START UPS DO NOT HAVE TO BE DIFFICULT

Transcription

1 Cpyright 2011 ISA. All Rights Reserved START UPS DO NOT HAVE TO BE DIFFICULT Michel Ruel 1 1 BBA Tp Cntrl Inc., Green Bay, WI, USA Keywrds: Prcess Mdel, PID Algrithm, Lp Tuning, Cntrl Strategy, Alarm Management, Perfrmance 1. ABSTRACT Between design and start up, much can be dne t ratinalize, standardize, and anticipate prblems. In additin, selecting the right ptins and parameters can greatly reduce issues at start up. During the design phase, engineers can astutely make the prper chices withut additinal csts t prepare a successful start up. Fr instance, traditinally, prgrammers have much flexibility during the prgramming phase; design dcuments d nt always specify the selectin f algrithms, default values, ptins, and selectins. Similarly, HMIs shuld be designed accrding t standards. This paper will suggest steps and actins t reduce start up time substantially. A series f tips and tricks are presented, especially t select mst parameters in advance. These steps will guide the prgrammers by prviding standardizatin, default values, and better chices fr ptins. Finally, a series f analyses are suggested fr the start up phase. It will be shwn that very few tests are necessary when using mdern tls, since peratinal data are usually sufficient. Examples f successful start ups will be presented using mdern tls and apprpriate prgramming standards. Perfrmance evaluatin shuld als be part f the start up phase and apprpriate measurement shuld be defined t detect bad actrs. 2. INTRODUCTION Many prjects are designed withut cnsidering cmmissining right frm the start. Start up and cmmissining shuld be cnsidered in every step, starting with design. The designers will decide n equipment selectin specificatin, prgramming language, cmmunicatin prtcls, prgramming standards, and cmmissining methdlgy. The budgets will include the time required t perfrm the steps described in this paper; fllwing these guidelines will reduce the start up time and savings will be used t ensure better planning and buy sftware tls. The start up and cmmissining budget will include the sftware tls required t manage the field equipment, as well as t tune and ptimize the lps. This paper will fcus n prcess cntrl but a similar apprach shuld be used fr mtrs, cnveyrs, pumps, and ther equipment. 3. EQUIPMENT SELECTION Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp 2011;

2 Cpyright 2011 ISA. All Rights Reserved The prper selectin f instruments and valves can significantly reduce the time required fr calibratin and cmmissining; digital instruments and digital cmmunicatins will allw the technician t manage this wrk. Detailed instructins fr cnfiguratin and calibratin will be prvided. All filters frm instruments will be remved; filters shuld be cnfigured in the cntrl system and be part f the tuning methdlgy. It is als the respnsibility f the designer t dcument special functins and specific cnfiguratins. The number f authrized suppliers shuld be limited. Cmmunicatins have t be standardized between the field devices and cntrl system, between the cntrl system and HMI, the cntrl system and histrian, and the cntrl netwrk and specialized sftware tls. Universal prtcls will be used and prprietary cmmunicatin netwrks shuld be avided. When selecting equipment, sizing and quality will determine cntrllability. An versized valve will crrespnd t a large prcess gain (%/%) while the valve characteristic will determine the prcess linearity. Table 1 suggests limits n equipment quality; this depends mstly n the valve r final cntrl element. Prcess gain and linearity will depend n equipment sizing and chice; fr example, selecting the prper valve characteristic will mdify linearity. Hysteresis and stictin will depend n valve/psitiner assembly and quality. Vendrs can guarantee values bit it will be true nly fr new equipment. Psitiner versht will be adjusted when cnfiguring the psitiner. TABLE 1 ACCEPTABLE EQUIPMENT PERFORMANCE Acceptable Value Target Value Prcess gain > 0.5 and < 2 1 Linearity (G prcess max/g prcess min) < 2 1 Hysteresis, Dead band < 2% 0% Stictin <0.2% 0% Psitinner versht < 20% 0% 4. PROGRAMMING Traditinally, prgrammers have much flexibility during the prgramming phase. Using well defined standards will guide the prgrammers fr the DCS (Distributed Cntrl System), PLC (Prgrammable Lgic Cntrller), and HMI (Human Machine Interface). These standards must be enfrced, even if many suppliers participate in the prject; this becmes even mre crucial when several grups prgram and design the cntrl strategies. Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

3 Cpyright 2011 ISA. All Rights Reserved 4.1 Pints t cnsider Pints t cnsider when defining standards: Tag numbering (fllwing ISA and ther standards) Data flw, data cmmunicatin, data exchange Netwrk hierarchy Tasks scheduling, lad sharing, scan time Prgramming language part f the IEC standard Functins defined in details, fr basic PID cntrllers fr instance (refer t Sectin 4.2.2) Cntrl strategies structure prperly defined and standardized Cascade, rati, verride, feedfrward, etc. Transfer switches standardized: Manual/Aut, Remte/Lcal, etc. Alarms Others 4.2 PID Functin The PID functin requires special attentin. Hence, mst systems ffer several PID structures and many parameters. One PLC manufacturer ffers 11 different PID algrithms and sme f these algrithms have defects and bugs in their cde. Anther manufacturer ffers a pwerful PID algrithm with mre than 200 parameters t select; unfrtunately, sme f the default values are unacceptable, a prprtinal gain f zer fr instance PID algrithm structure Which algrithm shuld be preferred? Ideal (als frequently named ISA)? Parallel? Series? Unfrtunately, the names f these algrithms are nt standard: ne named independent culd crrespnd t a Parallel algrithm with a given manufacturer but be Ideal with anther. The designer will prefer Ideal r Series algrithm since the Parallel algrithm is nt intuitive when manipulating the parameters. Fr mst lps, the derivative is nt used and if s, the Ideal and Series algrithms becme identical Parameters and selectins The designer will nt nly impse the use f a ne and nly algrithm, but will decide hw t select all key ptins and parameters. The designer shuld cnsider building a meta blck with all prper selectins and ptins already in place. Sme manufacturers use nn sense default values. Fr instance, the fllwing parameters will be selected: Measurement filter Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

4 Cpyright 2011 ISA. All Rights Reserved First rder r secnd rder Mving average is used nly if the nise perid is cnstant; this filter adds mre dead time t the measurement and is a nn linear functin. Output slew rate, limiting rate, maximum step change per scan Nt used Used exceptinally t prtect equipment r prcess. Filter n derivative Always used. If adjustable, use a filter time cnstant apprximately 1/10 f the derivative time If n filter is available fr the derivative blck, remve the derivative ptin. Derivative n Prcess Variable (PV) r Errr Unless specified, the default value shuld be n PV t avid brutal utput changes when manipulating the setpint. Integral n PV r Errr Unless specified, the default value shuld be n PV t avid slw SP changes when manipulating the setpint Hwever, useful fr level lps r large lag t avid versht n SP changes. Lead/Lag filter n SP Default value shuld be NO which is equivalent t Integral n Errr. If a Lag value= Lead value; this is equivalent t NO If Lead=0 and Lag=Integral time (nt true fr Parallel algrithm); this is equivalent t Integral n PV. Scaling If the PID cntrl blck manipulates numbers withut scaling them 0 100%, add a scaling functin fr the prcess variable, setpint, and cntrller utput. Special functins: P2ID, PI2D, PID gap, PID nn linear, PID, gap n Integral, etc. Shuld be cnsidered as special algrithms and be prgrammed as cntrl strategies. Default parameter values (refer t next sectin) fr Prprtinal, Integral, Derivative, and Filter Summary f PID cntrller chices: Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

5 Cpyright 2011 ISA. All Rights Reserved Algrithm specified, Series r Ideal preferred All selectins and basic parameters defined Adding a measurement filter functin if nt part f the PID basic functin Adding a characterizer at the utput (r in digital psitiner) Avid limiters and special nn linear functins Cntrl strategies The designer will develp prgramming standards fr all cntrl strategies; structure, parameters, default values, and transfer switches. Fllwing are the cmmn cntrl strategies t be defined: Cascade Rati Cnstraint r verride Supervisry Feedfrward PIDF default parameters The prgrammers shuld use the suggested values fr each type f lp. Table 2 suggests the default value fr cmmn lps; these values are cnservative and will prduce a smth sluggish respnse. Befre start up a prcess cntrl engineer shuld revise the values using prcess knwledge. This persn will als validate the cntrller actin (direct r reverse) depending n the prcess and equipment Table 2 Usual tuning parameter values fr cmmn lps P I D F Scan rate ~settling time K p T i T d T F t scan units secnd secnd secnd secnd minute Flw, Pressure (liquid) <0.5 1 Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

6 Cpyright 2011 ISA. All Rights Reserved Flw, Pressure (gas) <0.5 1 Level Level (buffer) Temperature minutes Analysis minutes Others 0.2 5t d 0 t d /5 < t d 10 t d Tests The prgrammers shuld test every functin and then encapsulate the cde and default values. An easy way t test the PID functins cnsists in simulating the prcess using a series f lag functins Variable speed drives used as cntrl element All variable speed drives (VSD) used as a final cntrl element shuld be cnfigured adequately, withut excessive ramps and limiters. The cntrller utput limits are taken int accunt when adjusting the minimum and maximum values. The limits selected in PDI functin and the limits fr the cmmand speed sent t the drive shuld be identical. The lw limit shuld be adjusted t crrespnd t minimum flw; it is cmmn t bserve that the speed much reach a certain speed t build enugh pressure t vercme static pressure. Even thugh the electrical grup is respnsible fr VSD, a prcess cntrl engineer shuld review the parameters selectin. 5. COMMISSIONING AND START UP At start up, the punch lists and detailed checking are defined in advance. The malfunctin, imprper design, and wrng selectin shuld generate immediate warnings and an escalatin prcess shuld be in place. Mst cmmissining prcedures will detail visual inspectin, pwer and grunding prcedures, calibratin, instrument and valve checkut, sftware tests, etc. What is missing? At start up, it is essential t validate: Valve perfrmance Cntrl lp perfrmance in different cnditins Cntrller stability, cntrl lp respnse Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

7 Cpyright 2011 ISA. All Rights Reserved Hence, all lps must perfrm adequately, be stable, and be tuned t btain the expected settling time. 6. OPTIMIZING AND TUNING LOOPS AT START UP Hw t cnduct the tests? Hw t analyze equipment behaviur? Which steps are required? If default values frm Table 2 are used, the lps will be stable. In additin, since peratrs manipulate setpints, switch mdes, start and stp, etc. at start up, all these steps generate sufficient data t mdel the prcess, t detect equipment prblems, and t select tuning parameters. Mdern tls will detect whether the prcess mdel is linear; therwise, the valve perfrmance test shuld be cnducted. Indeed, nrmal peratinal data and their analysis using apprpriate sftware tls will guarantee a smth and very quick start up. 6.1 Open Lp r Clsed Lp test? Traditinally, lps were tuned in pen lp using bump tests. Mst tuning tls use this technique. When analyzing prcess data in pen lp, ne must be careful since everything that is prgrammed in the cntrl system will nt be part f the test. Fr instance, the cntrller executin time, filters, ramp limiters, and characterizers will nt be cnsidered. Als, simple bump tests require special techniques t detect prcess defects such as dead band, backlash, hysteresis, stictin, and nn linearities. When analyzing in clsed lp, the cntrller sampling time is included, prcess defects are part f mdeling (if setpint mves in bth directins), and special cnfiguratin and prgramming are included. Hence, when using clsed lp testing techniques, the results btained include every part f the lp and if the tl is pwerful enugh, even nn linearities will be included in the mdel. Nn linearities are usually amplitude dependent. Als, the amplitude f setpint changes shuld reflect usual prcess variable excursins frm setpint. 6.2 DATA SUITABLE FOR CLOSED LOOP TESTING What is needed t have sufficient data t mdel a prcess? The first simple answer is: mves in the cntrller utput sufficient t cause prcess variable mvement superir t nise. Frtunately, this happens all the time at start up. Mdern tls will detect if the prcess mdel is linear; if nt, valve perfrmance shuld be verified, if nt this step is skipped An example A flw lp where setpint changes ccur is used t demnstrate hw t tune this lp withut extra tests. Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

8 Cpyright 2011 ISA. All Rights Reserved Figure 1: Data used fr analyses. (The grey area has been sliced ut since the pump was stpped.) We bserve very aggressive tuning n this lp. Mst sftware will disregard this data and will be unable t mdel the prcess since scillatins and nn linearities are present. Figure 2: Mdel fund, and uncertainty. Strng nn linearities are present but the sftware identifies a grade A mdel (high quality) with a small errr. Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

9 Cpyright 2011 ISA. All Rights Reserved Figure 3: Tuning pane and simulatin. Even with nn linearities, thse tuning parameters will maintain this lp stable in all situatins. 6.3 Cntrl strategies Advanced regulatry cntrl strategies such as cascade, rati, feedfrward will als be tuned under nrmal peratin withut bump tests. Fr instance, cascade lps are tuned while the master cntrller is in autmatic mde and the secndary in cascade mde. SP changes n the master cntrller frm nrmal peratin r with small steps generated by the sftware prduce sufficient mvement in the cntrller utputs t mdel the prcess, verify nn linearities, and tune the lps. Figure 4. The mdel quality fr bth the primary and secndary lps in a cascade temperature cntrl scheme is an A Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

10 Cpyright 2011 ISA. All Rights Reserved 7. CONTROL PERFORMANCE MONITOR Cntrl perfrmance mnitring cnsists f analyzing incming signals (prcess variables, set pints, Mnitring PID cntrl lps is anther very useful tl fr start up. and state/mde) and utging signals (cntrller utputs) in rder t determine if the expected perfrmance is reached. All signals are read frm the cntrl system (distributed cntrl system, prgrammable lgic cntrller, quality cntrl system, etc.) via digital cmmunicatins. The system detects scillatins and equipments (valves transmitters, variable speed drives, etc.) that d nt behave as benchmarked, as well as prcess cntrl prblems, prcess prblems, peratin prblems, etc. The system must detect all prblems related t cntrl lps, prcess equipments, peratins, and prductin. It must als handle special cntrl strategies (cascade, feedfrward, verride, rati, etc.) and generate predefined reprts. It is a cnditin based applicatin that mnitrs, identifies, diagnses, and remedies cntrl asset issues acrss all plant layers. This sftware tl als ffers mdeling and tuning tls. This technlgy nt nly helps imprve cntrl perfrmance, but it als helps t sustain it. It cntinuusly mnitrs all regulatry cntrl assets, detects and priritize prblems, and ntify the apprpriate persnnel. The system mnitrs PID lps, advance prcess cntrl (APC), analyzers, and sft sensrs. Since the PID lp layer is particularly imprtant fr pulp and paper mills, we will fcus n that layer in this article. Figures 5, 6, and 7 shw examples f simple web reprts. Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

11 Cpyright 2011 ISA. All Rights Reserved Figure 5. Summary reprt f aggregated areas Figure 6. Summary reprt f areas, using filters t identify lps with prblems Figure 7. Treemap screensht All graphics have been generated using the Matrikn TaiJi PID tuning tl. 8. CONCLUSION FROM design TO start up, much can be dne t ratinalize, standardize, anticipate prblems, and select right ptins and parameters t reduce the pain at start up. Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp

12 Cpyright 2011 ISA. All Rights Reserved The fllwing list is used: Design Standardize, limit number f suppliers, determine in advance parameters and functins O Define netwrk architecture, cmmunicatin and data flw Prgramming Standardize, impse algrithm, parameter default values Select PIDF parameters accrdingly t lp type Standardize cntrl strategies implementatin Cmmissining and start up Use punch lists Use mdern sftware tls t verify equipment perfrmance, mdel the prcess and determine ptimal tuning parameters. 9. REFERENCES Brisk, M. L., Prcess Cntrl: Ptential Benefits and Wasted Opprtunities, 5 th Asian Cntrl Cnference, vl. 1, 2004, Brittain, H. and Ruel, M. Optimize Yur Prcess Using Nrmal Operatin Data, NPRA Cnference, Hustn TX, Octber 2008 Gsselin, C., and Ruel, M., Advantages f Mnitring the Perfrmance f Industrial Prcess, ISA Management Newsletter, January 2007, 6 8. Ruel M., Cntrl Valve Perfrmance, Chemical Engineering, Octber 2000, Zhu, Y.C. Multivariable prcess identificatin fr MPC: the asympttic methd and its applicatins. Jurnal f Prcess Cntrl (1998), Vl. 8, N. 2, Instrument Engineers' Handbk, chapter 5.6 Plantwide Cntrl Lp Optimizatin (3 rd Editin, edited by Bela G Liptak, CRC Press, Bca Ratn FI, 2002), 40 pages Distributed with permissin f authrs by ISA 2011 Presented at ISA Exp