forward siemens control

8/14/2019 Forward Siemens Control

1/12

APPLICATION DATA

AD353-129Rev 2

April 2012

Procidia Control SolutionsFeedforward Control

This application data sheet describes severalfeedforward control methods. Any of theseconfigurations can be readily developed within aSiemens 353 controller.

1

Feedforward control is a method of compensating forprocess load disturbances before they can affect theprimary control variable. This control strategy is

particularly beneficial for critical process control loopsthat are often slow to respond to load disturbances.By reducing the effect of these load disturbances,feedforward control minimizes the magnitude andduration of the control errors that would normallyoccur without this control strategy.

In the heat exchanger example shown in Figure 1,exit temperature is the primary control variable. Thetemperature controller (TC) manipulates the flow of

1See Applications Support at the back of this publication

for a list of controllers.

steam to the heat exchanger to control thetemperature at setpoint. An increase in the feed flowto the heat exchanger will cause a decrease in theexit temperature. The temperature controllerresponds to this control error by increasing the steamflow as required to return the temperature to setpoint.The change in feed flow represents a change in loadon the heat exchanger and is referred to as a load

disturbance. Other load disturbances that can affectthe exit temperature are changes in inlettemperature, ambient temperature, and steampressure.

The temperature controller can compensate for anyand all load changes. That is the essentialadvantage of feedback control. No matter what iscausing a change in exit temperature, the controllerwill detect the error and change the steam flow asrequired to reduce the error to zero. Figure 2 is anexample of the response of a single-loop PI controllerto a change in feed flow. As expected, the controller

returns the exit temperature to setpoint after the loaddisturbance from the change in feed flow.

Steam

Feed

TT

T

Heat

Exchanger

Condensate

TC

ExitTemperature

The disadvantage of feedback control is that it candetect the load disturbance only after the disturbancehas occurred. Due to the dynamics of the process,the load disturbance will go undetected for some timebefore the exit temperature begins to change, andthen additional time will be required for the exittemperature to respond to the change in steam flowmanipulated by the controller to compensate for thenew load.

Another disadvantage of feedback control is thepotential for unstable operation. Since all feedbackcontrol loops will cycle if tuned to respond too quickly,the speed at which the temperature controller cancompensate for the disturbance is restricted by thestability limit of the control loop.

Figure 1 Heat Exchanger Temperature Control


2/12

AD353-129

Exit Temperature

Steam Flow

Feed Flow

Figure 3 shows an example of feedforward controlapplied to the heat exchanger temperature controlloop. The load disturbance, caused by a change infeed flow, is measured and added directly to the exittemperature controller output. This allows anincrease in feed flow to provide an immediateincrease in steam flow. With appropriate signal

scaling and dynamic compensation, it is theoreticallypossible to match the change in valve position to thechange in feed flow so that there is no effectwhatsoever on the exit temperature.

Figure 2 Single-Loop Response to Feed Flow Change

It should be noted that feedforward is an open loopcontrol strategy. The change in the manipulatedvariable (steam flow) has no effect on the loadvariable (feed flow). Therefore, it is not possible for afeedforward loop to become unstable.

Steam

Feed

TT

T

HeatExchanger

Condensate

TC

ExitTemperature

FT

+

+ For feedforward to be effective, it must be configuredso that the change in load moves the manipulatedvariable in the right direction, the right amount, and atthe right time. The right direction can be deducedfrom an understanding of the process. The rightamount and the timing may be estimated fromprocess knowledge, but is more often determined byempirical testing of the control loop and its response.

If feedforward adds too much steam for an increasein feed flow, the exit temperature will increase, andthere will be a control error in the opposite direction

from that caused by the load change. The controlconfiguration must include a gain adjustment so thatthe magnitude of the change in the manipulatedvariable can be adjusted relative to the change in theload variable.

Figure 3 Heat Exchanger Steady StateFeedforward Control

A similar problem occurs if the change in steam flowis too soon or too late. Since exit temperature willrespond at different rates to a change in feed flowand a change in steam flow, it may be necessary toslow down (or speed up) the response of the

2


3/12

AD353-129

manipulated variable to a change in feed flow. Thecontrol configuration must include dynamiccompensation to synchronize the response of themanipulated variable with the response of the loadvariable. This is accomplished by adding lead or lag,and perhaps dead time, to the feedforward signal.

Although feedforward control provides significant

benefits, it is not practical to control a process byfeedforward alone. Feedforward can onlycompensate for measured load variables. It is notpractical to measure every conceivable load variablethat can disturb a process. In addition, it is notpossible to provide perfect feedforward compensationfor the load variables that are being measured. Eventhe most sophisticated feedforward control scheme isnot capable of controlling the process variable exactlyat setpoint. Therefore, it is always necessary to usefeedback control in conjunction with feedforwardcontrol to trim or bias the feedforward model.

For feedforward control, consider one of thetechniques below, which are discussed in detail in theremainder of this publication.

Steady State Feedforward

Impulse Feedforward

Model Predictive Feedforward

STEADY STATE FEEDFORWARD

The steady state feedforward configuration is shownin Figures 4 and 5. The PID function block in loop 1manipulates its output to control the exit temperature

at the setpoint value from the SETPT block.

The feed flow signal is the load variable forimplementing feedforward control. In this applicationexample the feed flow signal is acquired usingEthernet I/O and brought into the 353 (loop 2)through the AIE1 function block. The lead/lag block(LL1) provides adjustable dynamic compensation. Itcan be configured to provide lag or leadcompensation. A dead time block (DTM1) has alsobeen included for any dead time compensation thatmay be required (see AD353-127 for dead timecompensation techniques).

The proportional-only controller (PD) in loop 2provides signal scaling and bias for the feedforwardsignal. If the PD controller is configured for directaction, the output signal is determined by thefollowing equation:

MRSPPGCO += )(

where:CO is controller outputPG is proportional gainP is process variable (feed flow)S is setpoint from SETPT blockMR is manual reset

The values for S and MR are initialized whenever theloop is switched to the manual mode. The not auto(NA) status signal from the A/M transfer switch inloop 1 will force the SETPT block in loop 2 to trackthe feed flow signal and the manual reset of the PDblock to track the output of the A/M block. When theloop is returned to auto, the values of S and MR areheld at the values of their respective track variablesat the moment the transfer is made. Since P=S andMR equals the valve signal, the output of the PDblock is initialized at the current valve output. Anysubsequent change in the feed flow will change theoutput of the PD block by an amount equal to PG(P

S).

Math block MTH1 in loop 1 adds the feedforwardcomponent from the PD block to the output of the PIDexit temperature controller. Except as filtered by thelag block (LL1), this provides an immediate change inthe valve signal for changes in feed flow. Thecharacterizer block (CHR1) is available tocompensate for any nonlinearity in the relationshipbetween valve position and steam flow to the heatexchanger.

Notice that MTH1 is configured with a 50% bias

value. This allows the PID controller output to trimthe feedforward component in both a positive andnegative direction. When the PID block output is atmidscale (50%), the controller output signal iscancelled by the bias and the feedforward componentis passed through without bias. However, thecontroller can adjust the valve loading signal up ordown by varying its output above or below midscalerespectively.

The A/M transfer block in loop 1 allows the operatorto switch between auto and manual modes ofoperation. To provide bumpless transfer from

manual to auto, the PID controller must be initializedat a value that will allow the MTH1 function block totrack the output of the A/M block in the manual mode.This is accomplished by back calculating theappropriate reset feedback in math block MTH2. Inmanual, the PID controller tracks the feedback signalprovided by the MTH2 block and the PD controller

3


4/12

AD353-129

Figure 4 Exit Temperature Controller w ith Steady State Feedforward Control (CF353-129SS)

Figure 5 Steady State Feedforward Computation for Exit Temperature Contro ller (CF353-129SS)

4


5/12

AD353-129

tracks the output of the A/M block. This forces theoutput of the PID block to midscale (50%); whichforces the output of the MTH1 block to match theoutput of the A/M block.

An example of the response to a feed flow change forthis example is shown in Figure 6. Compared withthe single loop only response in Figure 2, the

feedforward compensation has reduced the effect ofa feed flow load disturbance significantly. Dependingon the dynamic interaction between the feed forwardvariable (feed flow) and the controlled variable (exittemperature), the amount of improvement may belimited. The tuning of the feedforward compensationis also critical to the performance of the feedforwardcompensation.

Tuning

In a feedforward loop, the primary feedback controllermust be tuned before tuning the feedforwardcompensation. To minimize the effect of feedforwardwhile tuning the primary controller, set the lag block(LL1) time constant (TLAG) and the PD controllergain (PG) to their minimum values. The feedbackcontroller can then be tuned using any of the

standard methods including AUTOTUNE in a 353.

Next, the feedforward gain and time constant aretuned for the best feedforward compensation. Withexact compensation, a change in feed flow wouldhave little or no effect at all on the exit temperature.However, exact compensation is often not possiblebecause a simple lead or lag response cannotcompensate for the more complex process dynamicsnormally encountered. Therefore, the objective is to

minimize the change in the controlled variable for achange in the load variable.

For tuning the feedforward response, it is mostconvenient if the load can be adjusted to change thefeedforward variable. When this is not possible, it isnecessary to observe and, preferably, record theeffect of changes as they occur. A record of the

valve position is particularly useful for tuning.

Figure 7 shows four feedforward tuning examplesand the resulting valve positions. In all cases, a stepchange occurred at the instant the valve plotundergoes a step change upward. The horizontaldashed line represents the new steady state valveposition that returns the controlled variable to

setpoint. Dynamic compensation has beenminimized for these examples.

Exit Temperature

Steam Flow

Feed Flow

PG = 0.58, TLAG = 0.01 min, TLEAD = 0.18 min.

Figure 6 Response to a Feed Flow Change with Feedforward

Example A shows the effect of too much feedforwardgain. Feedforward moves the valve too far initially,and the feedback controller must reduce its output tocompensate for the over-correction. The only waythe controller can accomplish this is to integrate acontrol error caused by the initial overcorrection.

Example B shows the effect of too little feedforwardgain. Feedforward does not move the valve farenough initially, and the feedback controller mustincrease its output to compensate for the under-correction. Again, integrating a control error isrequired.

In example C, the feedforward gain is set correctly.The initial valve position achieved by feedforward isthe same position that satisfies the feedback

5


6/12

AD353-129

controller at steady state. However, the transientincrease in valve position after the initial stepindicates that the controlled variable did not remain atsetpoint initially. Feedforward was too late to preventthe transient control error. Clearly, feedforwardcannot be expected to respond before the loadchange occurs. However, lead compensationprovides an initial kick to minimize the effect of being

late.

In example D, the feedforward gain is also setcorrectly. However, in this case, the transientdecrease in valve position indicates that thefeedforward response was too early. Lagcompensation can minimize the effect of being tooearly.

Typically, the feedforward gain (PG) should beadjusted to the correct value before adjusting thelead or lag settings. Then, find the dynamiccompensation that minimizes the transient effect.

The dynamic compensation will rarely be perfect dueto the complex nature of the process dynamics.

Comparison with Other Methods

Steady state feedforward is relatively simple toimplement and tune. However, it is important torealize that this method of feedforward control can

prevent the feedback controller from stroking thevalve completely. If, for example, the feedforwardcomponent is contributing more than 50% to thevalve signal, the feedback controller will not be ableto bias the feedforward sufficiently in the negative toclose an air-to-open valve. This does not normallyoccur unless the feedforward component is badlymisadjusted.

If a sustained control offset develops due to thebiasing problem described above, momentarily switchthe loop to manual to re-initialize the controllers andthen switch the loop back to auto to resume control.

Figure 7 Feedforward Tuning Examples

6


7/12

AD353-129

IMPULSE FEEDFORWARD

Impulse feedforward derives its name from the factthat the feedforward (load) variable generates animpulse that provides a transient bias to the controlleroutput. The valve loading signal is the sum of thePID controller output and the feedforward impulse.This feedforward technique is shown in Figure 8.

Steam

Feed

TT

T

HeatExchanger

Condensate

TC

ExitTemperature

FT

+

+IMP

ResetFeedback

Figure 8 Impulse Feedforw ard

Figure 9 shows the combination of a deviationamplifier (DAM_) block and a lag (LL_) block that isused to implement the impulse feedforward. Theload variable is connected to the A and B inputs (Bvia the lag block), and the controller output isconnected to the C input. At steady state, A and Bcancel one another, and the output of the deviationamplifier is simply the value of the C input (controller

output).

For a step change in the load variable, there is animmediate change (impulse) in the output of thedeviation amplifier. However, as the lag responds tothe step change, the A input is increasingly cancelledby the B input, until at steady state the load variablecontributes nothing to the output of the deviationamplifier. The initial impulse decays to zero, and theoutput of the deviation amplifier returns to the valueof the controller output (input C).

An impulse that decays to zero does not provide the

steady state change in the manipulated variable thatis necessary to compensate for the load disturbance.However, as shown in Figure 10, the output of thedeviation amplifier provides external reset feedbackto the controller. This allows the impulse to drive thereset (integral) component and force the controlleroutput to a new steady state operating value. If thelag time (TLAG) is set equal to the integral time, thereset component will complement the decayingimpulse. The sum of these two transients will providea sustained step change in the output of the deviation

Load

Variable

Controller

Output

Lag

DEVAMP

A

B

C

GAIN(A-B) + C0

100%

t

L

TL

LGx

DEV AMP Output

Load Variable

Figure 9 Impulse Function

7


8/12

AD353-129

Lag

DEVAMP

A

B

C

L

TL

LGx

Load Variable

LoadVariable

Lag

DEVAMP

A

B

C

G(A-B) + C0

100%

t

L

TL

PIDFB

Impulse

Reset

DEV AMP Output

amplifier. This achieves a steady state change invalve loading without integrating a control error oradding a steady state bias.

Impulse feedforward can also provide dynamiccompensation by setting the lag time (TLAG) to beless than or greater than the integral setting (TI) ofthe controller. For lead response, TLAG must be lessthan TI, and for lag response, TLAG must be greaterthan TI.

Configuration

Figure 11 shows the configuration for impulsefeedforward. The PID controller manipulates itsoutput to control the exit temperature at setpoint.The feed flow signal is the load variable forimplementing feedforward control. This signal isapplied directly to one side of a deviation amplifier(DAM1) and indirectly through a lag block (LL1) to theother side of the deviation amplifier. If the applicationrequires an inverse relationship between the valvesignal and the load variable, these connections to thedeviation amplifier must be reversed. A change infeed flow generates an impulse as described above.

The A/M transfer switch allows the operator to switchbetween auto and manual modes of operation. Toprovide bumpless transfer from manual to auto, theauto status (AS) signal from the A/M block forces thePID block to track the valve signal and the lag block(LL1) to bypass the lag function.

If desired, a characterizer block can be inserted in thevalve signal, as shown in Figure 4 for the steadystate feedforward configuration. However, theexternal feedback for the PID block must tee offupstream of the characterizer block.

Figure 10 Impulse Plus Reset Feedback

Tuning

Impulse feedforward is tuned in much the same wayas steady state feedforward. The main difference isthe manner in which dynamic compensation isaccomplished when using impulse feedforward.

With the gain (GAIN) of the deviation amplifier(DAM1) and the time constant (TLAG) of the lagblock (LL1) set at minimum values, the feedbackcontroller (PID) can be tuned using any of thepreferred methods.

Before doing tests to determine the feedforward gain,the lag time constant should be set equal to theintegral time constant of the controller ( TLAG = TI).This adjusts the impulse for no lead or lag. Useexamples A and B in Figure 7 as a guide to find thecorrect feedforward gain setting.

With the correct feedforward gain, use examples Cand D in Figure 7 to determine the need for lead orlag action. If lead is required, decrease the lag timeconstant (TLAG) and if lag is required, increase thelag time constant.

8


9/12

AD353-129

Figure 11 Impulse Feedforward Conf iguration (CF353-129IM)

The steady state gain (G) of the impulse function isaffected by both the deviation amplifier gainadjustment (GAIN) and the ratio of the integral andlag time constants as follows:

)(TI

TLAGGAING =

To retain the effective feedforward gain for changesin TLAG, recalculate the gain of the deviationamplifier (GAIN) per the following equation:

)(TLAG

TIGGAIN =

Suppose, for example, that the feedforward gain wasset for 1.2 when TLAG = TI. If the lag time (TLAG) isreadjusted to 0.8TI to achieve a leading response,the GAIN of the deviation amplifier should bereadjusted to (1.2)/(0.8) or 1.5 to retain the originalfeedforward gain. It will be necessary to readjustGAIN for every value of TLAG that is tried whileconducting experiments to determine the optimumvalue of TLAG.


Impulse feedforward is accomplished with a simpleconfiguration. It also eliminates completely thepotential bias problem that can occur with steadystate feedforward.

The only disadvantage of impulse feedforward is thepotentially disconcerting interaction among the tuningconstants during the tuning process. The tuningprocedure is essentially the same in both cases, butthe interaction must be understood to avoidfrustration.

MODEL PREDICTIVE FEEDFORWARD

Model predictive feedforward uses a rigorousmathematical model of the process to calculate the

value of the manipulated variable as a function of oneor more load variables. The primary feedbackcontroller trims the model to compensate formodeling errors or unmeasured load disturbances.

As shown in Figure 12, both the feed flow and thefeed temperature are measured. The output of thetemperature controller (TC) represents the exittemperature. In this application, the heat required toraise the feed flow to the desired exit temperaturecan be calculated using the following equation.

TFcQ fp `=

where:Q is heat transfer rate [e.g. BTU/hr]cp is specific heat [e.g. BTU/(lbF)]Ff is feed flow (mass) [e.g. lb/hr]T is change in temperature [e.g.F]

9


10/12

AD353-129

The heat supplied by the steam can be calculatedusing the following equation:

SFQ =

Where:Q is heat transfer rate [e.g. BTU/hr]

is the latent heat of steam [e.g. BTU/lb]Fs is steam flow (mass) [lb/hr]

This equation is used to calculate the value of themanipulated variable in a model predictivefeedforward configuration. For other applications, itwill be necessary to use whichever first principles(heat balance, material balance, etc.) are appropriateto derive a rigorous mathematical model.

Steam

Feed

TT

T

HeatExchanger

Condensate

TC

ExitTemperature

FT

_

+XTFf T

TT

Figure 12 Model Predictive Feedforward

Configuration

The configuration for model predictive feedforward isshown in Figure 13. The math block (MTH1) is usedto calculate the steam flow as a function of feed flow,feed temperature, and exit temperature setpoint.Note that the output of the exit temperature controlleris used instead of the actual exit temperature setpointas a convenient method of providing feedback trim tothe feedforward calculation. Otherwise, another mathblock would be required to implement feedback trim.

In addition, if both the setpoint and the controlleroutput are used in the feedforward calculation, therewill be a double correction whenever the setpoint ischanged. This configuration avoids that problem. Ifthe model provides perfect steady state feedforwardcompensation, the temperature controller will besatisfied when the output equals the desired exittemperature. Otherwise the PID block will bias the

T calculation as required to drive the temperature tosetpoint.

The equation derived above can be implementedusing a single configurable math block (MTH_). Alead/lag block (LL_) provides dynamic lead or lagcompensation. A dead time block (DTM_) providesany dead time compensation that may be required.

Although all load variables may require dynamiccompensation, it is usually more critical on loadvariables that move more quickly such as flow.

The A/M transfer switch allows the operator to switch

between auto and manual modes of operation. Toprovide bumpless transfer from manual to auto, thePID block must be initialized at a value that will allowthe math block (MTH1) to track the output of the A/Mblock in the manual mode. This is accomplished byback calculating the appropriate reset feedback inmath block MTH2. In manual, the PID block tracksthe feedback signal provide by MTH2 and the ASstatus from the A/M block bypasses the dynamicfunctions of the LL_ and DTM_ blocks.

If desired, a characterizer block can be inserted in thevalve signal, as shown in Figure 4 for the steady

state feedforward configuration. However, input A toMTH2 must tee off upstream of the characterizerblock to calculate the correct external feedback forthe PID block. Alternatively, if a steam flowmeasurement is available, this configuration can bemodified to provide a setpoint to a secondary steamflow controller in cascade. See AD353-128 for moreinformation on Cascade Control.

The Siemens 353 controller uses real (floating point)numbers between blocks. This enables engineeringunits to be used for block interconnection signals.The PID block output is scaled to represent the exittemperature with a range of 0 to 200F. The signalout of the MTH1 block represents steam flow. Scalerblock SCL1 provides a steam flow range to theanalog output block so it can convert the steam flowrange of 0 to 6000 lbs/hr to 0 to 100% valve signal.The MTH2 block output signal represents exittemperature for reset feedback to the PID block. Thecalculations for the MTH_ blocks were derived asfollows:

10


11/12

AD353-129

Figure 13 Model Predictive Feedforward Conf iguration (CF353-129MP)

MTH1 ))(60()( fefPs TTFCF =

MTH2 ff

s

p

e TF

F

CT += ))((

60

Ff (Feed Flow) ................ 0 to 400 GPMTf (Feed Temp) ............... 0 to 200 FTe (Exit Temp).................0 to 200 F

Fs (Steam Flow) .............. 0 to 6000 lbs/hrCp (specific heat) ............ 0.8 BTU/lb (feed) F(density)........................8.25 lb (feed)/ gallon(latent heat of steam)....853.5 BTU/lb of steam

Tuning

With model predictive feedforward, the mathematicalmodel establishes the relationship between themanipulated variable and the load variables.Therefore, there is no feedforward gain adjustment,except that which is already built into the model. It isnecessary to determine only the type of dynamic

compensation required and to tune it in much thesame manner as described for steady statefeedforward.


Model predictive feedforward is the mostsophisticated of the three methods described and hasthe potential to provide the best performance. It cancompensate for more than one load variable and isbased on fundamental knowledge of the process.

This method requires more design effort than eitherthe steady state or impulse feedforward methods.However, the effort required will usually be rewardedby significant improvement in control loopperformance.

APPLICATIONSThe feedforward control techniques described in thispublication are applicable to any process control loopwith measurable load disturbances. Examples areboiler drum level control, composition control indistillation columns, and pH control in waste watertreatment.

Appl ication Support

User manuals for controllers and transmitters,addresses of Siemens sales representatives, andmore application data sheets can be found atwww.usa.siemens.com/ia. To reach the processcontroller page, click Process Instrumentationandthen Process Controllers and Recorders. To selectthe type of assistance desired, click Support(in the

right-hand column). See AD353-138 for a list ofApplication Data sheets.

The configuration(s) shown in this publication werecreated in Siemens i|configGraphicalConfiguration Utility. Those with CF353 inparenthesis in the Figure title are available using theabove navigation, then click Software Downloads>353 Feedforward Control (Reference AD353-129).

11
http://www.usa.siemens.com/iahttp://www.usa.siemens.com/ia


12/12

AD353-129

The configuration(s) can be created and run in a:

Model 353 Process Automation Controller

Model 353R Rack Mount Process AutomationController*

i|pacInternet Control System*

Model 352PlusSingle-Loop Digital Controller** Discontinued model

i|pac, i|config, Procidia, and 352Plus are trademarks of Siemens Industry, Inc. Other trademarks are the property of their respective owners.All product designations may be trademarks or product names of Siemens Industry, Inc. or other supplier companies whose use by thirdparties for their own purposes could violate the rights of the owners.

Siemens Industry, Inc. assumes no liability for errors or omissions in this document or for the application and use of information in thisdocument. The information herein is subject to change without notice.

Siemens Industry, Inc. is not responsible for changes to product functionality after the publication of this document. Customers are urged toconsult with a Siemens Industry, Inc. sales representative to confirm the applicability of the information in this document to the product theypurchased.

Control circuits are provided only to assist customers in developing individual applications. Before implementing any control circuit, it shouldbe thoroughly tested under all process conditions.

Copyright 2012, Siemens Industry, Inc.

12

forward siemens control

Documents