Sch i en eihe CIplus, Band 4/2013
He ausgebe : T. Ba z-Beiels ein, W. Konen, H. S enzel, B. Naujoks
Simula ion and Op imiza ion o
Cyclone Dus Sepa a o s
Bea e B eide ho , Thomas Ba z-Beiels ein, Bo is Naujoks,
Ma in Zae e e , And eas Fischbach, Oli e Flasch,
Ma ina F iese, Ola Me smann and J¨o g S o k
Simula ion and Op imiza ion o Cyclone Dus
Sepa a o s?
Bea e B eide ho , Thomas Ba z-Beiels ein, Bo is Naujoks, Ma in Zae e e ,
And eas Fischbach, Oli e Flasch, Ma ina F iese, Ola Me smann, and Jö g S o k
Facul y o Compu e and Enginee ing Sciences
Cologne Uni e si y o Applied Sciences, 51643 Gumme sbach, Ge many
[email p o ec ed]
Abs ac . Cyclone Dus Sepa a o s a e de ices o en used o il e solid pa icles
om lue gas. Such cyclones a e supposed o il e as much solid pa icles om
he ca ying gas as possible. A he same ime, hey should only in oduce a mini-
mal p essu e loss o he sys em. Hence, collec ion e iciency has o be maximized
and p essu e loss minimized. Bo h he collec ion e iciency and p essu e loss a e
hea ily in luenced by he cyclones geome y. In his pape , we op imize se en
geome ical pa ame e s o an analy ical cyclone model. Fu he mo e, noise a i-
ables a e in oduced o he model, ep esen ing he non-de e minis ic s uc u e o
he eal-wo ld p oblem. This is used o in es iga e obus ness and sensi i i y o
solu ions. Bo h he de e minis ic as well as he s ochas ic model a e op imized
wi h an SMS-EMOA. The SMS-EMOA is compa ed o a single objec i e op i-
miza ion algo i hm. Fo he ha de , s ochas ic op imiza ion p oblem, a su oga e-
model-suppo ed SMS-EMOA is compa ed agains he model- ee SMS-EMOA.
The model suppo ed app oach yields be e solu ions wi h he same un- ime
budge .
1 In oduc ion
The educ ion o emissions om coal- i ed powe plan s is a demanding ask. Cyclone
sepa a o s a e equen ly used de ices o il e ing he lue gas o such plan s. They e-
mo e dispe sed pa icles om gas. Thei ad an ages a e simple s uc u e, low cos s and
ease o ope a ion. Collec ion e iciency and p essu e loss a e he wo mos impo an
pe o mance pa ame e s. They a e hea ily in luenced by he choice o se e al geome -
ical design pa ame e s, like heigh o diame e . This esul s in o a Mul i-Objec i e Op-
imiza ion (MOO) p oblem, he so called Cyclone Op imiza ion P oblem (COP). This
s udy shows how a COP can be sol ed and analyzed, based on an analy ical, de e -
minis ic model. Fu he mo e, he analy ical model is ex ended by adding se e al noise
?This is he p e-p in e sion o he ollowing a icle: B eide ho , B.; Ba z-Beiels ein, T.;
Naujoks, B.; Zae e e , M.; Fischbach, A.; Flasch, O.; F iese, M.; Me smann, O. and S o k,
J.: Simula ion and Op imiza ion o Cyclone Dus Sepa a o s, in P oceedings 23. Wo kshop
Compu a ional In elligence, H sg. Ho mann, F. and Hülle meie , E., Ka ls uhe, 2013.
2 B eide ho e al.
a iables. These enable o e alua e obus ness o solu ions, and yield a be e es ima e
o how noisy eal-wo ld ci cums ances a ec he p oblem. Techniques like a classical
as well as a model-suppo ed SMS-EMOA a e used o handle he MOO p oblem.
The emainde o he pape is s uc u ed as ollows. Sec ion 2 p o ides an o e iew on
p e ious esea ch and me hods w. . . he modeling and he COP in pa icula as well
as MOO. This sec ion is ollowed by a sho summa y o he esea ch ques ions and
goals in Sec ion 3. Sec ion 4 desc ibes he p oblem se up, in oducing pa ame e iza ions
o he COP. A sensi i i y analysis o he p oblem is depic ed in Sec ion 5, which is
ollowed by he desc ip ion o he pe o med op imiza ion expe imen s and hei esul s
in Sec ion 6. A sho summa y and a discussion o he indings is gi en in Sec ion 7.
The pape closes wi h an ou look on u u e esea ch in Sec ion 8.
2 P e ious Resea ch and Me hods
2.1 Cyclone Op imiza ion
Da
D
Be
HeHe
H
ε
Be
Da
D
F on View Top View
Fig. 1: Schema ic ep esen a ion o a cyclone dus sepa a o .
Cyclones exis in di e en shapes bu he e e se low cyclone ep esen ed in Fig. 1 is
he mos common design in indus y. The p inciple o cyclone sepa a ion is simple: he
Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 3
gas-solid mix u e en e s a he op sec ion angen ially. The cylind ical body induces a
spinning, o exed low pa e n o he gas-dus mix u e. Cen i ugal o ce sepa a es he
dus om he gas s eam: he dus is mo ed o he walls o he cylinde and down he
conical sec ion o he dus ou le while he gas exi s h ough he ou le pipe.
Signi ican pa ame e s o a cyclone Se e al cha ac e is ics cons i u e a COP.
1. Geome ic shape
Se en geome ic pa ame e s allow o desc ibe he cyclone as shown in Fig. 1.
2. Fluid/Gas p ope ies
Pa ame e s like iscosi y o densi y desc ibe he ca ie subs ance.
3. Pa icle P ope ies
Densi y, concen a ion and dis ibu ion o pa icle sizes desc ibe he pa icle com-
posi ion.
4. Collec ion e iciency (CE)
The o e all CE o he cyclone desc ibes he amoun o pa icles il e ed om he
gas.
5. P essu e Loss (PL)
The P essu e Loss is he di e ence in p essu e be ween inle and ou le .
These di e en cha ac e is ics a e summa ized in Table 1. P essu e loss and collec ion
e iciency a e he main c i e ia used o e alua e cyclone pe o mance. Bo h a e unc-
ions o he cyclone dimensions. No mally, he goal o cyclone design is o maximize
collec ion e iciency and o minimize p essu e loss by adjus ing he geome ic pa ame-
e s.
P e ious Op imiza ion S udies A i s mul i objec i e op imiza ion o cyclone sepa-
a o s was pe o med by Ra i e al. [2]. They used he Non Domina ed So ing Gene ic
Algo i hm NSGA II o op imize an analy ical model by Mo hes and Lö le [1], min-
imizing p essu e loss and maximizing o al collec ion e iciency o eigh geome ical
pa ame e s. Elsayed and Laco [3] op imized ou geome ical pa ame e s using compu-
a ional luid dynamics CFD models and he a model based on wo k by Ba h [4]. They
minimized p essu e loss only, using he esponse su ace me hodology. Pishbin and
Moghiman [5] op imized se en geome y pa ame e s wi h a gene ic algo i hm, mini-
mizing p essu e loss and maximizing e iciency. They used a CFD model o cons uc
he i ness unc ion. The bi-objec i e p oblem was ans e ed o a single-objec i e p ob-
lem using weigh s. Elsayed and Laco [6] minimized p essu e d op and cu -o diame-
e . They used a Pa e o op imiza ion app oach, u ilizing a Radial Basis Func ion Neu al
Ne wo k RBFNN. The RBFNN was ained wi h da a om li e a u e. A simila ap-
p oach was aken by Sa ikhani e al. [7], whe e he da a o ained neu al ne wo k
s emmed om CFD simula ions.
The he ein p esen ed wo k uses he analy ical model based on wo k by Ba h [4] and
Muschelknau z [8]. In con as o p e ious app oaches, we in oduce a s ochas ic sim-
ula ion based on he analy ical model, whe e se e al pa ame e s a e assumed o be
4 B eide ho e al.
Table 1: Table o luid, pa icle and geome ical pa ame e s used in he expe imen s. Mos alues
a e aken om an example by Lö le [1].
Pa ame e Symbol De aul Uni
Geome y Cyclone diame e Da1260 mm
Cyclone heigh H 2500 mm
Ou le pipe diame e D 420 mm
Ou le pipe imme sion H 640 mm
Cyclone cone angle 13.134 ◦
Inle heigh He600 mm
Inle wid h Be200 mm
Fluid Viscosi y µ18.5·10−6P a ·s
Flow Ra e Vp5000 m3
h
Gas densi y ρ 1.86 kg
l
Pa icle Pa icle densi y ρp2kg
l
Pa icle concen a ion ce50 g
m3
Ou pu P essu e Loss PL 2564 P a
Collec ion E iciency CE 0.89 (wi hou uni )
noisy. This allows o in es iga e obus ness o solu ions. Fu he mo e he mo e ecen
SMS-EMOA is used o sol e he mul i-objec i e COP. In case o he s ochas ic cyclone
simula ion, he model- ee SMS-EMOA compa ed o a model-suppo ed SMS-EMOA,
using a K iging su oga e model.
2.2 Analy ical Models o Dus Sepa a ion
Ba h [4] and Muschelknau z [8] p oposed a simple model based on a o ce balance,
as p esen ed by Lö le [1]. This model enables o ob ain he collec ion e iciency and
p essu e loss. The p inciple o calcula ion is based on he ac ha a pa icle ca ied by
he o ex is in luenced by wo o ces: a cen i ugal o ce and a low esis ance. They
a e exp essed a he ou le pipe adius iwhe e he highes angen ial eloci y occu s.
The model ep esen s a e e se low cyclone wi h a angen ial ec angula inle . This
is a simple and s ill use ul model, by which ic ion was i s in oduced in cyclone
modeling.
Collec ion E iciency The cyclone geome y, oge he wi h low a e, de ines he cu -
size o he cyclone. Cu -size de ines he pa icle size ha will be collec ed wi h 50%
e iciency. Smalle pa icles a e collec ed wi h lowe e iciency, la ge wi h highe e i-
ciency. Ba h [4] de eloped a ma hema ical model o he cu -size as ollows:
xG =s18µ i
(%p−%) 2
φi
(1)

Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 5
whe e iis he ou le pipe adius, is he adial gas eloci y on he ou le pipe and φi
is he cyclone inle eloci y.
The ac ional e iciency cu e assigns an e iciency o he pa icle diame e as shown
in Fig. 2. La ge pa icles a e collec ed mo e e icien ly han smalle pa icles. The
ac ional e iciency cu e is desc ibed by:
T(x) = 1 + 2
x
xG
3,564 !−1.235
(2)
whe e xis he pa icle size and xG equals o Eq. (1). The o e all collec ion e iciency
0 5 10 15 20 25 30 35
0.0 0.2 0.4 0.6 0.8 1.0
Pa icle Size [µm]
F ac ional E iciency
Fig. 2: F ac ional e iciency cu e.
is p edic ed acco ding o:
E=Zxmax
xmin
T(x)qe(x)dx =
xmax
X
xmin
T( ˜xi)∆Qe(xi)(3)
whe e xmin is he lowe bound o he pa icle size, xmax is he uppe bound o he
pa icle size, ˜xiis he mean pa icle size, ∆Qe(xi)is he change in dis ibu ion o
pa icle sizes and qe(x) = ∆Qe(xi)
∆xi.
P essu e Loss P essu e loss is de ined as he di e ence in p essu e be ween wo poin s
o a luid ca ying body. I occu s wi h ic ional o ces. I ela es di ec ly o ope a ion
cos . The e o e an exac p edic ion is e y impo an . To al p essu e loss equals o:
∆p =ρ
2 2
i(ξe−a+ξa−i+ξi−m)(4)
whe e ξe−ais he ic ion coe icien o he loss wi hin he inle (equals ze o because
o he angen ial ec angula inle ) , ξa−iis he ic ion coe icien o he loss wi hin
he cyclone body, ξi−mis he ic ion coe icien o he loss wi hin he ou le pipe and
ρ
2 2
iis he ela ionship be ween p essu e and eloci y.
6 B eide ho e al.
2.3 Model Al e na i es
The abo e desc ibed analy ical model is used as a p edic o o collec ion e iciency
and p essu e loss. Se e al o he analy ical models o he cyclone sepa a o exis [9].
Al hough all hese me hods ha e had a ema kable success, mo e ad anced ideas a e
needed o model cyclones. Uns eadiness and asymme y a e o example wo ea u es
no conside ed in classical cyclone heo y ha may a ec he eloci y dis ibu ion o a
g ea ex en , hus changing he model o he sepa a ion mechanism. On he o he hand,
as in many o he ields, CFD cu en ly eme ges as a po en ially accu a e modeling ech-
nique.
S ill, analy ical models p o ide a good s a ing poin o i s in es iga ions. Such mod-
els usually a e no as p ecise as he mo e complex CFD models, bu a e much as e
wi h espec o calcula ion ime and o he esou ces.
2.4 Mul i Objec i e Op imiza ion
In classical op imiza ion me hods only one objec i e is in es iga ed. This is di e en in
Mul i Objec i e Op imiza ion (MOO) whe e mo e han one objec i e can be op imized
in pa allel. Howe e , new concep s had o be de eloped because hese objec i es a e
o en con lic ing, i.e. an imp o emen in one objec i e au oma ically leads o a de e i-
o a ion in o he objec i es. He e, he concep o Pa e o dominance comes in o play. I
says ha solu ion adomina es solu ion bi ais no wo se in any objec i e and be e
han bin a leas one objec i e. Fo mally, in case o minimiza ion i eads
adomina es b⇔ ∀i: i(a)≤ i(b)∧ ∃j: j(a)< j(b)
o a i ness unc ion o mul iple objec i es, (x) := ( 1(x), 2(x), . . . )Based on
his concep , an op imiza ion p ocess sea ches o solu ions ha a e no domina ed by
any o he solu ion. This esul s in o a se o non-domina ed solu ions, called he Pa e o
on . The pe o mance o an op imiza ion p ocess can he e o e only be exp essed in
ela ion o a se o solu ion, a he han he quali y o a single bes solu ion.
E olu iona y Algo i hms (EA) ha e become a s anda d ool o sol ing MOO p oblems.
These algo i hms a e based on se s o solu ions. This coincides well wi h he challenge
o inding a se o solu ions in MOO p oblems. Op imiza ion Algo i hms (EMOA) a e
mode n MOO echniques ha op imize he space ha is co e ed by a Pa e o on wi h
espec o a p ede ined e e ence poin . Maximizing his space, also called he hype ol-
ume, pushes solu ions mo e and mo e owa ds he desi ed objec i e alues. Mo eo e ,
he hype olume ewa ds a high di e si y o solu ions, i.e. a wide sp ead, and a smoo h
dis ibu ion o solu ions along he bo de o he non-domina ed a ea. All hese p op-
e ies a e highly app ecia ed in MOO. One o he echniques employing hype olume
maximiza ion is he SMS-EMOA (c . Beume e al. [10]), which is also employed he e.
2.5 Expensi e Op imiza ion P oblems
In gene al, indus ial design asks can no be op imized by manu ac u ing mul iple de-
sign ins ances and deciding o he bes al e na i e a e wa ds. In almos all cases, his
Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 7
p ocedu e would simply be oo expensi e. As a consequence, models a e conside ed
o es ima e he pe o mance o di e en designs be o e he ac ual manu ac u ing p o-
cess. These simula ions o models hemsel es can become ime-consuming o e alua e.
De eloping su oga e app oaches is he mos impo an solu ion o ha issue. In such
app oaches, he op imiza ion p oblem (e.g. he cyclone model) is eplaced by a cheape ,
o easie o op imize su oga e model. A comp ehensi e su ey o su oga e modeling
in op imiza ion was p o ided by Jin [11].
A me hodical amewo k o su oga e model based op imiza ion o noisy and de e -
minis ic p oblems is Sequen ial Pa ame e Op imiza ion (SPO) in oduced by Ba z-
Beiels ein e al. [12]. SPO has been de eloped o sol ing expensi e algo i hm uning
p oblems bu can be di ec ly employed o sol ing eal wo ld enginee ing p oblems as
well.
One o he mos o en used su oga e-models is K iging. This is pa ly due o he ac
ha i poses an excellen p edic o o smoo h, con inuous p oblem landscapes. Mo e-
o e , i p o ides an unce ain y es ima e o i s own p edic ion, which can be used o cal-
cula e he Expec ed Imp o emen (EI) o a solu ion. This was used in E icien Global
Op imiza ion by Jones e al. [13] o balance exploi a ion and explo a ion in he op i-
miza ion p ocess.
In MOO, se e al app oaches employ su oga e modeling. An o e iew o su oga e
modeling in MCO is gi en by Knowles and Nakayama [14]. EGO has also been ex-
ended o MOO p oblems, as in he Pa EGO Algo i hm by Knowles [15] o he SMS-
EGO app oach sugges ed by Ponweise e al. [16]. Emme ich e al. [17] show how he
EI in hype olume can be calcula ed exac ly.
3 Ques ions and Goals
Real-wo ld indus ial mul i-objec i e es cases a e highly app ecia ed by he MOO e-
sea ch communi y because hese allow o a compa ison o me hods apa om a i icial
es cases. The la e a e usually used in he communi y bu do no yield he complexi y
o signi icance o eal-wo ld applica ions. The COP is p esen ed as a MOO es case and
i s esul s a e p esen ed. Mo eo e , he MOO esul s a e compa ed o esul s om a
single objec i e op imiza ion app oach o de e mine a possible lack o pe o mance due
o in ol ing mul iple objec i es in pa allel. The e o e, he single objec i e op imiza ion
esul s a e compa ed o he ex eme Pa e o non-dominan solu ions.
The cyclone model assumes a ce ain dis ibu ion o pa icle sizes as well as ixed,
undis u bed se ings o all o he ele an a iables. In p ac ice, a iables like he in low
speed o pa icle sizes will be noisy. Tha noise can be simula ed by epea ed sampling
om andom dis ibu ions, each sample leading o a new e alua ion o he cyclone
model. The epea ed e alua ion wi h di e en samples leads o a mo e ime consuming
a ge unc ion o he op imiza ion algo i hm. To alle ia e his issue, su oga e models
can be in oduced o suppo he op imiza ion p ocess. The e o e, he goals o his s udy
a e o:
– es he de e minis ic COP as a MOO p oblem.
–compa e a single-objec i e and a mul i-objec i e app oach.
8 B eide ho e al.
–analyze he in luence o noise on he solu ion quali y.
–de e mine in luence o geome y, luid and pa icle pa ame e s.
–compa e a model- ee and a model-suppo ed SMS-EMOA in case o op imizing a
s ochas ic cyclone simula ion.
4 P oblem Se up
4.1 Pa ame e s o he De e minis ic Cyclone Model
The he ein desc ibed expe imen s a e based on an example by Lö le [1]. This exam-
ple uses he geome ical, luid-speci ic, and pa icle-speci ic pa ame e s summa ized in
Table 1. The geome ical pa ame e s o be op imized a e a ied in ixed bounda ies,
which a e ±10 % o he de aul alues om Table 1. Geome ical pa ame e s could be
a ied in a much wide ange, howe e , his ange would no necessa ily be i ing o
he gi en luid and pa icle pa ame e s. The e o e, he 10% de ia ion was chosen as a
ypical expe imen al se up. A wo-dimensional case (wi h Daand Honly) as well as a
se en dimensional case wi h all se en pa ame e s a e in es iga ed.
4.2 S ochas ic Cyclone Simula ion
In addi ion o he pa ame e s om he de e minis ic model, h ee noise sou ces a e
in oduced. Tha is, low a e Vp, pa icle densi y ρp, and he pa icle sizes xiused in
he collec ion e iciency calcula ion can be subjec o noise. The low a e is allowed o
a y in be ween ±10% o he de aul alue, while pa icle densi y a ies wi hin ±5%.
In bo h cases, a uni o m dis ibu ion is used. Fo he collec ion e iciency calcula ion,
one sample is d awn in each o he in e als om Table 2, ins ead o simply using he
mids o each in e al. Tha is, he pa icle size in each in e al is de e mined andomly
wi h a uni o m dis ibu ion. These alues a e hen inse ed in Eq. (3), o calcula e he
o e all collec ion e iciency.
Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 15
ime es ic ion makes pu e exploi a ion he mo e desi able choice. This esul in o a
un- ime o oughly 210 seconds, consis en wi h he model- ee un- ime. He e, he
main con ibu o s o un- ime a e he aining o he K iging model, and he subsequen
op imiza ion on he K iging model.
Bo h he SMS-EMOA and he K iging-suppo ed SMS-EMOA a e un 20 imes. The
quali y o each poin in he esul ing Pa e o on es ima es a e alida ed by 10,000
uns o he s ochas ic simula ion. To compa e, he hype olume indica o is used wi h a
e e ence poin o 5000 PL and ze o CE.
Resul s om S ochas ic Op imiza ion Runs The op imiza ion esul s a e summa-
ized in Fig. 10. Fo he model- ee SMS-EMOA, inc easing he numbe o e alua ions
pe poin seems o yield no imp o emen . Explo ing mo e poin s is a leas as p o i able
as es ima ing he quali y o each poin mo e accu a ely. S ill, i can be seen, ha he
model suppo ed SMS-EMOA clea ly ou pe o ms he model- ee SMS-EMOA. This
is despi e o he ac , ha he model uses a much smalle budge han he SMS-EMOA.
The un- ime o bo h app oaches is abou equal. The K iging model seems o deal well
wi h he emaining noise in he objec i e unc ion, yielding a mo e easy o op imize
su oga e. This is especially in e es ing since he a ge - unc ion is no exac ly expen-
si e, which would be he usual case whe e K iging su oga es wo k well. S ill, 1000
e alua ions o each poin make his a semi-expensi e p oblem. One explana ion o he
●●
●●
● ●●
SMS−EMOA, 1 e alua ion
SMS−EMOA, 10 e alua ions
SMS−EMOA, 100 e alua ions
SMS−EMOA, 1000 e alua ions
K ig+SMS−EMOA, 1000 e alua ions
3120 3140 3160 3180 3200 3220 3240
Hype olume
Fig. 10: Boxplo s o he op imiza ion expe imen s wi h he s ochas ic simula ion. The x-axis
shows hype olume o he es ima ed on s, alida ed by 10,000 uns o each poin . Highe al-
ues a e be e . E alua ion numbe s e e o he epea ed e alua ion o each poin .
excellen pe o mance o he model-based SMS-EMOA, besides i s abili y o smoo hen
he noisy landscape, may be, ha e en in he se en-dimensional case, he ue on is
sp ead along se e al bounda ies o he decision space. This could al eady be obse ed
o he de e minis ic model, and holds he e as well. Tha may lead o a si ua ion, whe e
he su oga e model does no ha e o ha e high accu acy o e he whole sea ch space
because i su ices o p edic dec easing alues owa ds he bounda ies.
7 Summa y and Discussion
This pape p esen s a mul i objec i e op imiza ion p oblem, based on a de e minis ic,
analy ical model o a cyclone dus sepa a o . Noise in luence was added, hus gene a ing

16 B eide ho e al.
a s ochas ic simula ion model. This allowed no only o op imize, bu also o in es iga e
obus ness o solu ions, agains unce ain y in he noisy a iables.
In case o he de e minis ic model, use s can choose p e e able esul s om he Pa e o
on , depending on hei p e e ence o CE and PL combina ions. P e e ed design
poin s can be u he analyzed wi h he s ochas ic simula ion model. Fo ins ance, a
use migh choose he knee o he Pa e o on (see igh plo in Fig. 9) and an ex eme
poin on he uppe le pa o he Pa e o on , as summa ized in Table 3. Those se ings
could be ee alua ed wi h he s ochas ic model, yielding esul ing a iance es ima es
as depic ed in Fig. 11. In p ac ice, i may occu , ha a use would p e e he ex eme
solu ion. While his solu ion has a wo se expec ed alue o CE, he e is s ong o e lap
be ween bo h solu ions. On he o he hand, he PL alues show clea ly a signi ican di -
e ence, hus leading o he po en ial p e e ence o he ex eme poin .
I was also shown, ha he s ochas ic simula ion can be op imized di ec ly. While he
Table 3: Two Poin s om he Pa e o on ound by SMS-EMOA on he de e minis ic model, wi h
all se en geome ical pa ame e s conside ed. The whole on is shown in Fig. 9 on he igh .
Pa ame e Poin 1 (knee) Poin 2 (ex eme)
Da1134 1134
H 2750 2750
D 462 462
H 576 576
13.92 12.81
He540 660
Be180 220
PL 2103.87 1375.90
CE -0.921 -0.86
high noise le el makes his mo e cos ly o a simple SMS-EMOA, a su oga e model
based app oach seems o handle he issue mo e e icien ly.
The classical, single objec i e Nelde -Mead is able o iden i y he op ima o he de e -
minis ic objec i e, bu ails in case o he noisy simula ion.
The p esen ed cyclone model ep esen a ions a e o compa a i ly simple s uc u e, and
hus a e good candida es o eal-wo ld based mul i objec i e es p oblems.
Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 17
Knee
Ex eme
1200 1400 1600 1800 2000 2200 2400 2600
P essu e Loss
●●●●● ●● ●●● ●● ● ●● ●● ●● ●●●●●● ● ● ●●● ●●●● ●●
●● ●● ●● ●● ● ●● ●●● ● ● ●● ● ●●● ● ● ●●● ●● ●● ●●● ● ● ●● ●●● ●● ●● ●●●●●● ●● ●●●
Knee
Ex eme
−0.9 −0.8 −0.7 −0.6
Collec ion E iciency
Fig. 11: Boxplo s o he knee, and an ex eme poin o he Pa e o on ound o he de e minis ic
model. Co esponding o he pa ame e s in Table 3, ee alua ed 1000 imes wi h he s ochas ic
simula ion. Lowe alues a e be e .
8 Ou look
While he analy ical cyclone model does pose an in e es ing MOO p oblem, i lacks any
in o ma ion abou quali y o non-s anda d p oblems and solu ions. Tha means, when-
e e pa icle, luid o geome y a ibu es s ay o a om he s anda d, he models
quali y de e io a es. Fo ins ance, he model is unable o ep esen non-cen ic posi-
ions o he ou le o slan ed inle s. S ill, such changes o geome y a e o high in e es
o p ac i ione s in indus y. To ge a be e quali y es ima e o hese geome ies, CFD
models a e used.
CFD models o e a wide a ie y o model he dynamics o pa icles in any kind o low.
Howe e , such models need di e en p elimina ies like he disc e iza ion o he consid-
e ed space (meshing) and a sol e o he esul ing se o (pa ial) di e en ial equa ions.
This esul s in a he ime-consuming and hus expensi e simula ions o each design
al e na i e. Howe e , such simula ions can be e y p ecise, mapping he eal p ocess
wi h a e y high accu acy and hus migh be wo h he e o .
To suppo he op imiza ion o such ime consuming simula ions, he analy ical model
may s ill be o use. I can be used o mul i- ideli y op imiza ion o such CFD models,
using echniques like Co-K iging [20]. While hey ep esen only a pa o he possi-
ble numbe o geome ical pa ame e s, hey can s ill be used o imp o e he quali y o
such a Co-K iging model, which would o he wise ha e o ely on accu a e, bu spa se,
CFD simula ions only. Fu he mo e, he op ima o Pa e o on s ound on he analy ical
model can be used o gene a e s a ing poin s o he op imiza ion o he mo e complex
CFD models.
Acknowledgemen s
This wo k has been pa ially suppo ed by he Fede al Minis y o Educa ion and Re-
sea ch (BMBF) unde he g an s MCIOP (FKZ 17N0311) and CIMO (FKZ 17002X11).
18 B eide ho e al.
Re e ences
1. Lö le , F.: S aubabscheiden. Leh buch eihe Chemieingenieu wesen, Ve ah ens echnik.
Thieme. ISBN 9783137122012. 1988.
2. Ra i, G.; Gup a, S. K.; Ray, M. B.: Mul iobjec i e Op imiza ion o Cyclone Sepa a o s Using
Gene ic Algo i hm. Indus ial & Enginee ing Chemis y Resea ch 39 (2000) 11, S. 4272–
4286.
3. Elsayed, K.; Laco , C.: Op imiza ion o he cyclone sepa a o geome y o minimum p es-
su e d op using ma hema ical models and {CFD} simula ions. Chemical Enginee ing Sci-
ence 65 (2010) 22, S. 6048 – 6058.
4. Ba h, W.: Be echnung und Auslegung on Zyklonabscheide n au G und neue e Un e -
suchungen. B enns o -Wä me-K a 8 (1956) 1, S. 1–9.
5. Pishbin, S. I.; Moghiman, M.: Op imiza ion o Cyclone Sepa a o s Using Gene ic Algo i hm.
In e na ional Re iew o Chemical Enginee ing 2 (2010) 6, S. 683–691.
6. Elsayed, K.; Laco , C.: Modeling and Pa e o op imiza ion o gas cyclone sepa a o pe o -
mance using {RBF} ype a i icial neu al ne wo ks and gene ic algo i hms. Powde Technol-
ogy 217 (2012) 0, S. 84 – 99.
7. Sa ikhani, H.; Hajiloo, A.; Ranjba , M.: Modeling and mul i-objec i e op imiza ion o cy-
clone sepa a o s using {CFD} and gene ic algo i hms. Compu e s & Chemical Enginee ing
35 (2011) 6, S. 1064 – 1071.
8. Muschelknau z, E.: V -Hochschulku s 2 [zwei], mechanische Ve ah ens echnik. Ve ah en-
s echnik in e na ional ; Sonde h. K ausskop . 1972.
9. Co es, C. Gil, A.: Modeling he gas and pa icle low inside cyclone sepe a o s. P og ess in
Ene gy and Combus ion Science 33 (2007), S. 409–452.
10. Beume, N.; Naujoks, B.; Emme ich, M.: SMS-EMOA: Mul iobjec i e selec ion based on
domina ed hype olume. Eu opean Jou nal o Ope a ional Resea ch 181 (2007) 3, S. 1653–
1669.
11. Jin, Y.: A comp ehensi e su ey o i ness app oxima ion in e olu iona y compu a ion. So
Compu ing 9 (2005) 1, S. 3–12.
12. Ba z-Beiels ein, T.; Pa sopoulos, K. E.; V aha is, M. N.: Design and analysis o op imiza ion
algo i hms using compu a ional s a is ics. Applied Nume ical Analysis and Compu a ional
Ma hema ics (ANACM) 1 (2004) 2, S. 413–433.
13. Jones, D.; Schonlau, M.; Welch, W.: E icien Global Op imiza ion o Expensi e Black-Box
Func ions. Jou nal o Global Op imiza ion 13 (1998), S. 455–492.
14. Knowles, J. D.; Nakayama, H.: Me a-Modeling in Mul iobjec i e Op imiza ion. In: Mul i-
objec i e Op imiza ion, S. 245–284. Sp inge . 2008.
15. Knowles, J.: Pa EGO: A hyb id algo i hm wi h on-line landscape app oxima ion o expen-
si e mul iobjec i e op imiza ion p oblems. IEEE T ansac ions on E olu iona y Compu a ion
10 (2006) 1, S. 50–66.
16. Ponweise , W.; Wagne , T.; Bie mann, D.; Vincze, M.: Mul iobjec i e Op imiza ion on a
Limi ed Budge o E alua ions Using Model-Assis ed -Me ic Selec ion. In: PPSN, S. 784–
794. 2008.
17. Emme ich, M.; Deu z, A.; Klinkenbe g, J.: Hype olume-based expec ed imp o emen :
Mono onici y p ope ies and exac compu a ion. In: E olu iona y Compu a ion (CEC), 2011
IEEE Cong ess on, S. 2147–2154. IEEE. 2011.
18. Nelde , J. A.; Mead, R.: A Simplex Me hod o Func ion Minimiza ion. The Compu e
Jou nal 7 (1965) 4, S. 308–313.
19. Box, M. J.: A New Me hod o Cons ained Op imiza ion and a Compa ison Wi h O he
Me hods. The Compu e Jou nal 8 (1965) 1, S. 42–52.
Simula ion and Op imiza ion o Cyclone Dus Sepa a o s 19
20. Fo es e , A.; Sobes e , A.; Keane, A.: Enginee ing Design ia Su oga e Modelling. Wiley.
2008.
21. Lee, L. H.; Chew, E. P.; Teng, S.; Goldsman, D.: Op imal compu ing budge alloca ion o
mul i-objec i e simula ion models. In: Simula ion Con e ence, 2004. P oceedings o he
2004 Win e , Bd. 1, S. –594. 2004.
Kon ak /Imp essum
Diese Ve ¨o en lichungen e scheinen im Rahmen de Sch i en eihe ”CIplus”. Alle
Ve ¨o en lichungen diese Reihe k¨onnen un e
www.ciplus- esea ch.de
ode un e
h p://opus.bsz-bw.de/ hk/index.php?la=de
abge u en we den.
K¨oln, Janua 2012
He ausgebe / Edi o ship
P o . D . Thomas Ba z-Beiels ein,
P o . D . Wol gang Konen,
P o . D . Ho s S enzel,
D . Bo is Naujoks
Ins i u e o Compu e Science,
Facul y o Compu e Science and Enginee ing Science,
Cologne Uni e si y o Applied Sciences,
S einm¨ulle allee 1,
51643 Gumme sbach
u l: www.ciplus- esea ch.de
Sch i lei ung und Ansp echpa ne / Con ac edi o ’s o ice
P o . D . Thomas Ba z-Beiels ein,
Ins i u e o Compu e Science,
Facul y o Compu e Science and Enginee ing Science,
Cologne Uni e si y o Applied Sciences,
S einm¨ulle allee 1, 51643 Gumme sbach
phone: +49 2261 8196 6391
u l: h p://www.gm. h-koeln.de/~ba z/
eMail: homas.ba z-beiels ein@ h-koeln.de
ISSN (online) 2194-2870