utilizing deltav adaptive control. presenters peter wojsznis gregory mcmillan willy wojsznis terry...

84
Utilizing DeltaV Adaptive Control Utilizing DeltaV Adaptive Control

Upload: edith-lane

Post on 28-Dec-2015

227 views

Category:

Documents


3 download

TRANSCRIPT

  • Utilizing DeltaV Adaptive Control

  • PresentersPeter Wojsznis

    Gregory McMillan

    Willy Wojsznis

    Terry Blevins

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test Results Adaptive Control DemosActuator DiagnosticsSummary

  • Control Loop Performance A Never Ending CycleTestCalculateDeployOperateDegradationEvaluateThe more often you Tune, the better the performance.

  • Operating Condition ImpactProcess gain and dynamics may change as a function of operating condition as indicated by PV, OUT or other measured parameters e.g. plant throughput

  • There Must Be A Better WayWouldnt it be nice to have controllers use optimal tuning parameters all the time (continually) without having to tune at all, ever?

  • DeltaV Adapt Fully Adaptive PID Control Tuning Learns Process Dynamics While In Automatic Control No Bump Testing Required Works On Feedback And Feedforward Patents Awarded! See It HereNo Tuning Required!

  • DeltaV Adapt Applicable to Most Control Loops Today

  • DeltaV Adapt A Clear Difference

  • Not an overnight thingEMERSON technology developed in Austin.Patents have been awarded.1997 - Dr. Willy Wojsznis concept originated1998 - Research started at Tech Center - Austin1999 - Dr Seborg started work on formal proofs2002 - Development started2003 - Prototypes at Eastman Chemical, Solutia and UT with good results.

  • Patents Have Been Awarded!Mr. Terry BlevinsDr. Wilhelm Wojsznis

  • Solid theoretical background1999 - Dr. Seborg started working on formal proofs of convergence for us along with his Emerson funded grad student

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test Results Adaptive Control DemosActuator DiagnosticsSummary

  • Adaptive PID Techniques

  • Model Free vs Model Based Adaptation

  • Model Switching AdaptationUse N-models working in parallel Evaluate model errorSelect model with minimal errorShortcoming: TOO MANY MODELS

  • Parameter InterpolationEvery parameter value of the model is evaluated independentlyThe weight assigned to the parameter value is inverse of the squared errorAdapted parameter value is weighted average of all evaluated values

  • Advantages of Parameter InterpolationSequential parameter adaptation less models: Example: Model with 3 parameters (Gain, Lag, Dead Time) and 3 values for every parameter has 3x3x3 model variations for model switching adaptation and 3+3+3 model variations for sequential parameter adaptationBetter convergenceInterpolation gives better model due to continuous adaptation of the model parameter value over the whole assumed range

  • Parameter Interpolation - CalculationsFor each iteration, the squared error is computed for every model I each scanEi(t) = (y(t) Yi(t))2 Where: y(t) is the process output at the time t Yi(t) is i-th model output A norm is assigned to each parameter value k = 1,2,.,m in models l = 1,2,,n.Epkl(t) = Ni=1 (klEi(t))

    =1 if parameter value pkl is used in the model, otherwise is 0

    For an adaptation cycle of M scans sumEpkl = Mt=1 (Epkl(t)), Fkl = 1/sumEpkl

    pk(a) = pk1fk1++pklfkl++pknfkn

    fkl = Fkl / sumFk

  • Simple Example Pure Gain Process

  • First Order Plus Dead Time ProcessModel Parameter InterpolationFor a first order plus deadtime process, only nine (9) models are evaluated each sub-iteration, first gain is determined, then deadtime, and last time constant. After each iteration, the bank of models is re-centered using the new gain, time constant, and deadtime

  • First Order Plus Dead Time Process Model Parameter Interpolation

  • Model VerificationFinal stage of model adaptation and verification showing actual response and response calculated by the identified models.

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test Results Adaptive Control DemosActuator DiagnosticsSummary

  • DeltaV Adaptive ControlOperational FeaturesProcess models are automatically established for the feedback or feedforward paths. Model adaptation utilizes a data set captured after a setpoint change, or a significant change in the process input or output. Multiple models are evaluated and a new model is determined

  • DeltaV Adaptive ControlOperational FeaturesModel is internally validated by comparing the calculated and actual process response prior to its application in tuning. The user may select the tuning rule used with the feedback model to set the PID tuning.

  • Adaptive Control - Internal StructureControl

  • Defining Operating RegionsAdaptive control allows operating regions to be defined as a function of an input state parameter

    Define up to 5 regions

    When the state parameter changes from one region to another, the model values (and associated tuning) immediately change to the last model determined for the new region

    Limits on model parameter adjustment are defined independently for each region.

  • Configuration of Adaptive ControlNew control block in the advanced control palette. Parameters are automatically assigned to the historian.No more difficult to use than PID. Initial values for model, limits, and time to steady state are automatically defaulted based on block tuning.

  • Adaptive Control ApplicationUsed to view the operation of modules that include Adapt blocks.May modify adaptive operation, parameter limits, and default setup parameters from this view. Adapt blocks run independent of the DeltaV Adapt application.

  • Adaptive Control OperationAdapt Application 1. Select Window

    2. Observe loop plots (PV, OUT, SP)

    3. Observe Adaptation Status

    4. Operate PID loop SP, OUT, Mode

  • Feedback AdaptationAdapt Application 1. Select Window

    2. Observe model parameters trends and controller tuning parameters (Gain, Reset, Rate)

    3. Observe current process model and PID tuning parameters

    4. Select Adaptive operation mode

    5. Select tuning rules

  • Feedforward Adaptation Adapt Application Select WindowObserve model parameters trends and controller tuning parameters (Gain, Reset, Rate)Observe current process model and PID tuning parametersSelect Feedforward Adaptive operation modeSelect Gain FF Factor

  • Multi-range adaptation 1. Up to 5 ranges

    2. The last adapted process gain, time constant and dead time is displayed for every range

    3. State parameter: PV, OUT or feedforward inputRange 1Range 2State parameter

  • Adaptive Control SetupAdapt Application 1. Trigger to adapt

    2. Controller Output pulse injection

    3. How fast to adapt

    4. Process type: Integrating Non Integrating; Minimum time to steady state

    5. Adaptive mode of operation Defaults are set for a typical operation!

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test ResultsAdaptive Control DemosActuator DiagnosticsSummary

  • Simple Example Non-Linear Installed CharacteristicsProcess gain will change as a function of valve position if the final control element has non-linear installed characteristics.Valve position is used as the state parameter if ranges are applied

    Chart1

    0

    24

    42

    55

    70

    76

    82

    86

    89

    92

    93

    Flow

    Stem position %

    Flow %

    Flow vs Stem Position

    Sheet1

    Stem PositionFlow

    00

    1024

    2042

    3055

    4070

    5076

    6082

    7086

    8089

    9092

    10093

    Sheet1

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    0

    Flow

    Stem position %

    Flow %

    Flow vs Stem Position

    Sheet2

    Sheet3

  • Example Throughput Dependent Process The process deadtime for superheater outlet temperature control changes as a function of steam flow rateSteam flow rate is used as the state parameter

  • ExampleMultiple Valves - Split Range The process gain and dynamic response to a change valve position may be different for each valve.Typical example is heating/cooling of batch reactor, extruder, slaker, etc.Valve position is used as the state parameter.

  • ExampleColumn Temperature Control The sensitivity of tray temperature to changes in distillate to feed ratio is highly non-linear. Tray temperature is used as the state parameter.

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test ResultsAdaptive Control DemosActuator DiagnosticsSummary

  • Adaptive Field Test

    Emerson Process Management - Process Systems and Solutions LabUniversity of Texas reactive distillation columnEastman ChemicalSolutia

  • DeltaV Adaptive ControlField Trials Eastman ChemicalControl automatically adapts based on SP changes in Auto Caustic loop

  • DeltaV Adaptive ControlField Trials Solutia, Pensacola, FLCooling Tower WaterCooling Tower WaterHMD(Base)65AC681(pH)65TC68865TC68465TC68565LC682Strike Kettle Process and InstrumentationAcid feed from centrifuge splitter

  • DeltaV Adaptive ControlField Trials Solutia, Pensacola, FLKettle control regular PID control

  • DeltaV Adaptive Control Field Trials Solutia, Pensacola, FLKettle control Adapt control

  • DeltaV Adaptive Control - Field Trials - J.J.Pickle Research Campus, UT, Austin, TXFlow loop

  • DeltaV Adaptive Control - Field Trials - J.J.Pickle Research Campus, UT, Austin, TXFlow loop

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test ResultsAdaptive Control DemosActuator DiagnosticsSummary

  • Simple Reactor Demo - Adaptation Demo with model scheduling Adaptation with process state parameter defined for two ranges

  • Simple Reactor Demo - Feedforward Adaptive Control Feedforward dynamic model is adaptedFeedforward controller is automatically updatedUp to 5 ranges can be defined for feedforward adaptation and model scheduling

    Inlet temperature as feedforward input

  • Adaptive pH Control for Fun and ProfitThe pH electrode offers an extraordinary rangeability and sensitivity. The price is an extreme nonlinearity. This demo shows how the DeltaV adaptive controller can provide a more efficient and faster approach to set points and rejection of disturbances. A high fidelity dynamic simulation and online process performance indices embedded in DeltaV show reagent savings of 40%.

  • Top Ten Signs of a Rough pH StartupFood is burning in the operators kitchenThe only loop mode configured is manualAn operator puts his fist through the screenYou trip over a pile of used pH electrodes The technicians ask: what is a positioner?The technicians stick electrodes up your noseThe environmental engineer is wearing a maskThe plant manager leaves the countryLawyers pull the plugs on the consolesBob and Bob are on the phone holding for you

  • Tremendous Rangeability and Sensitivity of pH Creates Exceptional Control OpportunitiespHHydrogen Ion ConcentrationHydroxyl Ion Concentration01.00.0000000000000110.10.000000000000120.010.00000000000130.0010.0000000000140.00010.000000000150.000010.00000000160.0000010.0000000170.00000010.000000180.000000010.00000190.0000000010.00001100.00000000010.0001110.000000000010.001120.0000000000010.01130.00000000000010.1140.000000000000011.0

  • Severe Strong Acid - Strong Base Nonlinearity (Gain Changes by factor of 10 for each pH unit)Zoom in on 3 to 10 pHEntire Operating RangeThere are no straight lines in pH - graphical deception is common

  • Weak Acid and Base - Moderated Nonlinearity(Gain changes by factor of 50 from 9 to 7 pH)Optimum set pointFor acidic influent

  • Model and Tuning Settings are Scheduled Based on What is Learned in Operating RegionsModel and tuning is scheduled based on pH

  • User Sees Adapted Model Parameters and Chooses Tuning Method

  • Opportunity in pH is Huge When Moving to a Flatter Portion of Titration CurvepHReagent to Feed Flow Ratio Reagent SavingsOriginal set pointOptimum set point

  • Adaptive Control Achieves Optimum Set Point more Efficientlyhourly cost of excess reagenthourly cost of excess reagenttotal cost ofexcess reagenttotal cost ofexcess reagentpHpH

  • Adaptive Control Recovers from Upsets more Effectivelyhourly costof excesshourly cost of excesspHtotal costof excesstotal cost of excesspH

  • Adaptive Control Returns to Old Set Points with Less OscillationpHpH

  • Component Balance and Online Process Performance Indicator are Embedded in DeltaV

  • Charge Balance is Done in Excel Spreadsheet

  • Advantages of DeltaV Adaptive pH ControlAnticipates nonlinearity by recognizing old territoryModel and tuning settings are scheduled per operating regionRemembers what it has learned for preemptive correctionDemonstrates efficiency improvement during testingSteps can be in direction of optimum set pointExcess reagent useage rate and total cost can be displayed online Achieves optimum set point more efficientlyRapid approach to set point in new operating regionRecovers from upsets more effectivelyFaster correction to prevent violationMore efficient recovery when driven towards constraint Returns to old set points with less oscillation Faster and smoother return with less overshoot

  • 3rd Edition Features Online pH Estimators andAdaptive Control

  • OutlineIntroductionAdaptive Technique BackgroundAdaptive PID Design Application ExamplesField Test ResultsAdaptive Control DemosActuator DiagnosticsSummary

  • Components of the self-diagnosed adaptive control loopPerformance -Variablity, Standard DeviationFinal Element/Valve Hysteresis, StickinessLoop Adaptation Model qualityLoop Stability Monitor Oscillations IndexDiagnostic RoutineCorrective Action, Alarm or Message

  • Loop performanceDeltaV PID loop has two performance indexes as normal loop parameters:Variability IndexStandard deviation

    DeltaV Inspect application allows easy setting and review of the loop performance in the systemThe indexes will be used as a part of diagnostic information of the Adaptive loop

  • Model QualityFinal stage of model adaptation and verification showing actual response and response calculated by the identified models. The model error indicates model quality.Other factors include adaptation history and model convergence

  • Loop stabilityOscillation index:

    Loop oscillation amplitudeOscillation period

    The highest priority loop diagnostic parameter

  • Valve diagnosticsCalculation of the valve parameters:Valve backlashValve stickinessValve hysteresisTwo complementary techniques are used:Loop oscillation analysisUse of valve stem position (BKCAL)

  • Valve diagnostics based on oscillation analysisPIDAOProcessSPOUTVPPVOUT oscillations caused by valve backlash and stickinessPV oscillations caused by valve stickiness

  • Valve backlash causes oscillations on the controller outputOUTPV

  • Valve backlash and stickiness cause oscillations on the controller and process outputOUTPV

  • Valve diagnostics based on oscillation analysis

    Define oscillation amplitude on the controller output -

    Define oscillation amplitude on the controller input -

    Calculate hysteresis as

    Knowing process gain

    , calculate valve stickness (resolution) as:

    Calculate backlash as

    _1153035955.unknown

    _1153036582.unknown

    _1153038759.unknown

    _1153038812.unknown

    _1153036162.unknown

    _1153035850.unknown

  • Valve diagnostics based on known valve stem positionPIDAOProcessSPOUTVPPVBKCALParameter representing valve stem position

  • Hysteresis calculation based on known valve stem position

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (1)

    Selection of the highest values of

    averaged over certain period of time may be considered as actuator backlash and resolution estimate.

    maximum values selection

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (2)

    Where

    b valve backlash

    h - valve hysteresis

    - valve resolution

    i - back calculation signal delay in scans, accounting for valve speed of response (velocity limit)

    _1151836903.unknown

    _1151836940.unknown

    _1152084739.unknown

    _1149407275.unknown

  • Backlash calculation based on known valve stem position

    If

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (3)

    Then

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (4)

    The resolution then can be easy found as:

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (5)

    where

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (6)

    MACROBUTTON MTPlaceRef \* MERGEFORMAT (7)

    _1153034201.unknown

    _1153034329.unknown

    _1153036307.unknown

    _1153034225.unknown

    _1151827797.unknown

  • Adaptive Loop self-diagnostics overview window

  • Valve diagnostics summarySelf diagnosed control loop contains four basic components: Performance, Adaptation, Stability and Valve

    Model Based Adaptive Control extends diagnostic features and calculation options (valve parameters)

    Valve diagnostic is a key component of the loop diagnostic

  • SummaryAdaptive Control technique with model parameter interpolation is an unique, theoretically sound and practically proven technologyDeltaV Adaptive configuration is compatible with PID controller and can replace in principle PID in every loop Feedforward adaptation and model scheduling enhance adaptive features Easy to use adaptive application makes settings and operation of adaptive loops easy

  • How to take advantage of Adaptive Control in your plant soonIdentify difficult to tune loopsIdentify loops you want to improve operationContact: [email protected] [email protected] Become an adaptive control Beta installation

  • DeltaV Product Manager John Caldwell

    Title and description should match the submitted abstract information.

    The Emerson Exchange logo should not be modified.Insert the presenters and their company logos in the order that they will present.To control efficiently we need well tuned controllersTo tune controllers accurately current process dynamics must be known. This requires process testing.After testing, updated tuning parameters may be calculated and then used for control.Over time process dynamics will change leading to poor control with the current tuning parameters.This means more tuning which requires more testingThis entire process must be repeated again and again in an attempt to maintain acceptable control performance.

    Title and description should match the submitted abstract information.

    The Emerson Exchange logo should not be modified.To control efficiently we need well tuned controllersTo tune controllers accurately current process dynamics must be known. This requires process testing.After testing, updated tuning parameters may be calculated and then used for control.Over time process dynamics will change leading to poor control with the current tuning parameters.This means more tuning which requires more testingThis entire process must be repeated again and again in an attempt to maintain acceptable control performance.