RCS for process control: Is there anything new under the sun?

RCS for process control: Is there anything new under the sun?
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  18 th  European Symposium on Computer Aided Process Engineering – ESCAPE 18Bertrand Braunschweig and Xavier Joulia Editors!" #$$8 Elsevier B%&%'(td% All rights reserved% RCS for process control: Is there anything new under the sun? )anuel *odr+gue,- *icardo San,  ASLab-UPM, José Gutiérrez Abascal, Madrid 28006, Spain Abstract .he purpose o/ this paper is to e0plore the potential use in process control o/ cognitivearchitectures used in other domains% A well2nown cognitive architecture- used tocontrol comple0 systems o/ di//erent areas has 3een selected% 4t has 3een compared withthe current control strategy used in process plants% Conclusions on its applica3ility- itsstrengths and wea2nesses are also presented in the paper% Keywords 5 comple0 control- cognitive architecture- process control% 1. Introduction .he process industry is 6uite mature in many aspects% 7ne o/ these is process control%Although decentrali,ed control 3ased on P4 controllers still is e0tensively used-signi/icant advances and research! have 3een made5 multivaria3le predictive control-use o/ simulation models- realtime optimi,ation- change in communication protocolshy3rid or digital!% Even the implementation o/ the control architecture may change and/latten! in the /uture i/ standard protocols li2e 4ndustrial Ethernet apply to all the levelso/ the plant 91:% But still the classic hierarchical organi,ation in /our levels remains% ;owadays many /ields loo2 into other domains to see i/ the ideas developed /or thesrcinal domains can 3e success/ully applied to their domains% 4n this paper a cognitivearchitecture success/ully applied to implement comple0 controllers in di//erent areas isconsidered and its possi3le application to the process industry studied%.he paper is organised as /ollows5 ne0t section presents the *CS cognitive architecture-its components and organi,ation- section three descri3es how process control iscurrently implemented in most industrial plants- section /our compares 3oth approachesand /inally section /ive draws some conclusions out o/ the presented ideas% 2. RCS: the cognitie architecture *CS *ealtime Control System! 9#<: is a cognitive architecture designed to ena3leany level o/ intelligent 3ehavior% 4nitially 3ased on a theoretical model o/ thecere3ellum- it has 3een evolving over the last three decades% .oday it is a realtimecognitive control architecture with di//erent applications% 4t has 3een used /or intelligentmachine tools9=:- /actory automation systems9>: and intelligent autonomous systems9?:among others%*CS is a multilayered multiresolutional hierarchy o/ computational agents or nodes%*CS nodes have vertical hierarchical! as well as hori,ontal relationships% Each node/ollows a common design pattern- 3eing composed o/ the /ollowing elements5 sensory processing SP!- @orld )odeling @)!- value udgment &J!- 3ehavioral generationB! and 2nowledge data3ase !%  #  M !odri"uez et al Digure 1 shows the 3asic control node with its elements and relationships%  Dig 1% *CS Computational control! node% A 3rie/ description o/ the di//erent elements o/ each o/ the control agents /ollows% Sensory !rocessing:  .his element gets sensory input and compares these o3servationswith e0pectations generated 3y an internal world model% "orld #odel:  4t is the systems 3est estimate o/ the state o/ the world% .he worldmodel includes a data3ase o/ 2nowledge ! a3out the world% 4t also contains asimulation capa3ility which generates e0pectations and predictions% 4t can provideanswers to re6uests /or in/ormation a3out the past- present and pro3a3le /uture states o/ the world  #$at i%   and #$at is  6ueries!% .his in/ormation goes to the tas2 decomposition element in the B! and to the sensory processing element% $alue %udg&ent:  4t determines what is good and 3ad% 4t evaluates the o3served and predicted state o/ the world% 4t computes costs- ris2s and 3ene/its o/ o3served situationsand o/ planned activities% 'ehaior (eneration:  Behavior is generated in a tas2 decomposition element that plans and e0ecutes tas2s 3y decomposing them into su3tas2s- and y se6uencing thesesu3tas2s so as to achieve goals% oals are selected and plans generated 3y a loopinginteraction 3etween tas2 decomposition hypothesi,e plans!- world modeling predictresults! and value udgment evaluate the results!% Behavior generation is typically donevia Dinite State )achines or rules governing costs /or nodes and edges in a graphsearchmethod% .his node contains three su3elements5 the Jo3 Asigner- the Planner and the   !&S %or process control' (s t$ere an)t$in" ne* under t$e sun+ FE0ecutor% *CS systems are 3uilt /ollowing the *CS methodology that has several steps%.he /irst one is to gather domain 2nowledge with e0perts- and to generate with the helpo/ the e0perts! the hierarchical tas2 decomposition% Knowledge )atabase:  4t stores in/ormation a3out space- time- entities- events- states o/ the system environment and a3out the system itsel/- parameters o/ the algorithms-models /or simulation- etc% *. +he process control hierarchy .he process industry comprises mainly continuous 3ut 3atch processes also% .heindustries involved are chemical- petrochemical- pharmaceuticals- re/ineries- etc% .heseare usually very large and comple0 /acilities% .he main goal o/ any process plant is toget the ma0imum 3ene/it which means the demanded amount with the speci/ied 6ualityusing the less resources! assuring sa/ety and sta3ility o/ the plant% 4n order to achievethis goal- control strategies have 3een applied and evolved over the years as newcapa3ilities were availa3le% Drom the initial manual control to the current digitaldistri3uted control system CS!%.o handle the comple0ity o/ the plant and to still achieve the overall goal- a controlhierarchy has 3een developed and used /or many years% .his architecture gets thecompany policy several wee2s time resolution! and re/ines it to the current action to 3eapplied on any actuator o/ the plant mssec resolution time!% .he procedure is too3serve the state o/ the plant through thousands o/ sensors and evaluate the ne0t action/or any resolution time% 4mplicit- e0plicit- heuristic and /irst principles models are usedin order to generate the ade6uate action% .he common process control architecture has/our control levels% .he lower level o/ the architecture is the 3asic regulatory control-this control is achieved 3y single decentrali,ed loops% )ost o/ these loops are controlled 3y standard P4 controllers% .he actuating hori,on at this level is ust one% .he second level is the advanced and predictive control% .hese are two di//erent controlschemes that wor2 at the same level% 4n/ormation is transmitted hori,ontally andvertically in this and upper! level% )ore ela3orated control strategies as selectivecontrol- ratio control- /eed/orward control are implemented% 4n this second level implicitas well as e0plicit heuristic and /irst principles 3ased! models are used to generate theaction% .he action is the set point goal! to achieve at the lowest level% Predictionhori,on is in the case o/ model predictive control! o/ tens o/ movements% Gpper levels o/ control deals with optimi,ation- scheduling and planning% Gnitoptimi,ation can 3e made online with continuous in/ormation /low /rom and to thelower levels% Site optimi,ation- scheduling and planning are done o//line% &erydi//erent types o/ models are used in these levels% As commented- in/ormation /lowsvertically and hori,ontally through the architecture and each upper level is o/ lower timeresolution% ,. RCS s )CS )any similarities e0ist 3etween the two architectures- as can 3e o3served in /igure #% 7/ course in the process control system there is no a common identi/ied computationalagent with so well de/ined elements as in the *CS architecture- 3ut at any level a good  <  M !odri"uez et al matching can 3e esta3lished- as it is shown in the /ollowing comparison 3etween the process control and the *CS elements Dig #% CS vs% *CS architecture Regulatory control node. .his is the simplest node% 4t implements the simplest *CSnode- one in which the 3ehavior is purely reactive% 4t has a @orld model the P4algorithm is a model- an empirical or heuristic model- 3ut a model o/ the system under control!- 3ut this model does not predict the 3ehavior% 4t only reacts to the current valueso/ the plant and decides an action to 3e per/ormed there is no plan- it is ust an action/or the ne0t time!% 4t can 3e considered to have a  where the model parameters arestored% &ery simple preprocessing is per/ormed 3ut some it is done as signal /ailure-%%%! #odel predictie control node. )PC has several components% 4t has a model usuallyan identi/ied linear! o/ the world% 4t has a  where past values o/ the manipulatedvaria3les )&s! and controlled varia3les C&s! are stored% 4n this  other in/ormationis stored as )&s and C&s limitations- weighting /actors- etc% .he model uses the inputsto predict the /uture% .his state is used in Behavior eneration module% 4n this modulean optimi,ation is per/ormed to select the 3est action plan% .his plan a set o/ movements /or the )&s along with C&s values! is set and sent to the regulatory level%Some preprocessing is implemented as well% .he )PC module implements also a/eed3ac2 loop to correct model errors due to model mismatch with the actual plant!% Real +i&e -pti&iation. .his module receives the values o/ the varia3les o/ the plant- per/orms reconciliation on these values% .his node has a steady state mathematical- physically 3ased! model o/ the plant% An optimi,ation is made using that model everyhour or so% .he optimi,ation results are sent to the lower level- the supervisory control%.hese results are the new set points o/ the controlled varia3les% .he 3est operating pointo/ the plant which means a set o/ set points values! is calculated in each optimi,ation%.he optimi,ation ta2es into account constraints on the varia3les limited change inmanipulated varia3les- sa/ety- 6uality- etc% constraints in controlled varia3les!% .he nodeuses as well a historian module with past data o/ the plant%   !&S %or process control' (s t$ere an)t$in" ne* under t$e sun+ = !lanning and scheduling. .his module corresponds with the 3usiness part o/ thecontrol hierarchy% 4t has a 3usiness model and 3ased on plant data current and pastvalues!- on e0ternal data mar2et data- e0ternal plant in/o- etc%! and using the company 3usiness goals derives a production plan /or the plant% 4t gives capacity productionvalues as well as 6uality values to the lower- optimi,ation- level% .he resolution time atthis level is days or wee2s% Dig F% Process control levels as *CS agents .he control levels introduced a3ove are presented in the /igure F% /ollowing thestructure o/ a *CS node% 4t can 3e o3served that the node in any level complies with the*CS node% As a preliminary conclusion it can 3e said that the conventional controlstructure is *CS compliant- or can 3e considered as an implementation o/ it% So is t$erean)t$in" ne* under t$e sun+, *$ats t$e bene%it o% usin" !&S or ot$er t)pe o% co"niti.earc$itecture/ %or process control+ .he answer is that it depends on the application and on the point o/ view% 4n spite o/ thisand 2nowing that CS is *CS compliant some di//erences or capa3ilities must 3estressed5
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks