Beyond Single Loop PID Control

8/3/2019 Beyond Single Loop PID Control

1/40

1

Introduction

Beyond Single Loop PID

Control: Model-Based andCombined Feedforward-

Feedback ControlISA Philadelphia ChapterDavid B. LeachIndependent ConsultantINDUSTRIAL PROCESS OPTIMIZATION

March 19, 2003 2003 Industrial Process Optimization


2/40

2

OUTLINE

Outline Basic and Advanced Regulatory Control

Definitions

Combined Feedforward-Feedback Control Example 1: Combined Feedforward-Feedback Control of Distillation Column

Combined Feedforward-Feedback TuningMethodology

Model-Based Control and Controller Types Example 2: Cooling Tower Water Quality

Composition Control

Summary


3/40

3

Basic Regulatory Control

Primary Process Control Objective

Controlled Variable (CV) stays within apredefined limit around the setpointirrespective of routine disturbances thatroutinely affect the control loop

Feedback Control

Single loop feedback control isadequate to meet the primary controlobjective for most processes

Effect of disturbances is not taken intoaccount in advance


4/404

Basic Regulatory Control (Contd)

Basic Regulatory Control (BRC)

Basic instrumentation and controlsystem hardware, software, andconfiguration required to safely operatethe plant on a second-to-second basis

Should be able to handle routine loaddisturbances Includes sequential regulatory control

and batch logic if required

Includes required equipment interlocklogic and safety, health &

environmental controls


5/405

Advanced Regulatory Control

Advanced Regulatory Control (ARC)

Extends control system capabilitybeyond regulatory and sequentialcontrol to move the process closer toits economic optimum

Typically implemented to:Improve operating efficiency and

profitability

Increase plant production

Improve plant stability and operability

Better reject routine control loopdisturbances


6/406

Advanced Regulatory Control (Contd)

Advanced Regulatory Control (ARC)

Coordinates or ties together control formultiple loops

Typical Advanced Regulatory Controlindustrial applications:

Cascade control

Override control

Combined feedforward-feedback control

Model-based control (including ModelPredictive Control)

Inferential composition control


7/407

Feedforward Control

Feedforward Control

Sustained control error must haveenough economic impact to justify

higher design and implementationcosts

Can minimize adverse effects of:Large magnitude/frequent input

disturbances

To some degree significant process lag

Effect of disturbance variable(s) on CVmust be measurable

Cost/complexity trade-off


8/408

Feedforward and CombinedFeedforward-Feedback Control

Why Use Combined Feedforward-Feedback Control?

Feedforward control only is notpractical because it requires:Accurate modeling of the process

Ability to predict and model the effect ofall possible disturbance variable(s) onthe primary controlled variable (CV)

So Combined Feedforward-Feedbackcontrol is generally used


9/409

Feedforward Control Types

Steady-State Feedforward Control

Most simple and direct approach

No dynamic effects included

Instantaneous correction applied tomanipulated variable

May not achieve control objective ifdynamic effects are significant (and they

usually are)


10/40

10

Feedforward Control Types (Contd)

Dynamic Feedforward Control

Takes into account:

Process dynamics (usually mostsignificant)

Disturbance dynamics

Sensor dynamics

Can be implemented by: Generic dynamic compensator (most

common)

Application-specific feedforward

control strategy and calculation


11/40

11

Feedforward Dynamic Compensation

Generic FFD-FDBK Dyn. CompensatorDYNAMIC FEEDFORWARD CONTROL APPLIED AS AN ADDITIVE CORRECTION

TO PID FEEDBACK CONTROLLER OUTPUT

Process

Input

Disturbance

Variable (DV)

Steady-State

Feedforward

Computation

(Gain)

Lead-Lag +

Delay

Dynamic

Compensator

SensorSetpoint

PID

FeedbackControllerFeedforward

Additive

CorrectionControlled

Variable (CV)

Manipulated

Variable (CV)


12/40

12

Feedforward Dynamic Compensation(Contd)

Feedforward Dynamic Compensation Response of a Lead-Lag DynamicCompensator to Step Input Change

LEAD > LAG

CONSTANT

RESPONSE

LEAD < LAG

CONSTANT

RESPONSEDISTURBANCE

STEP INPUT


13/40

13

Example 1: Distillation Column CombinedFFD-FDBK Control Process Schematic

TIC

06

Solv-ent

Purifi-cationCol.A

Reboiler

DILUTESOLVENT

FEEDFIC

08

LIC

01

RECYCLED

IMPURESOLVENT

FIC

10

LIC

02TI

06

MAINFDBK

OUTER

FIC

03

FFD+FDBKTY06

FY

08

FFD1

DISTURB1

FY

10

FFD2

DISTURB2

LY

02

FFD3

DISTURB3

CONCENTRATED

SOLVENT VAPOR

TO CONDENSER

SIGNAL FROM

FROM COLUMN

B SUMP LT-02

RECYCLEDWASTE H2O FROM

COLUMN B

SP

MAIN

FDBKINNER

STEAM

PURIFIED

WASTE H2O

CONDENSATE


14/40

14

General Preparation for Tuning

BEFORE conducting any tuning exerciseswork closely with the operations personnel to:

Establish the control loop performance criteria

Determine allowable operating and understandsafety limits for the control loop and other affectedvariables

Obtain any necessary operations work and safetypermits if required

Irrespective of tuning method used:

Familiarize yourself with the process (there is nosubstitute for thorough process understanding!)

Understand in detail the data acquisition andcontrol system and algorithms used including

optional features


15/40

15

Feedforward Tuning Methodology

Recommended Procedure ALWAYS conduct at least 1-2 process

response tests

Using an appropriate input disturbancesuch as a step or pulse (symmetrical orasymmetrical doublet pulse preferred)

Conduct process response tests atdifferent parts of the normal operatingrange of the controlled variable

Average the results, assess nonlinearity

If cascades are present, conduct processresponse test(s) and tune inner feedbackloop first

Conduct process response test(s) and tunethe primary feedback controller

F df d T i M h d l


16/40

16

Feedforward Tuning Methodology(Contd)

Recommended Procedure (Contd) Continuously monitor and record the input

disturbance variable (DV) and the primary

feedback controlled variable (CV) Put primary feedback controller influencedby input disturbance (feedforward) intoManual mode and allow the controlledvariable to reach steady state

Manipulate the upstream variable thatcauses the input disturbance (e.g. vessel

feed flow controller, level controller output,etc.) to create a series of input steps orpulses of varying magnitude and duration

F df d T i M th d l


17/40

17

Feedforward Tuning Methodology(Contd)

Recommended Procedure (Contd) Observe effect of disturbance on the

primary CV and insure process response is

in direction expected and magnitude ofresponse is well above noise band

Put primary feedback controller influencedby input disturbance (feedforward) backinto Auto mode

Allow primary controlled variable to reachsteady state at same setpoint

If more than one input disturbance variable(DV) influences the primary feedbackcontroller, repeat this procedure for each

DV


18/40

18

Feedforward Tuning Methodology (Contd)

Recommended Procedure (Contd) Perform process response test resultsanalysis for each DV

Using a tuning or model identificationpackage [e.g., ExperTune, University ofConnecticuts (UConn) Control Station,MathWorks MATLAB + System ID Toolbox,etc.]

Estimate input disturbance process gain(including sign), deadtime, and first ordertime constant

Use feedforward tuning constant rule set* ortuning and simulation package to obtainfeedforward gain, lead, lag, and if reqddelay

Commission and test combinedfeedforward-feedback loop

*The authors rule set follows in next slide


19/40

19

Feedforward Tuning Methodology (Contd)

Recommended Procedure (Contd) 1st pass feedforward tuning constant rule set

Feedforward Gain = Load DisturbanceProcess Gain*/Controlled Variable ProcessGain**

Feedforward Lead = (1.3-1.5) x ControlledVariable 1st Order Process Time Constant**

Feedforward Lag = (1.1-1.3) x LoadDisturbance 1st Order Process TimeConstant*

Feedforward Delay = Load Disturbance

Process Deadtime* - Controlled VariableProcess Deadtime** (ignore if less than 0)

*Normalized effect of load disturbance variable (DV) change on primary

process control var. (CV)**Normalized effect of primary feedback controller manipulated variable(MV) move on primary process control var. (CV)

Si l t d F db k O l T Ctl L


20/40

20

Simulated Feedback Only Temp. Ctl. LoopPerformance FIC10 20% Load Disturbance

Sim lated Combined FFD FDBK Temp Ctl


21/40

21

Simulated Combined FFD-FDBK Temp. Ctl.Loop Performance FIC10 20% Load Disturb.

Example 1: Distillation Col Combined


22/40

22

Example 1: Distillation Col. CombinedFeedforward-Feedback Control Results

Adding three combined feedforward-feedback control loops with dynamiccompensation achieved:

Routine solvent purification columnoperation within environmental emissionsconstraints

Substantially reduced solvent loss Estimated savings:~ $100K/year in solvent recovery

Unestimated $/year in avoidance ofenvironmental emissions excursion fines


23/40

23

Model-Based Control

What is Model-Based Control? Embeds a process model in the control

algorithm to better achieve the control

objective Some Model-Based Control Examples

Internal Model Control

Smith Predictor with PID Feedback Control Adaptive Model-Based Control Feedback

Controller Tuning Constants OnlineAdjustment

Adaptive Model-Based Control - ProcessModel Parameters Online Adjustment

Model Predictive Control

Many Other Variants and CommercialProducts

Generic Model Based Control Block


24/40

24

Generic Model-Based Control BlockDiagram

Note: the above diagram was extracted from Techniques of Model-Based Control,Brosilow & Joseph 2002 Prentice-Hall, and was modified by the presenter.

Manipulated

Variable (MV

or MVs)

Model-Based

Controller or

Optimizer

Disturbance

Estimation and/or

Model Adaptation

Process

Model

ControlledVariable (CV

or CVs)

Operator

Target(SP or

SPs)Process

Updated

Model

Parameters

DisturbanceVariable (DV

or DVs)

Measure.

Smith Predictor with PID Feedback Control


25/40

25

Smith Predictor with PID Feedback ControlBlock Diagram

Process Model-Gain + 1st

Time Constant

TermKm

ms +1

ProcessModel-

Deadtime

Term

e-

ms

-+

Disturbance

Variable (DV)

ManipulatedVariable

(MV)

PIDFeedback

Controller

Controlled

Variable

(CV)

Std. Error

Kp e-

ps

ps +1Process

+ +

-

+

-

+

Operator

Target

(SP)

ModelCompensated

Error

Model Corrected

Controller

Feedback

Note: the above diagram was extracted from Reg. & Adv. Reg. Control: Sys. Dev, H. L.Wade 1994 ISA and Fundamentals of Process Control Theory 3rd e., P. W. Merrill2000 ISA, and was modified by the presenter.

Generic Adaptive Model Based Control


26/40

26

Generic Adaptive Model-Based ControlBlock Diagram

Manipulated

Variable (MV

or MVs)

Feedback orPredictive

Controller

Controlled

Variable (CVor CVs)

Operator

Target

(SP orSPs)Process

DisturbanceVariable (DV or

DVs)

Measure-

ment

Online

Process

Model &

Model ID &

ModelBuilder

OnlineController

and/or Model

Parameters

Calculation

Updated

Controller

Tuning

Parameters

Updated

Process

Model

Parameters

Note: descriptions in italics = capabilities of more sophisticated controllers

Model Predictive Control Definition and


27/40

27

Model Predictive Control Definition andFeatures of Interest

MPC- class of model-based controlalgorithms that compute a sequence ofmanipulated variable (MV) moves in order to

optimize the future behavior of a plant Solves control and optimization appmathematically online in real-time

Uses lineardynamic models to predict plantbehavior (Feedforward)

Corrects for mismatch between actual plantbehavior and model (Feedback)

May include operating constraints(constrained or unconstrained MPC)

May include dynamic cost optimization(objective) function

How Model Predictive Control Works


28/40

28

How Model Predictive Control Works

Manipulated

Variablesu(k)

Control Horizon

Projected Outputs

k k+1 k+2 k+p. . . . . .

TIME

Steady State TargetFuturePast

Actual

Outputs

y

V

AL

U

E

y(k + l k)

Generic Unconstrained Model Predictive


29/40

29

Generic Unconstrained Model PredictiveControl Block Diagram

Note: the above diagram was extracted from Advanced Control Unleashed, G. K.McMillan et al 2003 ISA, and was modified by the presenter.

Manipulated

Variables(MVs)

Future

SPsProcess

Model - SP

Prediction

Process

Model - CV

Prediction

ProcessMP Control

Algorithm

Disturb-

ance

Variables

(DVs)

Operator

Targets

(SPs)

Controlled

Variables

(CVs)

Model

ErrorVector

Model

Correction

Vector

CostOptimization

Obj. Function

(opt.-if avail.)

Raw Matl

& Product

Costs

-

+

-

+Controlled

Variables (CVs)

Feedback

Manipulated

Variable (MVs)Feedback

Predicted

CVsDVs

Model Predictive Control Advantages


30/40

30

Model Predictive Control Advantages

Controls complex multivariable processes andcan handle:

Large plant controller-related variable setTypical example: 100 CVs/30 MVs/10 DVs

Interactive variable and integrating processdynamics

Long dead time and time constant processes Can (depending on ctlr cap.) perform real-time

economic optimization Optimum SP range vs. single SP for each CV Includes raw material and product costs

Achieves fastest plant production rate ramping

Model-Based & Adaptive Controller


31/40

31

Model Based & Adaptive ControllerSuppliers

Model-Based (Non-Adaptive) Controller Supplier

ControlSoft, Inc. MMC - Modular MultivariableCtlr (2/3/0/0)

http://www.controlsoftinc.com/mmc.shtml

Adaptive Model-Based Controller Supplier

Universal Dynamics Technologies BrainWave

(1/1/3/0) + BrainWaveMultiMax (12/12/36/NA)

http://www.brainwave.com/product/product_index.html

Adaptive Model-Free Controller Supplier CyboSoft (General Cybernation Group, Inc.)

CyboCon (12/3/3/NA)

http://www.cybosoft.com/index.htmlNote: (_/_/_/_) = controller capability for CVs/MVs/DVs/AVs (AV=constraint)

Model Predictive Controller Suppliers
http://www.controlsoftinc.com/mmc.shtmlhttp://www.brainwave.com/product/product_index.htmlhttp://www.cybosoft.com/index.htmlhttp://www.cybosoft.com/index.htmlhttp://www.cybosoft.com/index.htmlhttp://www.brainwave.com/product/product_index.htmlhttp://www.controlsoftinc.com/mmc.shtmlhttp://www.controlsoftinc.com/mmc.shtml


32/40

32

Model Predictive Controller Suppliers

Model Predictive Controller Suppliers Aspen Technology DMCplus(100s/100s/etc.)

http://www.aspentech.com/

Emerson Proc. Auto. DeltaV Predict (4/4/4/4) http://www.easydeltav.com/

Honeywell ProfitRobust Multivariable

Predictive Controller (200/100/100/300)

http://www.acs.honeywell.com/ichome/

Intelligent Optimization GMAXC

(40/25/10/40) http://www.intellopt.com/GMAXC.htm

3rd party S/W solution from Siemens Energy and

Automation Div. for APACS+ systems

(http://www.sea.siemens.com/process/default.html)

Note: (_/_/_/_) = controller capability for CVs/MVs/DVs/AVs (AV=constraint)

Example 2: Cooling Tower Water Quality
http://www.aspentech.com/includes/product.cfm?IndustryID=0&ProductID=98http://www.easydeltav.com/pd/PDS_DV_Predict.pdfhttp://www.acs.honeywell.com/ichome/rooms/DisplayPages/LayoutInitial?PageName=Software&Type=Category&ParentCatalogName=ICHome&CategoryName=Software&ParentName=acs_products_container&SL=1&L=1http://www.intellopt.com/GMAXC.htmhttp://www.sea.siemens.com/process/default.htmlhttp://www.sea.siemens.com/process/default.htmlhttp://www.sea.siemens.com/process/default.htmlhttp://www.intellopt.com/GMAXC.htmhttp://www.acs.honeywell.com/ichome/rooms/DisplayPages/LayoutInitial?PageName=Software&Type=Category&ParentCatalogName=ICHome&CategoryName=Software&ParentName=acs_products_container&SL=1&L=1http://www.easydeltav.com/pd/PDS_DV_Predict.pdfhttp://www.aspentech.com/includes/product.cfm?IndustryID=0&ProductID=98http://www.aspentech.com/includes/product.cfm?IndustryID=0&ProductID=98


33/40

33

p g Q yComposition Control Process Schematic

CT CIRC.PUMP

WATER

QUALITY

CHEMICAL

TREATMENTSYSTEM

ORP

DOSINGPUMP

DOSINGTANKCOOLING

TOWER

RESERVOIR

COOLING

TOWERFRESH CTW

MAKE-UP(DISTURBANCE)

PROCESS

HEAT LOAD

AT

AC

PID OR MODEL-

BASEDCONTROLLER

LC



34/40

34

p g Q yComposition Control Process Identification

*The Laplace polynomial equation used to model and simulate this underdamped process in theExperTune Loop Simulator is: 1/(C0 + C1s + C2s2 + C3s3): C0=4.2; C1=92; C2=670; C3=0.

TIME UNITS (secs)

CONTROL OUTPUT (CO) STEP CHANGE

(OPEN LOOP)CO IS SENT TO A DOSING CHEMICAL

ADDITION CONTROL VALVE

POSITIONER

PROCESS

RESPONSE OR

REACTION CURVE

ORP PROCESS

VAR.s RESPONSETO BELOW CO

STEP CHANGE

EXPERTUNE IDENTIFIED & VALIDATED

PROCESS MODEL: UNDERDAMPED*

PROCESS WITH A SERIESINTEGRATOR.

FOLPDT PROCESS GAIN = 0.24

PROCESS DEADTIME = 12 MINS

SERIES INTEGRATOR TIME = 5 MINS

ENGR UNITS (%of PV or CO range)

Open Loop Process Response Test Results



35/40

35

p g Q yComposition Control Process Model Dev.

Simulated Process & Disturbance Models ControlStation



36/40

36

p g yComposition Control Controller Dev.

Simulated PID & Model-Based Controller ControlStation



37/40

37

Comp. Ctl Simulated Perform. SP Step

Optimally-TunedPID w/Smith

Predictor

Plant Originally-

Tuned 2-ModePID

Optimally-

Tuned 3-Mode

PID

Example 2: Cooling Tower Water QualityC C S f


38/40

38

Comp. Ctl Simulated Perform. Load Pulse

Optimally-Tuned

PID w/Smith

Predictor

Plant Originally-

Tuned 2-ModePID

Optimally-Tuned 3-Mode

PID

Example 2: Cooling Tower Water QualityC iti C t l R lt


39/40

39

Composition Control Results

Optimal PID Control Tuning Results: Retuned with step and pulse response

testing using ExperTune

Retuned loop reduced average ORPControlled Var. (CV) variance from +/- 20-45% before retuning to ~ average of +/- 5%after retuning

Estimated savings:Avoidance of need to shut down the plant

and manually chemically clean heat

exchange equipment ~ once/year - $100K$5K per year in reduced microbiocide

usage

Summary


40/40

40

Summary

Advanced Regulatory Control - globallyproven to provide major competitiveadvantage if properly applied and maintained

Combined Feedforward-Feedback control canminimize the negative impact of routinedisturbances for high profit control loops

Model-Based and Model Predictive Controlcan handle difficult or complex applicationswhere single loop feedback control is notadequate

These techniques have been successfullyapplied to greatly improve plant performancefor many decades

Very capable commercial tools are now

available to facilitate the process of movingbeyond single loop control

Documents

Beyond Single Loop PID Control