Jihad Daoud, Igor Nipl
Enhanced Control by Visualisation of Process Characteristics:
Video Monitoring of Coal Powder Injection in a Blast Furnace
MASTER'S THESIS
Civilingenjörsprogrammet Institutionen för Systemteknik Avdelningen för Reglerteknik
2000:076 • ISSN: 1402-1617 • ISRN: LTU-EX--00/076--SE
Characteristics: Video Monitoring of Coal Powder
Injection in a Blast Furnace
Jihad Daoud
Igor Nipl
2000-02-29
Authorsaddress:
LuleåUniversity of Technology
Department of Computer Science andElectrical Engineering
Control Engineering Group
S-971 87LULEÅ,Sweden
jihdao-6@student.luth.se
igonip-6@student.luth.se
blastfurnace. Somedayyouwill become coal,andifyouarelucky,
youmightbeusedinablastfurnace.
Nothingisstatic,notus,notyou,notanyoftheimagesweanalysed.
Fightingagainstchanges,beingstatic,islike.... Noopiniononthat
one!
A language canbeused to control people, a computercanbeused
to control muchmore. It is onlystupidpeople/thingsthat are easy
to control,butagaineverythingisrelative.
Donotaskus aboutthetruth,wearestill searching. Whenwend
it,youwillknowitoryouarealreadydead. Ifyouarenotdead,you
aretoolazyoryouknowsomethingwedonotknow.
Tothepoorpeople.
Jihad
ToLindaandmyparents.
Igor
Thismasterthesis,basedonworkperformedatLuleåUniversityofTechnology
in cooperation withMefos, is aboutmeasurementof pulverisedcoal owinjected
intoablastfurnace,compensatingforsomeoftheusuallyusedcoke. Coalisdrawn
from aninjection vessel andtransported under pressurewiththehelp ofnitrogen
gastoablastfurnace. Itisblownthroughpipestothetuyereswhereitisinjected
intotheironmakingprocess.
Irregular coal supply to the furnace has bad inuence on the quality of the
produced ironso reliablecontrol is needed. In controllingthe ow, it is of great
importancethattheon-lineowmeasurementisaccurate. Enhancing theexisting
measurementwould be benecial for thequality of the produced iron. Therefore
newmeansofblastfurnaceprocesssurveillanceandowmeasurement,usingcam-
erasandimageprocessing,arestudied. Theideabehindcamerasurveillanceisalso
benecial forestimationofotherprocessparameters.
The main goal isobtaining relevantinformation from image data in order to
estimate thepulverisedcoalow. Methodsforachievingthisareinvestigatedand
discussed. Acomparisontooldmeasurementdataismade. Alsovalidationofdata
retrievedwiththehelpofimageprocessingismentioned.
Ithasbeenshownthatvideomonitoringinconjunctionwithimageprocessing
is a feasible option when it comes to coal ow estimation. The images include
potential information for other purposes likedetermining thetemperature of the
ame and how well the coal is distributed inside the blast furnace. This would
solve some of the problems and eliminate obstacles caused by the nature of the
steel makingprocess.
Peoplehavealwaysaskedus: Whystudyautomaticcontrol? Therealquestion
is: Why not? There are not many elds that aect our modern life as much as
automaticcontroldoes,ofcoursemathematicsandpossiblyphysicsarecornerstones
in anynutritiousstudy. Theyarehardtocompetewith. Appliedsciencein allits
forms isthewaytogoto enhanceproductsand toolsthat areessentialin today's
society. Theachievementsin the eld ofautomatic control aresurrounding us in
our everyday lives, no matter if we like it or not. A lot of things out there are
alreadydone,manymorearewaitingto bedone. Thereisalsoalotofnetuning
totakecareof,whichissometimesevenmorechallenging. Wewantedtobeapart
of this evolvingdevelopment. Wewant tothank AndersGrennberg,withouthim
wewouldnotbeclosingloopsthesedays.
Thisworkisapartofabiggerprojectwithinvolvementfromtheindustrialand
research world, backed up byPROSA - Centre for Process and System Automa-
tion 1
. Ourmaster thesiswascarriedoutat theDepartmentofComputerScience
andElectricalEngineering,ControlEngineeringGroupatLuleåUniversityofTech-
nology 2
in cooperationwithMefos 3
. Theworkyouholdin yourhands isbrought
to you by twohuman beings, but is aresult of many more human participants.
Peoplewithoutwhoseknowledgeandwillingnesstohelp,youwouldnotbeableto
read thisreport.
Duringthe time wespent onthis research welearnedto knowseveral people
withdierentbackgroundsfromdierentcompanies,gainedmoreunderstandingof
the complicated coalinjection process in ablastfurnace, and improvedourskills
in imageprocessing.
Wehad greathelpfrom ourexaminerProfessorAlexanderMedvedevandour
supervisorPh. D. OlovMarklund, bothatpresentworkingfortheDepartmentof
Computer Science and Electrical Engineeringat Luleå University of Technology.
Thank you for oering us a part of your valuable time. We would also like to
thank Andreas Johansson, Wolfgang Birk and other researchers at the Control
EngineeringGroup. RolandLindforsattheAV-centre. PerMäkikaltioand others
at the Division of Industrial Electronics and Robotics. Krister Engberg at the
Division ofSignal Processing forputting upwith us. Thesystem administrators,
Mattias Pantzare and Jonas Stahre for their indispensable help. All working at
LuleåUniversityofTechnology. TheFreeSoftwareFoundation 4
oeringtheworld
the best they can achieve, without them we would be dependent on commercial
software, exceptforMATLABwhere wehad notimeto writetheneeded toolbox
forOctave 5
. NottoforgetthehelpfulpeopleatMefos,LKAB,SecuritasandBjörn
OlssonatSSAB. Re-Tekmembersforkeepingupthespirit ofbeingatrociousand
takingthecomputersweborrowedbeforewenishedtheproject. El-Tekmembers
1
PROSA'shomepage. URL:http://www.sm.luth.se/csee/prosa/html/
2
Department of Computer Science and Electrical Engineering's home page. URL:
http://www.sm.luth.se/
3
Mefos'homepage.URL:http://www.mefos.se/
4
FreeSoftwareFoundation'shomepage.URL:http://www.fsf.org/
5
Octave'shomepage.URL:http://www.che.wisc.edu/octave/octave.html
formakingitpossibletobuyprovisionsduringthetimewespentwritingthisreport.
Lastbutnotleast,wewanttothankourfriendsfortheirpsychologicalsupport.
Thankyouall,youmadeusdoit.
Chapter 1. Introduction 11
Chapter 2. ProcessDescription 13
2.1. TheBlastFurnace 13
2.2. CoalInjection 15
2.3. CurrentControl 16
2.3.1. Sensors 16
2.3.2. Controllers 16
2.4. VideoSurveillance 17
Chapter 3. CollectingData 19
3.1. Available SignalsandEquipment 19
3.2. MeasuredSignals 20
3.3. VideoRecording 22
3.4. VideoDigitising 23
Chapter 4. DataProcessing 27
4.1. AnalysisofSignals 27
4.1.1. Filter Identication 33
4.2. Danalyzer 34
Chapter 5. ImageProcessing 35
5.1. ImageDecomposition 35
5.1.1. RGBandHSIspaces 35
5.1.2. ImageQuality 35
5.2. ImageContent 38
5.2.1. ImageHistograms 38
5.2.2. ImageThreshold 41
Chapter 6. Algorithms 45
6.1. FindingtheBackground 45
6.1.1. OneBackgroundApproach 45
6.1.2. CustomisedBackgroundApproach 46
6.2. FindingtheCoalPlume 52
6.3. EstimationofPlume'sArea 55
6.4. EstimationofPlume'sVolume 55
6.4.1. WeightedPixelEstimation 55
6.4.2. RotatedPlumeEstimation 56
6.4.3. ApproximatedShapeEstimation 56
Chapter 7. DataExtractionandValidation 59
7.1. RelationsBetweenAlgorithms 59
7.2. ExtractedDataCharacteristics 60
7.3. AndreasTestProperties 64
Chapter 8. ConclusionsandSuggestions 69
8.1. Conclusions 69
8.2. Suggestions 70
Bibliography 73
AppendixA. MATLABCode 75
A.1. mam.m 75
A.2. mip.m 75
A.3. imframe.m 75
A.4. imends.m 75
A.5. line2pixel.m 76
A.6. imconnect.m 77
A.7. imlter.m 79
A.8. imback.m 79
A.9. dyncrop.m 79
A.10. dynbg.m 80
A.11. bgmulti.m 80
A.12. imarea.m 81
A.13. algox.m 81
A.14. ndame.m 82
A.15. imame.m 82
A.16. algo2vol.m 82
A.17. evalvol.m 83
A.18. countpixels.m 83
A.19. ndvolume2.m 83
AppendixB. CCode 85
B.1. ssnap.c 85
Introduction
Heavy industries are the backbone of our society, any improvements in this
areamean indirectlyabetterstandardof living. Steel andironproduction isone
of these industries. Steelmakinghasevolveddramaticallysince mankindlearned
howto produceit. Yet thereis still alot todo because theprocessitself is quite
complicated and notfully understood. Newtechnology has contributed in many
ways to improve the steel making procedure, where involvement of people with
dierentbackgroundsandacademicknowledgeisessential.
In existing blast furnaces there are many problems, which remain unsolved
despite many yearsof thoroughresearch. New improvements and breakthroughs
are made everyday, but there will always be more work to be done due to the
sophisticated process nature. Eciency, quality, environmental issues and cost
reduction requirements are the main objectives. The fuel used in the furnace is
one of the targets. Changing the kind of fuel used has shown verygood results.
Traditionallycokeisused. Manyotheralternativefuels[10]havebeentested,such
aspulverisedcoal, naturalgas, oilbut alsowastematerials. Thefuture supplyof
coke[1]isanotherproblemthatmightleadtosteadilyincreasingprices. Pulverised
coalhasbecomeagoodalternative. Itis30-40%cheaperandmoreenvironmentally
friendly [12] than coke. Using pulverised coal resulted in a 40% saving in coke
requirementsatBritishSteel,Scunthorpeworks[11],[13]. Inaddition,pulverised
coalhasaquickerimpactonthereactionin theactivezoneofthefurnace.
Beside choosinganalternativefuel,steelmakersneedabetteroverviewof the
process. Controllingthe product quality relies on identifying the process param-
eters from a metallurgical point of view and how well the process is controlled.
Controllingtheprocess,besideunidentiedprocessparameters,runsintoproblems
relatedtoowmeasurement,temperaturemeasurementandfueldistributioninthe
blast furnacewhich are hardto dealwith usingold-fashion techniques becauseof
thehighprocess complexityandtheverydemandingenvironment.
Quick development in computer hardware has opened new perspectives and
possibilities. At present it is an easy task to process alarge amount of data to
extractusefulinformationinordertocontrolandsupervisethesteelproductionin
ablastfurnace. Onewaytodothisistheuseofcameraspointedtothepulverised
coal outlet from thetuyeres into thefurnace. Cameraimages canbeanalysed in
ordertodeterminedierentimportantprocessparameters. Thegoalistocalculate
high qualityparametersthat reect whathappensin thefurnace. This willmake
life easierformetallurgicalandautomaticcontrolpeople.
A prestudy [2] hasshown thatthere is asignicantrelationbetweenthe pul-
verised coalmass owestimation and the size of the coal plume in theanalysed
video recorded series of images. Therecording was donewith a black and white
camera. Deeperinvestigationis requiredto verify theresultsand ndalgorithms
for calculation of the coal plume volume and coal owestimation, but also tem-
perature and coal powder distribution. It is also interesting to study the result
obtainedwithacolourcamerainordertouseseveralindependentestimationchan-
nels. Inthis report we will focus on therst part, i.e. coal ow, but our results
will hopefullybeusefulfortheothertargetstoo.
Process Description
WehadanopportunitytoworkonLKAB'sexperimentalblastfurnaceatMefos
[17],wherewehadasetupofcamerasandapossibilitytocollectneededdata. In
this chapter we will briey describethe dierent parts of the plant and the coal
injection partoftheprocess.
A D B
C
Figure2.0.1. AschematicoverviewoftheplantatMefos.
2.1. The BlastFurnace
TheblastfurnaceatMefosismarkedwithanA inFigure2.0.1. Apictureof
theactualinstallationisinFigure2.1.1. Ithasthreetuyeresandadiameteratthe
tuyerelevelof1.2m. Theworkingvolumeis8.2m 3
. Thehotblastisproducedin
pebbleheaters,capableofsupplying1300 Æ
Cofblasttemperature. Thefurnaceis
designed foroperating withatop pressureof1.5 bar. It hasabell-typecharging
systemwithoutmovablearmour. Thecoalinjectionsystem,seeBinFigure2.0.1,
features individual controlof coalowforeachtuyere. Gascleaning system,part
D ofFigure2.0.1,consistsofdustcatcherandelectrostaticprecipitator. Material
istransportedtothetopoftheblastfurnacethroughCasshowninFigure2.0.1.
A tapping machine withdrill andmud gunis installed. Theblastfurnaceis well
equippedwithsensorsandmeasuringdevices,andanadvancedsystemforprocess
control. Probesfortakingmaterialsamplesfrom thefurnaceduring operationare
beingdeveloped.
Figure 2.1.1. Theblastfurnacebody.
LKABusethefurnaceprimarilyfordevelopmentofthenextgenerationofblast
furnace pellets. The furnace performance shows that it is a good tool for other
development projects. An importantarea is recycling of waste oxides. Injection
of wasteoxidesis another research areaaswell asinjection of slagformers. The
interestingpartsoftheplantarepresentedinFigure2.1.2.
Air Lock Vessel
Vessel Coal Injection Slag Vessel
000000000000000 000000000000000 000000000000000 000000000000000 111111111111111 111111111111111 111111111111111 111111111111111
Blast Furnace
Camera Video Tuyeres
Figure 2.1.2. The most relevant parts, for this project, of the
experimental blastfurnaceatMefos.
Figure 2.2.1. Controlscreenforcoalinjection,atMefos.
2.2. Coal Injection
The coalinjection arrangement, Figure 2.2.1,consists of twocoalvesselsand
threepipeseachendingwithatuyere. Thethreetuyeresareevenlyspacedaround
theblastfurnace,asshownin Figure2.2.2.
Blast Furnace Video Camera
Pulverized Coal Pipe
Tuyere
Coal Plume
Furnace Wall
Figure2.2.2. Thethreetuyeresaresurroundingtheblastfurnace
withthesurveillancecamerassupportingframework.
LookingagainatFigure2.1.2. Theuppervessel,theonethatislledwithcoal
when needed, works asan airlock vessel used to pressurisethe coalvessel below.
From the lowervessel, called theinjection vessel, coal is divided and distributed
under pressurethroughpipesto the tuyeres in theblast furnacewith thehelp of
when desired. A big problem is determining the behaviour of the coal particles
travellingthrough the pipes. That is because of the various size of theparticles,
turbulenceeectsinthepipesandpipes'characteristics. Coalparticlescanclogin
apipe,whichcandisturbtheprocessbeforebeingdiscovered. Anotherproblemis
leakageofthecarriergas. Solutionforthelatterisproposedin[6].
2.3. CurrentControl
Theexistingcontrolofthepulverisedcoalowtotheblastfurnaceisbasedon
acontinuouson-linemeasurementofthecoalowitself. Althoughthisistrue,itis
notthewholetruth. Ithasbeenshownthattheowmeasurementdeviceisnotvery
accurate,thatiswhythecurrentcontrolisdependentonaweightmeasurementof
theinjectionvessel.
2.3.1. Sensors. Mainly we cantalk about three ow measurement devices,
everyone ofthem connectedto pipestransporting pulverisedcoal tothe tuyeres.
ThedevicesusedareRamseyDMK270industrialmassowrateandvelocitymea-
suringinstrumentsfornonhomogeneousmedia[16]. Internallythedeviceconsists
ofasolidsconcentrationsensorandavelocitysensorwithtwomeasurementpoints
separatedby adistanceS, basedonthecapacitivemeasuringprinciple. Thepar-
ticle streamismeasuredintwopointswhichthevelocitytransmittercorrelatesto
ndtheclosestsimilaritybetweenthem. Fromthiscorrelationfunction,thetransit
time T from point oneto point two canbe determined. Inthe solid concentra-
tion sensor the change in capacitanceis proportionalto the solids concentration,
this voltage signal is transformedinto afrequency signaland is Pulse Frequency
Modulated(PFM).
TheowrateQ isgivenbyQ=CV A
sensor
,whereC istheconcentration
of the medium, V its velocityand A
sensor
is the sensor cross-sectionareaand is
calculatedwithA
sensor
= d
2
sensor
4
,whered
sensor
isthesensordiameter.
TheconcentrationC= Ka
10
(f
PFM f
PFM0
)K,whereK
a
isanadaptionfactor
to concentration sensor, f
PFM
is the measured frequency of PFM concentration
signal, f
PFM0
is the frequency of PFM signal at concentration zero and K is a
calibration factor. Finallythe velocity is calculatedwith thewell known formula
V = S
T .
Nottoforgetthatthebothvesselsareequippedwithweightgaugesandpressure
meters. Other importantsensorsarethepressuremeasurementdevicesinthecoal
injection pipes.
2.3.2. Controllers. Figure 2.3.1 is a schematic overview of the main con-
trollers. Because the owmeter is not reliable, the on-line ow measurement is
multipliedbyacorrectionfactorcalculatedaccordingtotheinjectionvesselweight
loss deviation, during acertain period of time. See Figure 2.3.2for details. Do-
ing sothe idea ofon-line measurementis lost, in ashort term perspective,while
it is relatively accurate considering a longer period of time. The corrected ow
measurementitself is theoutput of aPI-controllerusing 1 asitssetup valueand
feedbackwiththe owmeters tocoal vessel weight ratio. Beforedividingtheow
measurementbythevesselweightbothsignalsarewindowedwithawindowlength
of 10minutesbefore performingthedivision. This isdonebecausethescaleused
in thevessel has alimitedsensitivityand a bigerror margin, if compared to the
owmeasurementduringashort time.
ThecoalinjectioniscontrolledwiththreePID-controllers,oneforeachtuyere.
An operator issupposed to feedthe systemwith the desiredamountof coalow
needed in the furnace, the amount is divided bythree and the resultserves asa
x
x
x
−
−
−
DMK 270 kg/s
DMK 270 kg/s
DMK 270 kg/s
DIV
− PI
DIV PID
PID PID Flow Meter
Tuyere Valve
3 Operator
Vessel Coal Injection
SUM
XXX
1 Slag Vessel
Setup
WINDOW
Figure 2.3.1. Coalpowderinjectioncontrol.
.
DMK 270 kg/s
Flow Direction Concentration
Velocity
Mass Flow 1
Function
X
Weight Multiplicator
Mass Flow 1 Weight Corrected Coal Pipe 1
Mass Flow 2 Mass Flow 3 Coal Vessel Weight
Slag Vessel Weight
Figure 2.3.2. A principleschemeforpulverisedcoalmassowcalculation.
feedback. Thecontrolsignalfrom each ofthePID-controllers isused tocontrola
valveplacedbeforetheowmeteroneachpipe. Acontrolbasedonthesetermscan
notbeperfecthavinginmindthebadqualityoftheresultingmeasured/calculated
signals. Dependingonhowthemultiplicatorchanges,thecontrolsignalwillbehave
dierently. We will show later in Chapter 4.1 that some control signals have a
strangebehaviour.
2.4. Video Surveillance
Theconditionssurroundingthesteel makingprocedure areratherrough. The
veryhigh temperature of theame and thehigh brightnessfrom inside the blast
furnace makelife hard for those whowant to control or study this process. The
existingcamerasatMefoscannothandletheincominghighlightintensityandthey
thevideopictureviewable. Thelteritselfisadarkgreenpieceofthickglassthat
is placedabit away from thecameralens. Thecamerasthemselvesare mounted
abouttwometersfrom theouterwalloftheblastfurnaceandarebuilt-ininsidea
protectingcover. Thelightfrom inside thefurnace isled to thecamerasthrough
peekholesinthewallneareachtuyereviaprotectingpipes.
Tuyere
Coal Plume
Pulverized Coal Pipe Furnace
Wall
Video Camera
Protecting Cover
Filter
Protecting Pipe
Panasonic
WV−CL410
Figure 2.4.1. Thevideocamerasetup.
For a full comprehension of the whole video camera apparatus, Figure 2.4.1
mightbehelpful forthedevouredreaders. Onitswaytothelens lightpasses the
previously mentioned glass lter. The characteristics of this lter have notbeen
examinedingreatdetailbuthavingacloserlookatthethreecolourbuersshows,
for the human eye, that the red and in particular the green light pass the lter
almost unaected while the blue light is ltered out to the extent that the blue
buer becomesnearlyuseless,asdiscussedin Chapter5. Itisthereforedesiredto
solvethelteringproblem. Wehadthepossibilitytousetransparentglassinstead
ofthegreenone,whichwebelievedwouldleaveallthethreecolourbuersunspoilt.
Bydoingthisweriskedintroducingoverexposuretothevideosurveillancesystem.
Possiblesolutionswillbediscussedlaterinthisreport.
(a)Withgreenglass. (b)Withtransparentglass.
Figure 2.4.2. Sampleimagestakenwithgreenglassaslterand
withtransparentglass.
An example of what is seen with the help of the cameras is in Figure 2.4.2,
where themouthof thetuyere isseentogetherwithadark,ellipticshapedcloud,
thepulverisedcoalinjectedinsidethefurnace.