1 Sergej Jegorov, Piotr Wasiewicz 5-th DAMADICS Workshop Łagów, Poland, April 5th-7th, 2004 ACTUATOR BENCHMARK RESULTS: STEP I AND II.

1

Sergej Jegorov, Piotr Wasiewicz

5-th DAMADICS Workshop Łagów, Poland, April 5th-7th, 2004

ACTUATOR BENCHMARK RESULTS: STEP I AND II

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

2

Presentation Plan

• Control Valve Introduction

• Step I

• Step II

• Conclusions

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

3

Actuator structureActuator structure

Process VariableProcess Variable

• CV – Control Value

• Z – Valve Position• P1 – Valve Input Pressure • P2 – Valve Output Pressure• F – Medium Flow Rate

• CV – Control Value

• Z – Valve Position• P1 – Valve Input Pressure • P2 – Valve Output Pressure• F – Medium Flow Rate

Control Valve Introduction

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

4

Considered models of control valve

• servomotor rod movement model(1)

• control valve model (2)

• actuator model (3)

• simplified model of the control valve (4)

• simplified model of actuator(5)

)(ˆ CVfZ

),,(ˆ21 PPZfF

),,(ˆ21 PPCVfF

)(ˆ ZfF

)(ˆ CVfF

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

5

Diagnostic Matrix (theoretical)

f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14 f15 f16 f17 f18 f19

S1 ±1 ±1 -1 ±1 -1 -1 ±1 X ±1 -1

S2 -1 1 -1 1 ±1 ±1 X 1 -1

S3 ±1 -1 1 ±1 -1 1 ±1 ±1 1 1 ±1 ±1 X ±1 1 1 -1

S4 -1 1 -1 1 ±1 ±1 X ±1 1 -1

S5 ±1 -1 1 ±1 -1 1 ±1 ±1 1 1 ±1 ±1 X ±1 1 ±1 1 -1

Control valve vaultsPneumatic servo-motor

faultsPositioner faults External faults

S1 = f(r1), (11)S2 = f(r2), (12)S3 = f(r3), (13)S4 = f(r4), (14)S5 = f(r5). (15)

Val

ve c

logg

ing

Val

ve p

lug

or v

alve

sea

t se

dim

enta

tion

Val

ve p

lug

or v

alve

sea

t er

osio

n

Incr

ease

d of

val

ve o

r bu

shin

g fr

icti

on

Ext

erna

l lea

kage

Inte

rnal

leak

age

Val

ve c

logg

ing

Med

ium

eva

pora

tion

or

crit

ical

fl

ow

Tw

iste

d se

rvo-

mot

or's

pis

ton

rod

Serv

o-m

otor

's h

ousi

ng o

r te

rmin

als

tigh

tnes

s

Serv

o-m

otor

's d

iaph

ragm

pe

rfor

atio

n

Serv

o-m

otor

's s

prin

g fa

ult

Ele

ctro

-pne

umat

ic t

rans

duce

r fa

ult

Rod

dis

plac

emen

t se

nsor

fau

lt

Pre

ssur

e se

nsor

fau

lt

Pos

itio

ner

feed

back

fau

lt

Pos

itio

ner

supp

ly p

ress

ure

drop

Une

xpec

ted

pres

sure

cha

nge

acro

ss

the

valv

e

Ful

ly o

r pa

rtly

ope

ned

bypa

ss

valv

es

Flo

w r

ate

sens

or f

ault

r1 = Z – Z*(CV), (6)r2 = F – F*(Z, P1, P2), (7)r3 = F – F*(CV, P1, P2), (8)r4 = F – F*(Z), (9)r5 = F – F*(CV), (10)

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

6

FDI Structure

Inputs Outputs

Residuals Faults+

-

Fault detection Fault isolation

Process

Neural Network

Fuzzy logicController

FD using NN, FI using Fuzzy LogicFD using NN, FI using Fuzzy Logic

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

7

Search of Optimal NN Architecture

Name Delays1

Layer2

Layer3

Layer

Net 2 10 5 1

Net1 2 30 8 1

Net3 2 6 3 1

Net4 2 6 1 0

• Transfer functions of all layers are logsig

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

9

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

f_3f

1/24

f_t

Data Filtering

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

z

1

f_2f1/12

f_t

z

1

z

1

z

1

z

1

z

1

input_f_f4i1/6

input_f4i

filter1 filter2

filter3

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

10

NN Architecture Search. Training data without filtering

0 1000 2000 3000 4000 5000 6000-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

0 1000 2000 3000 4000 5000 6000-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8 net1

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6 net2

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6 net3

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

net4

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

11

Results achieved by applying filtering of measurements

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Fault f4Trained on filter1 but works on filter2

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0 1000 2000 3000 4000 5000 6000-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Fault f4 Trained on filter1 and filter2, but works on filter3

Blue – NN trained on filtr1Red – NN trained on filtr2

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

12Time Diagrams of real and modeled signals. NN trained applying filter2, but examined using filter3

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 NNr2

0 0.5 1 1.5 2 2.5

x 104

-0.35

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0 0.5 1 1.5 2 2.5

x 104

-0.35

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1 r2 res

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 NNr3

0 0.5 1 1.5 2 2.5

x 104

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0 0.5 1 1.5 2 2.5

x 104

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4 r3 res

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

13Time Diagrams of real and modeled signals. NN trained applying filter2, but examined using filter3

0 0.5 1 1.5 2 2.5

x 104

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0 0.5 1 1.5 2 2.5

x 104

-0.7

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

NNr4

r4 residual

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.5 1 1.5 2 2.5

x 104

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

NNr5

0 0.5 1 1.5 2 2.5

x 104

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

0 0.5 1 1.5 2 2.5

x 104

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

r5 residual

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

14

Measures of a Neural Networks "health"

-11 -9 -7 -5 -3 -1 1 3 5 7 9 110

20

40

60

80

100

120All Weights and Biases

-11 -9 -7 -5 -3 -1 1 3 5 7 9 110

5

10

15

20

25

30

35

40All Weights and Biases

-11 -9 -7 -5 -3 -1 1 3 5 7 9 110

5

10

15

20

25

30

35

40All Weights and Biases

-11 -9 -7 -5 -3 -1 1 3 5 7 9 110

5

10

15

20

25

30All Weights and Biases

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

15

Diagnostic Matrix (Practical)f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14 f15 f16 f17 f18 f19

S1 +1 +1 x +1 x x +1 x x X -1 +1

S2 -1 +1 x +1 x x x x +1 X +1 +1

S3 +1 -1 +1 +1 x +1 +1 x x +1 x x +1 X -1 +1 +1 +1

S4 -1 +1 x +1 x x x x +1 X -1 +1 +1

S5 +1 -1 +1 +1 x +1 +1 x x +1 x x +1 X -1 +1 -1 +1 +1

Control valve vaultsPneumatic servo-

motor faultsPositioner faults External faults

Val

ve c

logg

ing

Val

ve p

lug

or v

alve

sea

t se

dim

enta

tion

Val

ve p

lug

or v

alve

sea

t ero

sion

Incr

ease

d of

val

ve o

r bu

shin

g fr

ictio

n

Ext

erna

l lea

kage

Inte

rnal

leak

age

Val

ve c

logg

ing

Med

ium

eva

pora

tion

or c

ritic

al f

low

Tw

iste

d se

rvo-

mot

or's

pis

ton

rod

Serv

o-m

otor

's h

ousi

ng o

r te

rmin

als

tight

ness

Serv

o-m

otor

's d

iaph

ragm

per

fora

tion

Serv

o-m

otor

's s

prin

g fa

ult

Ele

ctro

-pne

umat

ic tr

ansd

ucer

fau

lt

Rod

dis

plac

emen

t sen

sor

faul

t

Pres

sure

sen

sor

faul

t

Posi

tione

r fe

edba

ck f

ault

Posi

tione

r su

pply

pre

ssur

e dr

op

Une

xpec

ted

pres

sure

cha

nge

acro

ss th

e va

lve

Fully

or

part

ly o

pene

d by

pass

val

ves

Flow

rat

e se

nsor

fau

lt .

DGN0 DGN1 DGN2 DGN3 DGN4 DGN5 DGN6

0 1 -1 0 0 -1 0

0 0 -1 1 0 0 0

0 1 -1 1 0 -1 0

0 0 -1 1 0 0 -1

0 1 -1 1 0 -1 -1

Fault free

f1, f4, f7, f10, f16

f2

f3, f6, f13, f18, f19

f5, f8, f9, f11, f12, f14

f15 f17

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

16

• Sugeno Type• Defuzzificatiom method is whatever (wtaver)

Fuzzification (Membership Functions)

Input Membership Functions Input Membership Functions r3, r5r3, r5

Input Membership Functions Input Membership Functions r1, r2, r4r1, r2, r4

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

17

Examples of results of isolation of fault 1

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

18

Examples of results of isolation of fault 3

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

19

Step I results (from forms S1-FF-fx)

f1 f1m f2 f2inc f3inc f4inc f5inc f6inc f7 f7m f8 f9 f10 f11

td 5s 15s 2s 8.278e+004s 5.378e+004s 4130s - 1.672e+004s 5s 5s - - 5s -

rfd 0% 0% 0% 0% 0% 0% - 0% 0% 0% - - 0% -

rtd0.98% 0.95% 0.99% 0.94% 0.72% 0.13% - 0.89% 0.98% 0.98% - - 0.98% -

fsd 0.75 0.5 0.75 0.98 0.64 1 - 0.19 0.75 0.5 - - 0.75 -

tit 5s 15s 2s 8.278e+004s 5.378e+004s 4130s - 1.672e+004s 5s 5s - - 5s -

rfi 0% 0% 0% 0% 0% 0% - 0% 0% 0% - - 0% -

rti 0.98% 0.95% 0.99% 0.92% 0.72% 0.02% - 0.89% 0.98% 0.98% - - 0.98% -

rmi 0% 0% 0.02% 0.38% 0.021% 0.12% - 0.01% 0.02% 0.01% - - 0.01% -

fsi 0.75 0.5 0.75 0.98 0.6403 1 - 0.19 0.75 0.5 - - 0.75 -

dacc 0.2 0.2 1 inf 0.2 0.2 - 0.2 0.2 0.2 - - 0.2 -

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

20

Step I results (from forms S1-FF-fx) continued

f12 f13 f13m f13inc f14 f15 f16 f17 f17 inc f18 f18m f18inc f19 f19m

td - 5s 6s 45s - 231s 7s 6s 640s 5s 6s 9409s 7s 7s

rfd - 0% 0% 0% - 0% 0% 0% 0% 0% 0% 0% 0% 0%

rtd- 0.98% 0.97% 0.97% - 0.92% 0.97% 0.97% 0.91% 0.98% 0.97% 0.93% 0.97% 0.97%

fsd - 0.75 0.5 0.07 - 0.75 0.75 0.76 0.17 0.75 0.5 0.112 0.75 0.5

tit - 5s 6s 45s - 240s 7s 35s 640s 5s 6s 9409s 7s 7s

rfi - 0% 0% 0% - 0% 0% 0% 0% 0% 0% 0% 0% 0%

rti - 0.98% 0.97% 0.97% - 0.66% 0.97% 0.90% 0.91% 0.98% 0.97% 0.93% 0.97% 0.97%

rmi - 0.01% 0% 0% - 0.28% 0% 0.07% 0.03% 0% 0% 0% 0% 0%

fsi - 0.75 0.5 0.07 - 0.75 0.75 0.75 0.17 0.75 0.5 0.112 0.75 0.5

dacc - 0.2 0.2 0.2 - 0.07 0.2 1 1 0.2 0.2 0.2 0.2 0.2

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

21

Step II results

scenario 1 2 3 4 5 6 7 8 9 10 11 12

Fault or group

DGN1 - DGN3 DGN5 - DGN5 DGN2 DGN5DGN5 or 1

DGN1 DGN1 DGN3

Start time 15073 - 37619 82491 - 80965 62438 24809 20778 24129 855 55196

Stop time inf - 73370 82519 - 80979 76487 24821 49064 35621 11963 80822

scenario 13 14 15 16 17 18 19 20 21 22 23 24

Fault or group

DGN6 DGN3 DGN5 DGN3 DGN3 DGN1 DGN1 - DGN2 DGN3 DGN3 DGN6

Start time 19817 58605 60001 30194 12430 1940 22556 - 18238 7400 5773 10717

Stop time 23743 77130 60014 Inf Inf 10039 35085 - Inf inf inf Inf

Łagów, Poland, April 5th-7th, 2004

5-th DAMADICS Workshop

22

Conclusions

• fault detection subsystem based on neural network technology was developed

• fault isolation subsystem based on fuzzy logic technology was developed

• Neuro-fuzzy FDI system is applicable for actuator fault diagnosis. Fault groups are distinguishable.

• Close to 1, true fault detection rates factors achieved in Step I confirms acceptable NN models quality

• High values of true fault isolation rates (Step I) confirms proper isolability features of fuzzy isolation scheme applied

• Results achieved in Actuator Benchmark Step I are highly acceptable

• The fault distinguishability problem exists because of limited availability of measurements when considering industrial benchmark Step II). In this case 6 fault groups are distinguishable.