0.15; dy_s = 1; t_init = 300; dt = 0.1; t y r u = = = = %Initial valve position in step-test %gain used for the step test %Initial valve position in step-test %gain used for the step test z40_141_142(:,1); z40_141_142(:,2); z40_141_142(:,3); z40_141_142(:,4); i_init = find(t==t_init); y_plant = y(i_init:end); t_plant = t(i_init:end); u_plant = u(i_init:end); 119 y_init = y(i_init-10); u_init = u(i_init-10); yp = max(y_plant); %Step change is positive dy_p = abs(yp - y_init); i_yp = find(y_plant==yp); t_p1 = mean(t_plant(i_yp)); yu = min(y_plant(i_yp:10*i_yp)); dy_u = abs(yu - y_init); i_yu = find(y_plant==yu); t_u = mean(t_plant(i_yu)); y_inf = y_plant(end); dy_inf = abs(y_inf - y_init); tp = t_p1 - t_init; deltat = t_u - t_p1; %% %%%%%%MODEL%%%%%% z = 0.3 g = 9.81; Wg_in=0.0024; Wl_in=0.39298; (Kg/sec) W = Wg_in+Wl_in; R = 8314; M_g = 29; %The operating point %Gravity (m/s2) %Inlet mass flow rate of gas (Kg/sec) %Inlet mass flow rate of liquid p_s=101325; p_vmin = 0; (Pa) T=15+273.15; par.r2 = 0.025; par.A2 = pi*par.r2^2; rho_g= (p_s+p_vmin)*M_g/(R*T) outlet(Kg/m3) rho_l=1000; %Seperator pressure (pa) %minimum Pressure drop over the valve %Inlet mass flow rate (Kg/sec) %Gas constant (J/(K.Kmol)) %Molecular weight of Gas (kg/kmol) %Riser temprature (K) %Radius of riser (m) %Cross section area of riser (m2) %Average gas density at the %Liquid density (Kg/m3) alpha_g = Wg_in/(Wg_in+Wl_in) %Average gas mass fraction alpha_l=(1-alpha_g)*rho_g/((1-alpha_g)*rho_g+alpha_g*rho_l) volume fraction rho = alpha_l*rho_l+(1-alpha_l)*rho_g L_r = 5.30 z_star= 0.26; cd = 0.31; valve K_pc = sqrt(2)* par.A2 a = (1/rho)*((W/K_pc)^2) p_star= (rho_l*g*L_r)+p_s+ p_vmin fz_star = z_star*cd/sqrt(1-z_star^2*cd^2) delta_p_star = a/fz_star^2 p_fo = p_star-delta_p_star position of the valve(at z=1) (pa) fz = z*cd/sqrt(1-z^2*cd^2) K_z = -2*a/z^3*cd^2 K_z0 = -2*a/z0^3*cd^2 %Average mixture density %Length of riser %Bifurcation point %Discharge coefficient of %Valve constant (m2) %Constant parameter %inlet pressure at fully open %static gain of the system (pa) %% %liquid 120 %%%PI tuning parameters%% T_osc= 140 %period of slugging oscillations in sec in the open loop system Betha=-log((dy_inf-dy_u)/(dy_p-dy_inf))/(2*deltat)+ Kc0*K_z0 *((dy_pdy_inf)/dy_inf)^2/(4*tp) Kc=Betha*T_osc/K_z*sqrt(z/z_star) tauI_z=3*T_osc*(z/z_star) disp('FEED TO OLGA AND FIND THE MAXIMUM STABILITY!') %% %%%%%%%%%%Plot%%%%%%% %MODEL z_t = 0.2:0.001:1; n = length(z_t); Pin = zeros(1,n); K_z_t = zeros(1,n); for i = 1:n fz = z_t(i)*cd/sqrt(1-z_t(i)^2*cd^2); Pin(i) =(a/fz^2 + p_fo)/1000; K_z_t(i) = (-2*a/z_t(i)^3*cd^2)/1000; end figure(1) clf plot(z_t,Pin,'r','LineWidth',2.5); xlabel('Z'); ylabel('Inlet Pressure [kPa]'); hold on grid on %% %%%%%%Openloop and Steady-state load Openloop z_olga = Openloop (:,1); P_max = Openloop (:,2); P_min = Openloop (:,3); P_ss = Openloop (:,4); figure (1) plot (z_olga,P_ss,'b','LineWidth',2.5); hold on legend('Simple static MATLAB model','OLGA case',2) plot hold plot grid (z_olga,P_max,'b','LineWidth',2.5); on (z_olga,P_min,'b','LineWidth',2.5); on 121