ANNUAL REPORT 2009University of Illinois, August 5, 2009
Bryan PetrusBG Thomas
Joseph Bentsman
Department of Mechanical Science and Engineering
University of Illinois at Urbana-Champaign
Online Control of Spray Cooling Using
Cononline
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 2
Overview• Cononline Overview
– Consensor: “software sensor”– Concontroller: PI controller bank– Monitor
• Controller performance comparison
• Future research
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 3
Project Motivation: Approaches to Cooling Spray Control
1) Manual control:– Operator sets of water flow rates– Difficult at high casting speeds when response times must be
short
2) Casting-speed-based control:– Set water flow rates according to casting speed– Results in non-optimal cooling during transient conditions
3) Conventional feedback control:– Limited measurement opportunities– Pyrometers etc. can be unreliable in spray zones
4) Software-sensor-based control:– Create “software sensor,” an accurate, real-time computational
model to base control on
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 4
OverviewHuman-Machine
Interface
Shell thickness and surface temperature estimation
Spray water flow rates
Setpointoptions
Software Sensor2-D transient thermal model
(200 moving 1-D slices)
steelsteel P steel
T TC k
t x xρ ∗ ∂ ∂ ∂ = ∂ ∂ ∂
ControllerSeparate PID controller for each spray zone
caster data
ΣΣΣΣ+
-
Caster
AUTOMATIC CONTROL LOOP
MAN/MACHINE SUPERVISORY
LOOP
SetpointGenerator
Surface temperature setpoint
pK
iK1
s
ΣΣΣΣ
ΣΣΣΣ
Saturation
ΣΣΣΣ +-
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 5
Computer Architecture
Controller Computer
(Slackware Linux)
CommServer
shared memory
CommServer
CONCONTROLLER
ActiveXServer Model Computer
(CentOS Linux)
shared memory
CommClientCONSENSOR
Windows Computers
CononlineMonitor
CononlineMonitor
TCP/IP connection
Shared memory connection
Legend:
Molten Steel
z
Meniscus
Slab
Torch Cutoff Point
Tundish
Mold
Ladle
Support Roll
Strand
Liquid Pool Metallurgical
Length
Spray Cooling Solidifying Shell
Submerged Entry Nozzle
Caster Automation
CommClient
Current control logic
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 6
Consensor Overview: CON1D
• Fundamentally based transient finite-difference model:
• CON1D predicts:– shell thickness– temperature distribution– heat flux profiles
• Suitable for real-time model– Can simulate entire
caster in < 1 second– “Restart mode”: Can stop
simulation at arbitrary point, continue later
22*
2 steel
steel steel steel
kT T TCp k
t x T x
∂ = + ∂
∂∂ ∂ρ∂ ∂ ∂
Molten Steel
z
Meniscus
Slab
Torch Cutoff Point
Tundish
Mold
Ladle
Support Roll
Strand
Liquid Pool Metallurgical
Length
Spray Cooling Solidifying Shell
Submerged Entry Nozzle
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 7
Consensor Overview
• Multiple “slices”– Each second, simulate
each slice for 1 second– 200 slice simulation for 1
second each takes ~ same time as 1 slice through entire caster: < 0.5 seconds
• Consensor– stores and manages 200
CON1D slices– Interpolates between
slices to estimate full shell & temperature profile
CONSENSOR output domain
CON1D ou
tput d
omain
Distance below meniscus, z
Tim
e, t
( )01 1, / cT x L t t z V= ± = +( )0
2 2, / cT x L t t z V= ± = +
( )ˆ ,T z t t∗=t∗
t t∗ − ∆ ( )ˆ ,T z t t t∗= − ∆
01t
02t
( ),z t∗ ∗
( )01, / cz t z V∗ ∗+
( )( )01 ,cV t t t∗ ∗⋅ −
CON1D slices
CONSENSOR updates
Delay interpolation
Exact estimate
Surface temperature output locations
Casting velocity, Vc
zx
CON1D “slices”
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 8
Concontroller Overview:Spray Zones
Zone 1
Zone 2
Zone 6
(Inner/Outer)
Zone 4
Zone 7
(Inner/Outer)
Zone 5
(Inner/Outer)
Zone 3
4 x 1 + 3 x 2 = 10 controllers
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 9
Concontroller Overview• Zone-based PI control: 10 individual PI
controllers, one for each spray zone
• Controller Algorithm: At each second of time:1. Obtain surface temperature profile from CONONLINE. 2. For all 10 zones:
i. Compute the zone-based average surface temperature error for current zone:
ii. Use Terr to compute the water flow rate command:
3. Send all water flow rate commands to Consensor, caster automation, and Monitor
( )zone
ˆ( , ) ,
( )
s
jj
j
T z t T z t dz
T tL
− ∆ =
∫
( ) ( ) ( )0
tP Ij j j j ju t k T t k T t dt= ∆ + ∆∫
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 10
Setpoint Methodologies
1. Speed-based spray flow setpoints – current Nucor spray practices
2. Temperature setpoints(zone-averages) based on steady states for flows in (1)
3. Vary (2) based on casting conditions• Casting speed• Mold exit temperature
(mold heat flux, superheat)
4. Operator chosen 0 5000 10000 15000
800
850
900
950
1000
1050
1100
1150
1200
Distance from meniscus(mm)
Tem
pera
ture
( o C
)
Three types of setpoints
Real
FilteredZone-based averaged
Pattern 1 Pattern 2 Pattern 3 Pattern 4
(i/min) (in/min) (in/min) (in/min)
(gal/min) (gal/min) (gal/min) (gal/min)
Zone 1, Speed 1 0 0 0 0
Zone 1, Flow Rate 1 0 0 0 0
Zone 1, Speed 2 15.7 15.7 15.7 15.7
Zone 1, Flow Rate 2 26 24 26 23
Zone 1, Speed 3 31.5 31.5 31.5 31.5
Zone 1, Flow Rate 3 26 24 26 23
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 11
Monitor Overview:Profile Screen
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 12
Monitor Overview:Parameter Screen
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 13
Monitor Overview
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 14
Ongoing work on Cononline
• Adding new features to Monitor (for plant operation)– “Passive” mode that displays without allowing
changes to setpoints– Automatic resize window from 1024x768 up– Options (servers, mode) set in configuration file
• Changing to production versions of software– Programs run as Linux “daemons”– Program log files can be used for debugging
• Fixing stability issues• Multi-threading Consensor for faster running
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 15
Controller Performance Comparison
• Based on caster data recorded at Nucor Decatur– thanks to Terri Morris, Rob Oldroyd, and Alan Hable
• Simulations run in real-time at UIUC– HP servers, Intel Xeon processors
• Test situation: sudden slowdown– Casting speed drops from 3.0 m/min to 2.5 m/min 30
seconds into simulation• Comparing four different control methodologies
– No control (constant spray rates)– Spray-table based control– PI control with speed-based setpoints– PI control with mold-exit-temperature-based setpoints
(“fixed setpoints”)• All videos are recorded at 6x playback speed
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 16
No Control (fixed spray rates)
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 17
Spray table control
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 18
PI Control: speed-based setpoints
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 19
PI Control: fixed setpoints
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 20
Controller Performance Comparison
• Spray-table control displays temperature overshoot during slowdown
• PI control with speed-varying setpoints reacts quickly at all points throughout caster
• PI control with fixed setpoints reduces sprays more gradually in zones further down the caster 0 50 100 150 200 250 300 350 400
1020
1030
1040
1050
1060
1070
1080
1090
1100
1110
1120
Time (s)
Out
er ra
dius
sur
face
tem
pera
ture
(o C)
speed-dependent temperature setpointsfixed temperature setpointsno control (constant spray rates)spray table controlPI control, speed-dependent setpointsPI control, fixed setpoints
0 50 100 150 200 250 300 350 4001020
1030
1040
1050
1060
1070
1080
1090
1100
1110
1120
Time (s)
Out
er r
adiu
s su
rfac
e te
mpe
ratu
re (o C
)
speed-dependent temperature setpointsfixed temperature setpointsno control (constant spray rates)spray table controlPI control, speed-dependent setpointsPI control, fixed setpoints
Zone 2 (outer radius) average surface temperature
Zone 8 average surface temperature
0 50 100 150 200 250 300 350 4000
5
10
15
20
25
30
35
40
45
Time (s)
Spr
ay-w
ater
flux
Qsw
(L/
s/m
2 )
no control (constant spray rate)spray table controlPI control, speed-dependent setpointsPI control, fixed setpoints
Zone 8
Zone 2
Spray rates
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 21
Future Research: Advanced Control Development
• Surface temperature control does not guarantee metallurgical length control
– Develop control algorithm for centerline temperature
– Switch between objectives?• solidification front (prevent
whales)• surface temperature (steel
quality)?
• New NSF grant: “Hybrid Control of Continuous Casting for Whale and Crack Prevention”
– Closed-loop measurements are very spatially localized (discrete)
• Mold heat removal rate• Pyrometer readings
– Need temperatures between measurements (continuous)
|e|
texit = Lcaster/v
Open-loop predictor error evolution
Closed-loop observer actions: predictor reinitialization through controlled discrete transitions
t1 t2
Pyrometer locationsSlice initiation time
t0
Pre
dic
tio
n e
rro
rSlice exit time
Initial condition from mold heat removal rate t
Boundary Actuation:Cooling water spray rate, u(t), generates heat flux hspray(Ti(±L ,t – Tamb)
Boundary Sensing:• Mold heat removal rate, Qmold(t)• Boundary point temperature
measurements from pyrometers, Ti(±L ,t)
Boundary Disturbances:Uncontrolled heat flux from: roll/shell contact points, radiation, natural convection, (hroll + hrad_spray+ hconv)(Ti(±L ,t) – Tamb)
Model:1D slices traversing 2D cross-section of 3D strand, Ti(x,t)
Control Objective:Temperature along shell surface,Ti(±L ,t)
Slice velocity, Vc
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 22
Future Research: Model Development
• Make model robust to casting conditions and data errors
• Improve accuracy of model by adding physical behavior– More accurate heat transfer coefficients
(Sami, Xiaoxu’s research)– Possible hysteresis effects during spray
changes
University of Illinois at Urbana-Champaign • Metals Processing Simulation Lab • Bryan Petrus, BG Thomas, & Joseph Bentsman 23
Thank you!
• Continuous Casting Consortium Members
• Nucor Decatur– Terri Morris, Rob Oldroyd, Kris Sledge, Ron
O’Malley
• National Science Foundation– GOALI DMI 05-00453 (Completed July 29, 2009)– GOALI CMMI-0900138 (Received July 14, 2009)
• Other CCC grad students– Sami Vapalahti, Xiaoxu Zhou