the tno mission - fhi, federatie van technologiebranches tno egbert jan sol.pdf · 2 7...
Post on 11-Jun-2020
2 Views
Preview:
TRANSCRIPT
1
Egbert-Jan.Sol@TNO.nl
Egbert-Jan Sol, CTO TNO Industrie & Techniek
Production Process Automation
Paradigm shifts in sensors & control models
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl2
The TNO missionTo enable scientific knowledge to strengthen the capabilityof businesses and government to innovate.
Strategy: what is behind the horizon - value creation by BV NL in 2010-15
0%
10%20%
30%40%
50%
60%70%
80%90%
100%
1750 1800 1850 1900 1950 2000
China
WestOthers
India
Russia
Brazil
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl3
Where do we create value in the Euro State of the Netherlands(Industry: High-Tech Syst., Process, Autom. Supply)
Expo
rt (2
37)30
161
28
18
(c) Berenschot, Min EZ (adapted by E-J Sol)
Dutch (internal) market (324)
100 155
Financial, Media and
ICT services(620.000)
Infra, transport& construction
(917.000)
25
Food & Health Industry
(149.000)
44
Basis, Product, Machine & System Ind
(554.000)Total labor:
(2.240.000)
Process: - Rijnmond- Moerdijk- Terneuzen- Delftzijl- Geleen
Manuf. Industry:High Tech Syst. & Auto. Toelever.
- ZO NL
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl4
Anyone can compete on the old factors of economic growthNL should not compete on labor, capital, materials, but on knowledge
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl5
How? Not on costs!, so change the rules of the game
CASHCOWmarket leaders(- cost reduction
- life extension)
Price&
size ofmarket
STARfocus on speed
and market share
Old
Current
New
STARfocus on speed& market share
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl6
Redox disturbances in container glass production
Disturbance
Product
-4.75
-4.70
-4.65
-4.60
-4.55
-4.50
-4.45
-4.40
-4.35
-4.309:279:5
910:2
311:0
311:3512
:0713:1
114:3
115:1116
:2616
:5518:0
719:1120:3
121
:2722:3
923:430:391:43
2:393:434:395:43
6:397:438:399:43
9:2110:3
911:4312:3
913
:4314:3
915:4316:3
917
:4318:3
919:4
320:3
921
:4322:3
923:4
30:391:3
62:403:364:405:366:40
7:368:409:3610
:4011:3
612:4
013:3
614
:4015:3
616:4
017:3
618
:4019:3620:4
021:3
622:4
023:360:401:362:403:36
4:405:36
6:407:368:409:3610:3
411
:3812:3413:3
814:3
415:3
816:3417:3
818:3
419:3
820
:3421:3
822:3
423:3
80:341:382:383:384:34
5:386:347:388:349:38
10:34
11:38
12:34
13:3814:3
415:3
816:3
417
:3818:3419:3
820:3
421
:3822:3423:3
80:341:382:34
3:384:345:386:347:388:3
49:3810:3411:3812:3
413:3
814
:3415:3816:3
417:3
818:3
419:3820:3
421
:3822:3
423:380:341:3
82:343:384:345:3
86:347:388:349:3
810:3
411:3
812:3413:3
814
:3415:3
816:3417:3
818
:3419:3
820:3421:3822:3
423:3
80:341:382:3
43:384:345:38
6:34
7:38
8:349:3810:3
411
:3412:3
813:3714:3315
:3716:3
317:3
718:3319:3
720:3
321:3
722:3323:3
70:3
31:372:333:3
74:3
35:376:337:378:3
39:3310:3
311:3712
:3313:3
314:3
915:3516:3
917
:3518:3
919:3520:3
921
:3522:3
923:350:391:3
52:393:354:395:3
56:397:358:39
9:38
10:35
11:39
12:3513:3
914
:3515:3
916:3517:3
918
:3519:3
920:3
521:3922
:3523:3
90:351:392:3
53:394:355:396:357:398:359:3610:3
211
:3612:3
213:3
614:3
215
:3616:3
217:3
618:3219
:3620
:3221:3
622:3223:3
60:321:362:323:284:325:366:327:368:32
9:3610:3
211:3
612
:3713:3314:3
715:3316:3
717
:3318:3
719:3320:3
721
:3322:3
723:3
30:371:332:373:334:375:3
36:377:338:379:3310:3711:3
312:3713:3
314
:3815:3
416:3
817:3
418
:3819:3420:3
821:3
422
:3823:340:381:342:383:34
4:385:346:387:348:389:3
410:3
811
:3412:3813:3
414:3
815
:3416:3817:3
418:3
819:3
420:3821:3
422:3
823:3
40:381:342:3
83:344:385:346:3
87:348:389:3410:3
311:3
7
29-jul 30-jul 31-jul 1-aug 2-aug 3-aug 4-aug 5-aug 6-aug 7-aug 8-aug 9-aug 10-aug 11-aug 12-aug
log
(pO
2 (ba
r))
19
20
21
22
23
24
25
Fe2+
/Fe to
t (%
)
log(pO2)Fe 2+/Fetot (Opt.)
29 ju
li
30 ju
li
31 ju
li
1 au
g
2 au
g
3 au
g
4 au
g
5 au
g
6 au
g
7 au
g
8 au
g
9 au
g
10 a
ug
11 a
ug
12 a
ug
Process monitoring (sensor)
Process monitoring (model)
2
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl7
Glass Furnessbatch and feeder sensor measure redox state of melt
Change in batch sensor signal leads 10 hours later to change in glass product colour
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl8
-5.5
-5.0
-4.5
-4.0
-3.5
-3.0
-2.5
6-8-2003 0:00 8-8-2003 0:00 10-8-2003 0:00 12-8-2003 0:00 14-8-2003 0:00 16-8-2003 0:00
date (dd/mm/yyyy hh:mm)
log(
pO2(
bar))
pO2 batch area (around 1300°C)pO2 feeder (around 1200°C)feeder calculated from batch
Simultaneous measurement of redox near batch and in feeder
soft sensor
measured
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl9
GTM-X results:Refractory corrosion
GTM-XProcess
and designanalysis
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl10
GTM-X
Main Model•Navier-Stokes
- Finite Volume- Energy (buoyancy)
•Steady-state & transient•State-of-the-art solvers•Multiple Domains•Grid:
- body-fitted- multi-block- multi-level grids- structured- collocated- p-modifiers
•Parallel•Materials Database
GUI•Pre-processor •Post-processor
Glass•Batch models
- 2.5D, 3D, 2-phase•Radiation models:
- Rosseland, (spectral) DOM•Electrical boosting•Bubbling•Foam model•Crown model•Stirring•Energy sources•Particle trace•Redox•Glass colour change•Non-linear mixing•Volatilization•3D & 1D walls•Glass surface height•Thermal homogeneity•Refractory wear•Glass quality indices
Combustion•Radiation models:
- (spectral) DOM•Combustion models:
- soot, NOx, Oil- oxy-fuel, oxy-boosting- f, f-g, dissociation- FLAME
•Turbulence models:- k-ε, RSM, durbin, ellip. Blend.
•Refractory corrosion•3D & 1D walls
Design, Optimization, Trouble-shooting
CVD•Interface with Chemkin•Rarefied flow model (DSMC)•Compressible flow•Surface reactions•Gas Phase reactions•Efficient stiff systems solver•Multi-component diffusion:
- Stefan-Maxwell, Wilke•Plasma…
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl11
Model prediction of glass melt in the furnace after disturbances such as:– Furnace load– Fuel consumption and fuel distribution over the different burners– Electric boosting supply– Ambient air temperature– Quality of recycling cullet– Intensity of bubbling and stirring– Changes in raw material composition.
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl12
POD - Proper Orthogonal Decomposition:Mathematical technique to make CFD models very fast
+ α2(t) + …
i=1
N
i=1
N
~
POD time functioncoefficients
Basis functions
Reconstructedvariables
10 20 30 40 50 60
0
0
010 20 30 40 50 60
= α1(t)
• Weighted sum of dominant profiles obtained by Singular Value Decomposition (SVD) of a snapshot matrix filled with simulation results over a long period of time with many changes in process conditions:
T(x,y,z,t) = Σ φi (x,y,z) αi (t) , N << original number of unknowns
• Solve original equations for αi(t) after substitution of the weighted sum of (known) dominant profiles.
• Reconstruct solution by substitution of αi(t). This leads to a speed up of computations by a factor of 100 to 10.000!
3
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl13
This way you do not become “Weltmeister PPA”
What does this means for you?Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl14
The Paradigm shift from intelligent box to distribution?
70-80
Hardw.(IBM)
Appli. &Services(You ?!)
Death of distance
OpenSource
GiveAway
Hardware
2000-2020
MB/day
Fixed
Mobile
€/ MB/day
1 10
100
3 90 9891800
Legend
MIPS
Costs/MIPS
MainframesPC's
80-90
Softw.(Micro-Soft)
Hardw.
Comm.(KPN)
90-2000
Softw.
Proces-sing isfree.
Intelli.BoxParadig.
GridNetwork
NetworkedServices
Main-Frame 1970 Mini
1979
PC-AT 1984 Pentium
1992Notebk 1997
PDA 2001
S-i-P 2010 Push-Pin 2020
02468
1012
0 5 10 15
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl15
Learning curve for smart (punaiske) devices(from mainframe to ambient push-pin computer
Main-Frame 1970 Mini
1979
PC-AT 1984 Pentium
1992Notebk 1997
PDA 2001
S-i-P 2010 Push-Pin 2020
02468
1012
0 5 10 15
Log 10 (Amount of devices) 5 = 100K, 10=10B
Log
10 (V
olum
e (h
xbxl
) mm
3)
500 B 1x1x1 cmdevices by ? 2020
5B 20x20x20 cmdevices by today
(c) TNO Industrial Technologies, Egbert-Jan Sol, ejsol@dse.nl, 2004
Note: SiP = System in a Package
50 B 2x2x2 cmdevices by ? 2010
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl16
Agricult.monitoringCrowd
monitoring
Sensor explosion and sensor synthesis/fusion
# of sensor (motes) from a few till thousands and beyond
# of DomainsSynthesisOr Sensor
Fusion
Cameraarray andanteArray
microphone
smartwells
tegelaar
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl17
Monitoring: traditional
Sensor Signalprocessing
Featureextraction
Presentation
Control
PROCES Actuator
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl18
Model based monitoring
sensor signalprocessing
Featureextraction
Presentation
Control
PROCES Actuator
M
M
M
M
M
4
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl19
Let’s get into action
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl20
Provide process knowledge to controller: PID >>> MPC >>> RMPC
Z1W3Z1W2
Z1W1
PID
TC2
PID
TC1
PID
TC3
Z3M1
Zone 3 Zone 2 Zone 1
setpoint setpoint setpoint
Z1W3Z1W2
Z1W1
PID
TC2
PID
TC1
PID
TC3
Z3M1
Zone 3 Zone 2 Zone 1
RMPC
RMPCRigorous
Model basedPredictive
Control
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl21
RMPC results (improved temperature control glass melt:
1145
1150
1155
1160
1165
1170
1175
1180
1185
1190
0 2000 4000 6000 8000 10000 12000 14000
Time [days]
Tem
pera
ture
[C]
Grid temperature
Temperature set point
MPC controller switched on
Set point changes due to job change
No control actions taken, controller switched off
1186.0
1186.5
1187.0
1187.5
1188.0
13000 13050 13100 13150 13200 13250 13300
1145
1150
1155
1160
1165
1170
1175
1180
1185
1190
0 2000 4000 6000 8000 10000 12000 14000
Time [days]
Tem
pera
ture
[C]
Grid temperature
Temperature set point
MPC controller switched on
Set point changes due to job change
No control actions taken, controller switched off
1186.0
1186.5
1187.0
1187.5
1188.0
13000 13050 13100 13150 13200 13250 133001186.0
1186.5
1187.0
1187.5
1188.0
13000 13050 13100 13150 13200 13250 13300
RMPCRigorous
Model basedPredictive
Control
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl22
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x 104
1400
1402
1404
1406
1408
1410
1412
1414
1416
1418
1420Measured throat temperature
time
Tem
pera
ture
(0 C)
00.511.522.53
prob
abili
ty d
ensi
ty fu
nctio
nprobability density
Cpk
= 0.
96
024681012
Cpk
= 0.
96C
pk=
4.3
024681012
Cpk
= 0.
96C
pk=
1.6
Economicbenefit
RMPC benefit (stabilize temperature allowing lower average temperature energy savings): RMPC
RigorousModel based
PredictiveControl
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl23
Conclusion: new production automation opportunitiesmany low cost sensors combined with advanced process models
80-90 70-80
Comm.
Softw.
Hardw.Hardw.
90-2000
Softw.
Proces-sing isfree.
2000-2020
Paradigm shift:From early adaptorTo visionaries,…To main street
Appli. &Services(You ?!)
Death of distance
OpenSource
GiveAway
HardwareEarly Market
Chasm
Tornado
BowlingAlley
Delft, 30 november 2005Egbert-Jan.Sol@TNO.nl24
TNO Theme: Maximasing performance in proces industry
• Modeling of physical process (fluidics, heat transfer, acoustics, separation, ..)
• Process (micro)Sensors & Control Models
From a single state variable to FEM models with multiple state
variables at every node combined with many (micro) sensors & sensor synthesis
top related