06 model robustness cts

84
7/21/2019 06 Model Robustness CTS http://slidepdf.com/reader/full/06-model-robustness-cts 1/84 February 21, 2016 Reliance Technology Group – APC / RTO  For  Internal Circulation 1 Process Engineering Group Process Engineering Group Model Robustness Model Robustness Rajalingam R Rajalingam R

Upload: sreekanthmylavarapu

Post on 08-Mar-2016

216 views

Category:

Documents


0 download

DESCRIPTION

d

TRANSCRIPT

Page 1: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 1/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation1

Process Engineering GroupProcess Engineering Group

Model RobustnessModel Robustness

Rajalingam RRajalingam R

Page 2: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 2/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation2

1) Importance of Model Robustness

2) Tools for making models robust

) Model Identification ! "M#

FIR Identification

$ubspace identification

%) Model Robustness

Scope

Page 3: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 3/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation

&'ample

Condition No: 23957 - ILL Conditioned

Condition No: 35 - !ell Conditioned

Solution"#uation

Impo$tance o% Model Robustness

Page 4: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 4/84

Model Robustness

February 21, 2016 Reliance Technology Group – APC / RTO For Internal Circulation%

  Model Robustness is important from mat(ematical stability

* Model Identification+ Finite Impulse Response FIR) and $ubspace

* FIR is Multiple input and $ingle -utput Model MI$-)

* $ubspace is Multiple input Multiple -utput Model MIM-)

* Model .ncertainty+* Time "omain and Fre/uency "omain nalysis

* #orrelations+

*  uto correlation #ross correlation analysis

* #ondition number and R+

* #ondition 3umber * Relati4e ain analysis

Page 5: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 5/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation5

Model IdentifcationModel Identifcation

Page 6: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 6/84

Page 7: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 7/84

Linea$ Models

February 21, 2016 Reliance Technology Group – APC / RTO For Internal Circulation8

$tructure &mpirical9data!based First principles

Mat(ematicalform

:inear9non!linear :inear9non!linear 

"'amples o% Linea$ models+

 $tatic pressure ;it( respect to dept( of li/uid

 <olume of a 4ertical cylinder ;it( respect to le4el

"'amples o% Non-linea$ models:

In distillation column, t(e c(ange in purity ;it( Reboiler "uty ;ill be more ;(en

Reflu' Ratio is small= and 4ice 4ersa

Linea$i(ation

If non!linear, t(en use a suitable transform like :- transform) to make t(erelation linear

Page 8: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 8/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>

)ene%its o% Linea$ models

Page 9: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 9/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation?

)ene%its o% Linea$ models

Page 10: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 10/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation10

)ene%its o% Linea$ models

Page 11: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 11/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation11

)ene%its o% Linea$ models

 @I1 *+ I+,+

I2,+*+

 @I2

 1

C. C+ C2 C3 C/ C5 C0 C7

+/

70 7 7 7

 2+

253 5 5 5

2

90

+2

#<

+2 +2 +2

Page 12: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 12/84

Identi%ication

δ #<  A B ∆I, IAM<

Cno;n

Calculate

∆∆

∆∆

=

∂∂∂∂∂∂∂

%

2

1

%%%%

%%%

2%%

12%

12

12

1

8

6

5

%

2

1

I

I

I

I

aaaa

aaaa

aaaa

aaaa

aaa

aa

a

#<

#<

#<

#<

#<

#<

#<

*

Cno;n

Page 13: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 13/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation1

Response to a c(ange in t(e independent+

#<1 ! #<0  A a1 B DI)

#<2 ! #<0  A a2 B DI)

#< ! #<0  A a B DI)#<% ! #<0  A a% B DI)

#<5 ! #<0  A a5 B DI)

#<6 ! #<0  A a5 B DI)

DIInd

a1a2

aa% a5

Ereser4ation of $cale

#<1

#<2

#<

#<% #<5

#<0

Time 0 1 2 % 5

"ep

Identi%ication

Page 14: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 14/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation1%

0 2 % 6 >

Ind 1!2

1

%%

!1

!5

!8!8

!1%

86 8 8 8

!>

!2

!12

!1%

#<

!8,!2)B8,0)B8,1)B8#<#<

!8,!2)B8,0)B8,1)B8#<#<

!5,!2)B6,0)B8,1)B8#<#<

!1,!2)B%,0)B6,1)B8#<#<

%,!2)1B,0)B%,1)B6#<#<

%,0)1B,1)B%#<#<

1,1)1B#<#<

08

06

05

0%

0

02

01

=++=−

=++=−

=++=−

=++=−

=++=−=+=−

==−

Identi%ication

Time 0 1 2 % 5 6 8

Page 15: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 15/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation15

Finite Impulse Response FIR) Identification

 ny impulse can be represented like t(is+ G1,0,0,H0

Page 16: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 16/84February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation16

Finite Impulse Response FIR) Identification

-utput 4ector can be ;ritten as a combination of inputs 4ector+

yJtK A (0 'JtK (1 'Jt!1K (2 'Jt!K HHHH (k!1 'Jt!k!1)K (k 'Jt!kK

 * 'JtK L c(ange in independent 4ariable* yJtK L c(ange in dependent 4ariable* k A Time to $teady $tate

* t A time

We usually assume that h0= 0, i.e., the system

 does not react immediately to the input.

3o;, by least s/uare obecti4e function, ;e can find

out t(e step response coefficients G(0,(1,(2,H,(k

where N = total number of samples in a dataset

Page 17: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 17/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation18

FIR I"H$tep Response #oefficients

Page 18: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 18/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation1>

$tep!Response coefficients

FIR I"Hstep response coefficients

Impulse $esponse coe%%icients :

Step $esponse coe%%icients +

 ny step response can be represented as superposition of 4arious impulses+

G1,1,1H,1 A G1,0,0,H,0 G0,1,0,H,0 G0,0,1H,0 H G 0,0,0H,1

  G0,(1,(2,(,H,0 G0,0,(1,(2,(H,0 G0,0,0,(1,(2,(H,0 H G 0,0,0H,(n

 

Page 19: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 19/84

FIR Identification+ T(e Matri'

Steps to dete$mine t1e unit step $esponses

4u$ing a test pe$iod, collect time-stamped p$ocess data

%$om t1e sstem %o$ all independent and dependent

6a$iables

&$om t1e collected data, calculate C and I

Sol6e t1e cont$ol e#uation, C 8 I, %o$ t1e unit step

$esponse coe%%icients, t1e mat$i'

Page 20: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 20/84

Erocess reac(es steady!state

after % inter4als $o, t(ere are

only % NaO coefficients

FIR Identification

"#uations %o$ p$edicting t1e linea$ sstem;

0B)IIaB)IIaB)IIaB)IIaB)II#<#<

aB)IIaB)IIaB)IIaB)II#<#<

aB)IIaB)IIaB)II#<#<

aB)IIaB)II#<#<

aB)II#<#<

01%1222%1%505

%0112221%0%

01212120

20111202

10101

−+−+−+−+−=−−+−+−+−=−

−+−+−=−−+−=−

−=−

Page 21: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 21/84

FIR Identification+ N$tep!ResponseO Form

Step-Response &o$m o% t1e Identi%ication <$oblem

%1222%1%505

%0112221%0%

01212120

20111202

10101

aB)IIaB)IIaB)IIaB)II#<#<

aB)IIaB)IIaB)IIaB)II#<#<

aB)IIaB)IIaB)II#<#<

aB)IIaB)II#<#<aB)II#<#<

−+−+−+−=−−+−+−+−=−

−+−+−=−−+−=−

−=−

∆∆

=

∂∂

%

2

1

%%%%

%%%

2%%

12%

12

12

1

8

6

5

%

2

1

I

I

II

aaaa

aaaa

aaaa

aaaa

aaa

aa

a

#<

#<

#<

#<

#<

#<

#<

*

∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆

∆∆∆

∆∆∆

=

∂∂∂∂∂

∂∂

%

2

1

%568

%56

2%5

12%

12

12

1

8

6

5

%

2

1

IIII

IIII

IIII

IIII

III

II

I

#<

#<

#<

#<

#<

#<

#<

a

a

aa

*

#omplete e/uations

useful)

Page 22: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 22/84

FIR Identification L Pad "ata

  period of bad data s(ould not in4alidate t(e entire data set

In t(e step!form of t(e algorit(m

#<0 appears in eac( e/uation

-nly one continuous section of data can be analyQed using t(is algorit(m L step

response algorit(m

Pad data ;it(in a data set in4alidates t(e data set

 L Erocess problem9upset

 L #omputer problem

 L ard;are9instrumentation failure

 L <al4e saturation

 L  typical disturbance

-t(er forms of t(e algorit(m L eliminate #<07

Page 23: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 23/84

FIR Identification L Impulse Form 1)

0)))

)))

))

)

+−−+−−+−−+−=−

−−−−−−−−=+

−+−+−+−+−=−

−−+−−+−−+−=−

−−−−−−=+

−+−+−+−=−

−−+−−+−=−−−−−=+

−+−+−=−

−+−=−

−−=−

−+−=−

−=−

%122212%1%5%5

%0112221%0%

%01%1222%1%505

%012121221%%

01212120

%0112221%0%

2011212122

20111202

01212120

120111212

10101

20111202

10101

a,aB)I,Ia,aB)I,Ia,aB)I,IaB)I,I#<#<

aB)I,IaB)I,IaB)I,IaB)I,I#<#<!

aB)I,IaB)I,IaB)I,IaB)I,IaB)I,I#<#<

a,aB)I,Ia,aB)I,Ia,aB)I,IaB)I,I#<#<

aB)I,IaB)I,IaB)I,I#<#<!

aB)I,IaB)I,IaB)I,IaB)I,I#<#<

a,aB)I,Ia,aB)I,IaB)I,I#<#<aB)I,IaB)I,I#<#<!

aB)I,IaB)I,IaB)I,I#<#<

a!,aB)I,IaB)I,I#<#<

aB)I,I#<#<!

aB)I,IaB)I,I#<#<

aB)I,I#<#<

$ubtract pairs of

e/uations to

remo4e #<0

Page 24: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 24/84

FIR Identification L Impulse Form 2)

"efine NbiO as t(e impulse coefficient+ ai ! ai!1

3o;, t(e Nimpulse!formO of t(e "M#plus Model e/uations

%1222%1%5%5

%0112221%%

01212122

20111212

10101

bB)IIbB)IIbB)IIbB)II#<#<

bB)IIbB)IIbB)IIbB)II#<#<

bB)IIbB)IIbB)II#<#<

bB)IIbB)II#<#<

bB)II#<#<

−+−+−+−=−−+−+−+−=−

−+−+−=−

−+−=−−=−

%%

2

122

11

aab

aab

aab

ab

−=

−=−=

=

Page 25: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 25/84

FIR Identification+ Impulse Response Model

#(aracteristics

Penefit+

"ata $licing is llo;ed+ #(ange in t(e dependent 4ariable is only a

function of t(e past c(anges in independent 4ariables for a time e/ual

to t(e time to steady!state

Eenalty+ dditional noise results from taking t(e deri4ati4e of t(e dependent

4ariable

Page 26: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 26/84

FIR Identification+

$imultaneous Independent <ariable Mo4es

In a typical process, it is generally impossible to N(oldO an independent step

for a full time to steady!state ;it(out ot(er mo4ement in ot(er independent

4ariables

-perations needs to make a mo4e

Feedfor;ard, disturbance, 4ariable mo4es

  simple, straig(t!for;ard solution to t(e control problem ;ill not (andle

t(ese additional mo4esH

e need a solution met(od t(at is tolerant of multiple mo4es ;it(in a single

Tss

Page 27: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 27/84

FIR Identification+

$imultaneous Identification of Models

Multiple Input, $ingle -utput MI$-) $imultaneousidentification

of model

coefficients

∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆

∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆

∆∆∆∆∆∆∆∆∆∆ ∆∆

=

∆∆∆∆

∆∆∆∆∆

2,%

2,

2,2

2,1

1,%

1,

1,2

1,1

2,62,82,>2,?1,61,81,>1,?

2,52,62,82,>1,51,61,81,>

2,%2,52,62,81,%1,51,61,8

2,2,%2,52,61,1,%1,51,6

2,22,2,%2,51,21,1,%1,5

2,12,22,2,%1,11,21,1,%

2,12,22,1,11,21,

2,12,21,11,2

2,11,1

?

>

8

6

5

%

2

1

IIIIIIII

IIIIIIII

IIIIIIIIIIIIIIII

IIIIIIII

IIIIIIII

IIIIII

IIIIII

#<

#<

#<#<

#<

#<

#<

#<#<

b

b

b

b

b

b

b

b

*

Page 28: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 28/84

FIR Identification+

$imultaneous Identification of Models

Page 29: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 29/84

First, letSs con4ert t(e control e/uation to a

residual form+ C - I 8 R

FIR Identification+ $imultaneous Identification

=

∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆∆

∆∆∆∆∆∆∆∆∆∆

∆∆

∆∆∆∆∆∆∆∆

!

"

#

$

%

&

'

(

*

b

b

b

b

b

b

bb

2,%

2,

2,2

2,1

1,%

1,

1,2

1,1

2,62,82,>2,?1,61,81,>1,?

2,52,62,82,>1,51,61,81,>

2,%2,52,62,81,%1,51,61,8

2,2,%2,52,61,1,%1,51,6

2,22,2,%2,51,21,1,%1,5

2,12,22,2,%1,11,21,1,%

2,12,22,1,11,21,

2,12,21,11,2

2,11,1

?

>

8

6

5

%

2

1

IIIIIIII

IIIIIIII

IIIIIIII

IIIIIIII

IIIIIIII

IIIIIIII

IIIIII

IIII

II

#<

#<

#<

#<

#<

#<

#<

#<

#<

Page 30: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 30/84

FIR Identification+ $imultaneous Identification L $olution

3o;, ;e can use a N:east $/uaresO regression met(od to sol4e fort(e response coefficients ;(ile minimiQing NR2O

#alculate t(e sum of t(e s/uared residual terms+

Residuals$/uaredof $um=

=

++++++++=

∑   '

'

'

!

'

"

'

#

'

$

'

%

'

&

'

'

'

(

i

r r r r r r r r r r r 

Page 31: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 31/84

FIR Identification+ $imultaneous Identification L

$ummary

Penefit

$imultaneous solution allo;s for c(anges in more t(an one

independent at a time during t(e test

Implication9Eenalty

Identifying t(e indi4idual responses re/uires uncorrelated mo4ement in

t(e independent 4ariables during t(e plant test

Models are identified simultaneously so c(anging t(e list of

independents may c(ange t(e model coefficients

Page 32: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 32/84

FIR Identification "esign #riteria

  period of bad data s(ould not in4alidate t(e entire

data set

$ol4e t(e impulse!form rat(er t(an t(e step!form of t(e

algorit(m

More t(an one independent 4ariable s(ould be allo;ed

to c(ange at t(e same time

$ol4e for all independent9#< response coefficients

simultaneously as a least!s/uares MI$- problem

Page 33: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 33/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation

FIR parameters

+ =ime to stead state ==SS

 2 Numbe$ o% coe%%icients

3 Smoot1ing %acto$ 

Page 34: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 34/84

FIR Identification Earameters

Erocess settling time steady!state time)

:engt( of t(e step response models in time

3umber of coefficients

More coefficients re/uired for accurate modeling of fast responses less time

bet;een coefficients)

$moot(ing

Eenalty for c(ange from one coefficient to ne't

Model fit is not significantly degraded

Page 35: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 35/84

Puild t(e #ontroller Model+ 3umber of Model #oefficients

T(e number of model coefficients, or t(e number of points used to Ndra;O t(e

line is related to t(e controller e'ecution fre/uency and t(e Tss+

#ontrol inter4al needs to be fast enoug( so t(at t(e controller can respond to

t(e fastest measured9 unmeasured independent disturbances in t(e system

Ideally, t(e controller is running fast enoug( t(at it (as 5 to 10 e'ecutions to (andle

a disturbance before it becomes a problem

   

 

 

 

 −

= Interval 

Collection

 Data

tsCoefficien Model of  Number 

StateSteadytoTime

 Interval 

 ExecutionController *

Page 36: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 36/84

Puild t(e #ontroller Model+ 3umber of Model #oefficients

f # ff

Page 37: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 37/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation8

FIR parametersH3umber of #oefficients

It decides o4ersampling ratio in model identifications

$ampling time A $ettling time 9 $tep!response coefficients

FIR 3 b f # ffi i

Page 38: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 38/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>

FIR parametersH3umber of #oefficients

-4ersampling Ratio A 1 -4ersampling Ratio A 2

3otice gain mismatc( in FIR $$ FIR and $$ results no; matc( better 

FIR t $ t(i F t

Page 39: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 39/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation?

FIR parametersH$moot(ing Factor 

FIR smoot(ing algorit(m ;ill ad4ersely affect model responses ;it( faster initial

dynamics, like 4al4e positions e can use un!smoot(ed response models

instead, and use cur4e operations to remo4e noise and s/uiggles from t(e model

FIR t $ t(i F t

Page 40: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 40/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%0

FIR parametersH$moot(ing Factor 

$FA 5 $FA 001

Ti " i (

Page 41: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 41/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%1

Time "omain pproac(

FI I is time domain approach. Consider the followin+ euations-

i/en the compleity of differential euations, why would we e/er want to wor1 in the time

domain2

3he 4aplace transform mo/es us out of the time5domain into the comple freuency domain, so

that

we can study and manipulate our systems as al+ebraic polynomials instead of linear 67s.8( 9 's 9 $s'):8s) = 8( 9 s);8s)

3hat<s ri+ht, the 4aplace transform is hidin+ the fact that we are actually dealin+ with second5order

differential euations. In the 4aplace or time domain, if we want to account for systems with

multiple inputs and multiple outputs, we are +oin+ to need to rely on the principle of

superposition to create a system of simultaneous 4aplace euations 8or time5domain based

euations) for each output and each input. For such systems, the classical approach doesn<tsimplify the situation in MIMO case.

FIR models are $ingle input single output models $I$-) and can be e'tended for

Multiple input single output MI$-) $o, a case containing multiple #<s is run as

combination of MI$- models for eac( #<

$t t (

Page 42: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 42/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%2

$tate!space pproac(

It turns out t(at if ;e decompose our (ig(er!order differential e/uations into multiple first!order

e/uations, ;e can find a ne; met(od for easily manipulating t(e system in MIM> models T(e

solution to t(is problem is state space app$oac1

State

#entral to t(e state!space notation is t(e idea of a state state of a system is t(e current 4alue

of internal elements of t(e system, t(at c(ange separately to but not completely unrelated to)

t(e output of t(e system ere are some e'amples+

*Consider a chemical reaction where certain reagents are poured into a mixing container, and the output is the amount of the

chemical product produced over time. The state variables may represent the amounts of un-reacted chemicals in the container,

or other properties such as the quantity of thermal energy in the container (that can serve to facilitate the reaction).

e denote t(e input 4ariables ;it( u, t(e output 4ariables ;it( y, and t(e state 4ariables ;it( ' In essence,

;e (a4e t(e follo;ing relations(ip+

y A f', u)

(ere f', u) is our system lso, t(e state 4ariables can c(ange ;it( respect to t(e current state and t(e

system input+

' A g', u)

(ere ' is t(e rate of c(ange of t(e state 4ariables

$tate "efinition

Page 43: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 43/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%

$tate "efinition

$tate 4ariables A memory elements A contains all t(e information from t(e

past t(at is rele4ant for predicting t(e future

If ;e describe system as an operating mapping from t(e space of input to

t(e space of output, t(en ;e may need t(e entire input!output (istory of

t(e systems toget(er ;it( t(e planned input in order to compute future

output 4alues lternati4ely, ;e may use state 4ariables ;(ic( (as all

(istory of input and -utput 4ariables

-rder of model A number of states n)

#< t i i $ b id tifi ti

Page 44: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 44/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%%

#<s uto!grouping in $ubspace identification

$ubspace is MIM- model identification= but ;(en number of unrelated #<s are large in

case, t(en it increases computation time, and benefits ;ere limited compared to running

multiple smaller cases

T(e subspace identification algorit(m can optimiQe case runs by using #< grouping T(e

#< cross correlation is used to determine groups of #< t(at s(are common dynamics,

and create internal sub!case for t(ose related #<s MIM- for eac( internal sub case)

 

Py keeping related #<s toget(er t(e underlying states ;ill be identified ;it( more

certainty Py remo4ing unrelated #<s t(e computation time per subgroup is decreased

#< grouping alternati4es

1)roup Related #<s uto!grouping, "efault) L ma'imum of 10 #<s for one sub!case

2)-ne #< per roup Forced MI$- I") L ;it( t(is option, t(e users are able to compare

t(e MI$- subspace I" ;it( t(e FIR I" in a similar internal setup

)ll #<s in -ne roup Forced :arge MIM- I") L set as one large MIM-

%)$et Ma'imum #< roup $iQe MIM- I" ;it( certain number #<s per group) L gi4es

fle'ibility to modify t(e default ma'imum #<s of 10 per group

   llows users to gain!validate their "nowledge about the C#s correlation in their

 process, compare models with different setups, and retune the default parameters when

necessary.

#<s to gro ping in $ bspace identification

Page 45: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 45/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%5

#<s uto!grouping in $ubspace identification

Time to steady state in $ubspace identification

Page 46: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 46/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%6

Time to steady state in $ubspace identification

T(eoretically, TT$$ 4alue (a4e no direct impacts on t(e subspace I", because TT$$ is

not a dependent parameter in subspace I" and it only determines (o; many model cur4e

coefficients ;ill be generated from a subspace model identified

o;e4er, "etrending filters t(at ;ill be calculated upon TT$$ 4alues affect $ubspace I"

for different TT$$

Time to steady state in $ubspace identification

Page 47: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 47/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%8

Time to steady state in $ubspace identification

 ll parameters are t(e same

e'cept TT$$

Time to steady state in $ubspace identification

Page 48: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 48/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%>

Time to steady state in $ubspace identification

Ma'imum states per #< group+ $ubspace identification

Page 49: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 49/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation%?

Ma'imum states per #< group+ $ubspace identification

$tates do not necessarily (a4e a direct p(ysical interpretations, but t(ey (a4e a

conceptual rele4ance

Ma'imum states per #< group, ;(ic( is an upper bound of t(e model order ie

ma'imum number of states allo;ed) for eac( #< group sub!case

T(is parameters allo;s user to set a model order searc( range, or force t(e subspace

algorit(m to fit a lo;!order model ;it( truncation error 

(en to reduce t(e ma'imum states per #< group7

Ma'imum states per #< group+ $ubspace identification

Page 50: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 50/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation50

Ma'imum states per #< group+ $ubspace identification

Ma'imum states per #< group+ $ubspace identification

Page 51: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 51/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation51

Ma'imum states per #< group+ $ubspace identification

ig(!fre/uency dynamic be(a4ior of model is caused by a (ig( model order, and a model

reduction can (elp to damp t(e (ig( fre/uency noise

Ma'imum states per #< group+ $ubspace identification

Page 52: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 52/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation52

Ma'imum states per #< group+ $ubspace identification

Ma'imum order per I9- pair+ $ubspace identification

Page 53: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 53/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation5

Ma'imum order per I9- pair+ $ubspace identification

T(is parameter is t(e identification (oriQon used in t(e identification It is e'actly

t(e number of data points in t(e future and past data (oriQon T(e larger t(e

Ma'imum order per I9- pair, t(e (ig(er order model ;ill be identified in order toco4er longer comple') dynamics T(e computation ;ill also be (ea4ier

Ma'imum order per I9- pair+ $ubspace identification

Page 54: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 54/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation5%

Ma'imum order per I9- pair+ $ubspace identification

If 4ery different ma'imum order per I9- are used in subspace case, t(en e4en if t(e same

final model order is selected, t(e calculated models are bound to be different

T(e amount of data used for initialiQing t(e identification run depends on ma'imum order

not t(e final order it e4entually selects), so if you select a (ig( ma' order, more data is

lost If useful data is lost, eg t(e only big step you (a4e) t(en t(e model can degrade If

bad data is lost, t(en t(e model can actually impro4e $o as before ;it( "MI, be careful

;it( slicing t least you lose muc( less data t(an before

:ets assume ma' order A n T(en n pre4ious n data points are t(en used to predict t(ene't n steps into t(e future T(e cost function is determined from t(e difference bet;een

predicted #<s and obser4ed #<s o4er t(ese n steps T(is means t(at a (ig( ma' order

fa4ors t(e long term prediction accuracy, ;(ile lo;er ma' orders ;ill fa4or t(e s(ort term

prediction accuracy of t(e models

Ma'imum order per I9- pair+ $ubspace identification

Page 55: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 55/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation55

Ma'imum order per I9- pair+ $ubspace identification

Ma'imum order per I9- pair+ $ubspace identification

Page 56: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 56/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation56

Ma'imum order per I9- pair+ $ubspace identification

Ma'imum order per I9- pair+ $ubspace identification

Page 57: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 57/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation58

Ma'imum order per I9- pair+ $ubspace identification

"rift Remo4al ! $ubspace

Page 58: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 58/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation5>

"rift Remo4al $ubspace

ll process units are disturbed by

se/eral drift disturbances.

rift disturbances are 1ind of

disturbances for which you don’tha/e measure, and it is not possible to

include them as FF si+nal. >ecause of

these drift disturbances, your process

+ain chan+es o/er time.

7ample- in heat echan+er, process

stream is bein+ heated by ?@ steam.

 Now, heat echan+er will be

 pro+ressi/ely foulin+. Ao o/er a

 period time , heat transfer coefficient

will come down, and for the same

amount of steam chan+e 8say (tph)Btemperature chan+e of the outlet

 process steam will not be the same

8will be the less as more and more

foulin+ happens).

"ata Ere!processing+ $ubspace identification

Page 59: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 59/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation5?

"ata Ere processing+ $ubspace identification

 4ailable option on data preprocessing are listed belo;+

1)"etrending

2)"ifferencing

)Uero!mean

%)"ouble "iff 

Detrending- @rocess test data often contains low freuency drift that is introduced by un1nown disturbances. 3hese low

freuency disturbances will create ne+ati/e impacts on identification and has to be eliminated. 3he option “etrendin+” is

desi+ned to do that ob.

ssume a measured si+nal can be di/ided into a process si+nal and a trend si+nal.

3he pre5process of data with etrendin+ will estimate the trend si+nal and remo/e it from the system .  >y default, the time5

constant of etrendin+ filter is calculated based on the caseDs 33AA-

>asis for &*33AA and #*33AA 5 how much etrendin+ filter should be used2

 If central average is calculated over a period of 3*TTSS t!en filter operation tends to remove only slo" drift

disturbances and leave dynamic information mostly intact# $ too small TTSS value for a case may lead to an “ over%

 Detrending ”  t!at may remove useful dynamic information from t!e data and cause inaccurate models#

"ata Ere!processing+ $ubspace identification

Page 60: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 60/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation60

"ata Ere processing+ $ubspace identification

"ata Ere!processing+ $ubspace identification

Page 61: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 61/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation61

"ata Ere processing+ $ubspace identification

"ifferencing+ yA't1) L 't)Aimilar to etrendin+, differencin+ is also intended to remo/e slow drifts from the process data. 3he

FI model I has been usin+ differencin+ as a data pre5processin+ strate+y since the ?C was first

 built. ifferencin+ can remo/e the slow disturbance, but it reduces Ai+nal to Noise 8AEN) ratio.

3herefore, for better model +ains and less strin+ent reuirements on AEN, the etrendin+ is

recommended in place of differencin+.

Uero mean+ It simply remo/es only an offset from raw data. It should be aware that there will be

multiple offsets if data consists of se/eral slices. 3he ero5mean operation will be performed on each

data slices.

"ata Ere!processing+ $ubspace identification

Page 62: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 62/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation62

p g p

"ouble "ifferencing+ Q A y t1) L y t) A 't1)! 't) L 't) ' t!1)

A 't1) L 2 't) ' t!1)

"ouble differencing ;as originally designed for Ramp and Eseudo!Ramp #< in FIR

I" operation For subspace I", it is no longer t(e only ;ay to preprocess Ramp

#<s &it(er "ifferencing or "ouble "ifferencing as an option ;as selected, a single

"ifferencing ;ill be first applied to a Ramp or Eseudo!Ramp #<

In addition, Ramp #< or Eseudo!Ramp #< are al;ays put into one or more Ramp

#< group by t(e uto!grouping algorit(m in order to separate Ramp #<s from non!

Ramp #<s

-4er $ampling Ratio+ $ubspace identification

Page 63: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 63/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation6

p g p

In t(e case ;(ere t(e process data are collected in a

(ig( fre/uency sampling and t(e final model for on!linecontrol ;ill be run in slo;er control sampling fre/uency, a

4alue of greater t(an one can be used to matc( t(e

modelSs sampling rate ;it( t(e controllersS

measurements &g L # analyQers

-4er $ampling Ratio+ $ubspace identification

Page 64: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 64/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation6%

p g p

T(e model you get by setting -4er!$ampling Ratio A 1 ;ill

 (a4e less precise gain but less delay ;it( t(e dynamics

"ynamic delay means t(e model calculated by $$ isslo;er t(an real system

:ast cur4e is ;it(

$RA10, and it is laggingbe(ind $RA1 cur4e

"irect Term "!matri')+ $ubspace identification

Page 65: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 65/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation65

) p

#ase ;it( non!e'istent in4erse response

Page 66: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 66/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation66

#ase ;it( non e'istent in4erse response

#ase ;it( non!e'istent initial kick

Page 67: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 67/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation68

In <al4e -9E, " term must

"ata $licing L FIR 4s $ubspace

Page 68: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 68/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation6>

g p

In subspace algorit(m, (o; muc( data loss ;ill (appen on data

slicing depends on t(e ma'imum order n) per I9- pair NnO data pointsare lost after eac( bad slice

In FIR algorit(m, (o; muc( data loss ;ill (appen on data slicing

depends on t(e time to steady state Tss) "ata of time period e/ual

to one Time to $teady state is lost after eac( bad slice

Identification speed L FIR 9$$ I"

Page 69: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 69/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation6?

p

*  $ubspace identification takes more computation time t(an t(e FIR

identification, in particular for models ;it( a large number of independent

and dependent 4ariables

*T(is occurs because t(e subspace identification does a true multi!input

multi!output MIM-) model identification t(at needs more intensi4e

computation t(an t(e multi!input single!output MI$-) FIR identification

*T(e benefit is t(at a true MIM- model (as less uncertainty, and is

statistically more accurate

*T(e computation time (as a cubical relation ;it( t(e NMa'imum -rderO

FIR I" 9 $$ I" $ummary 

Page 70: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 70/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation80

*  FIR algorit(m is like cur4e!fitting non!parametric) , $ubspace

algorit(m is finding parameter for pre!defined model structure

*  utomatic model order determination! $$ I" auto!determines t(eoptimal model order, to capture balanced (ig( and lo; fre/uency

dynamics , ;it(out o4er!fitting= ;(ereas in FIR I" model order is pre!

specified based on TT$$ and number of coefficients T(e goodness of

data fitting is t(e only obecti4e of FIR I"

*FIR is MI$-, $$ is truly MIM-*FIR fails ;(en $93 is lo;, $$ performs better *&fficient $licing L $$ I" is more efficient in term of slicing

o;e4er, because of (ig( degree of freedom t(at a FIR model can

offer ;(en fitting a model to a dataset, t(e FIR models can easily

capture (ig( order effects

Page 71: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 71/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation81

Model RobustnessModel Robustness

Model .ncertainty

Page 72: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 72/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation82

When we measure some physical quantity with an instrument and obtain anumerical value, we want to know how close this value is to the true value.

The difference between the true value and the measured value is the error.Unfortunately, the true value is in general unknown and unknowable. Sincethis is the case, the exact error is never known. We can only estimate error.

=1e$e?s no suc1 t1ing as a pe$%ect measu$ement@@=1e$e?s no suc1 t1ing as a pe$%ect measu$ement@@

Pode Elot L Fre/uency "omain .ncertainty

Page 73: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 73/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation8

Pode plot or fre/uency response) of any dynamic system, is defined as t(e

magnitude same as amplitude) of t(e sine ;a4e obser4ed in a specific #<, for a

constant fre/uency sine ;a4e ;it( a peak amplitude of 10 in t(e particular M< st(e fre/uency of sine ;a4e 4aries from t(e real slo; smaller TT$$) to 4ery fast

larger TT$$), t(e amplitude of t(e sine ;a4e 4aries in t(e #< significantly

* Input+ ct) A sinVt

* For simple first order systems) A Cp9Ws1)

* -utput ;ill be + yt) A S sinVt X)

* (ere S A f , V,W, Cp)

Pode Elot L Fre/uency "omain .ncertainty

Page 74: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 74/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation8%

Time "omain .ncertainty

Page 75: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 75/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation85

.ncertainty Results

Page 76: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 76/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation86

.ncertainty Results

Page 77: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 77/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation88

.ncertainty Results

Page 78: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 78/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation8>

T(e first in4erse response is in red Qone, and it is good to appro'imate

as dead time

Page 79: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 79/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation8?

Importance of Model .ncertainty analysis during $tep Test+1) .sed to monitor (o; accurate step test is going on

2) .sed to find out ;(en to slice out data

) Retuning step siQe

%) .sed to find out missing Feed for;ard 4ariables

5) .sed to take decision to ;indup step test

Page 80: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 80/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>0

$tep Test Monitoring

Page 81: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 81/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>1

Retuning $tep $iQe

Page 82: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 82/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>2

$tep test completion

Page 83: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 83/84

February 21, 2016 Reliance Technology Group – APC / RTO   For  Internal Circulation>

 ddition 9 "eletion of Feed for;ard 4ariables

Page 84: 06 Model Robustness CTS

7/21/2019 06 Model Robustness CTS

http://slidepdf.com/reader/full/06-model-robustness-cts 84/84