Академический Документы
Профессиональный Документы
Культура Документы
-. _Er!4r+\r. . 5@
@
eNIS S :ICOUdNOE UV CO UO ^H
ffi
IN eNtHSSNteNs
0r forne
I I
HOI srNt-I=ffng
voorl
NOS HV COUO^HÄ q p e q s1 lq n 6
.':'
".,,i;,""";',,
,. :,:r
00t ocuPr.rllollod lossalduo3 of t{ceojddv}sag
86 u6rso6 uorldrosqv ulnlurldo e.rn6rl
? 6 : 6 u r c u e ; e gd o o l Ä l 1 p n p a r i r l d r u r g
06 urelsÄSrotp^ burloog lsag u6rsaq
B8 ' ^ Ä e Ml o r s e 3u e p e ; n b r l x n l l a u u n u r u r y \
,8 DOleuV^q sJOlElaOO uollPls la/v\odulell
8L s d n y e l g e ; e g s l l e d gu o r l p l n u r s6 o 1 e u y
ZL lulod paol uorlellrlslg tsag alnduo3
99 elpC u/v\Osll soleraueg uel0or6 uotlelltlsrCstql
09 e1cÄ3uorleleue0egIEC u.rnurldgpurl
99 relndiuo3 Oo1euyqlrM JolcEoUuotlelnoltcagolelnr!tS ol MoH
h
s:e1ndulo36oleuy lnoqv ureal
'op e6e6
SINfINOC IO ]I8V1
suStndy\loceotvNV
LearnAbout
l 4
l
I 9NtSSfCOUd Noeuvcouc^H t^loul c3l_Ntud3u
ur ueql Jeql"r uorlnlos lcexe eql ol sno8oleu€ sr qlrq1v\ slecxe relndwoc So1eueer{J 'srs€qsnonurluoo B uo uoll
rrlroJ snonurluoJ ur pelueseJd sr Jalndtuoc 8o1eue uz -e"r8a1urSururro;-red firlrqedec eqtr sr <Je^e^roq,8o1eue
;o
ruo{ uorlnlos er{J 'J€eurluou ro r€ourl ere Äaql .reqlaq.tr eqt Jo ernl€eJ luecgru8rs lsour er{I 'c1e 'uorlerlueuod
;o sselp:e8er suorlenbe Ierlueregrp Jo uoqnlos eql ur -xe 'urqlr"re8ol eql 3ur1e1 'Sur.ienbs se qcns tsuorle.redo
uorlounJ fue"r1rq.resnonurluoJ s" IIe.&\se suorle.ledo crleur
-t{llr" aq} epnlcur paurro;.red eq usr qcrr{^{ suor}"1nduroc
ta3_aJau eq; 'dlsnonulluor pu" 1e11e.red ur suorlelnduroc sur^ro;rad
'relnduloc Sopue aqa
IetFIp eql 01 ls"rluoc d:eqs ur
'1eg;la;o
'uorlzcrldde e8uer slr pualxa
cr|z'0 ;o
d1qr1es.re.rSurdueduroccz
N0u3Nru
tvrJNSN0dxSreqtrnJ .talnduroc lelflp eql Jo
el{l pu€ anbruqcal IBOrJeurnN
'Jncco senle^ aleJcsrpeJeq./v\
/ lau \eA^, I ''c1a 'suor1e1nc1ec a1e1s-Äpea1s 'suorlenba crerqa8le
\räbol/""'rau- Jo uop
-n1os 'scr1sr1e1s 'uorlcnpar €l"p ur uorlecrldde peardsepr^{
/ lau \01^^.
l-l uvl --
z ls
. 'la8 pug s;elnduroc 'uollslnrl"c
ua00l/ IetlSl61 Lraaa "ro; ua.Lr8aq
'e8zn5uel pezrlercedsd-ra.l,e ur dlle.reue8'suorlcn.rlsur
lsnur
'uorl"Jedo asrcardL.raae sr .ralndtuoc
hrHJ.ruveor pelre1e1l lelr8rp e 8ur
-urer8or4 'pesn eq trsntusenl"A elercsrp Jequnu eF"re1z
Jo
1 3 , r{crq^. uI suralqo.rd.roy pa.lrnba-raq deur eunl elqereprsuoc
aa(lau)' 'd11er1uanbas euop er€ suo4"lncl"J pu" pesn eJB sanl?^
o+la Z g
ls elercsrp dluo acurs ',rena,uo11'surelqo.rdJo sselc epr.4r^Je^
zaala - l3
?4'rüj e 'c1e 'uor1e.r8e1ur'.ro.r.ra-pu"1el'r1
;o Suqpueq eql slrulred
.ro; senbruqcetrIscueurnu äql qlr^\ paldnoc ueq^\ qrrr{^4
N 0 r s t ^ r 0 Iool 1euorlelnduroc elqesra^ lsoru sr se4rlrc€J s,lelr8rp
"
ar{l Jo Unser erlJ 'rer{to aql u€ql ral"er8 ro u"q1 ssel
Z6 'o1
lznba sr euo JeqleqartSurprcep pu€ senl€A eleJJsrpoml
Sur.redurocÄq suorsrcap1zcr3o1e{eur ol pasn eq u"c relnd
-uror lelr8rp eqtr suorl"redo crleuqlrre 01 uolllpp€ uI
lg
'rot"lncler >1sapprde.r dlaurerlxe pue elq€r1arLrer, z
N0tlvctld u.tnn Jo luel?^rnba eql esuese ur sr relnduroc I€113Ipeql snr{I
'euqceur .rz1nc4:ed due uo rornlc"Jnu"ru eqr dq pexg
d11eraua8sr qrrr{rr uorsrca.rd;o eer8ap qBH otr peuuoJred
eq u€J suorleredo Jrleuqlrr€ eql (re^oerotrI "'senlel elerrsrp
oo-1
pIzszs4
t.tc t- =€a €g Jo Jequnu ssallturl lsowl€ u" Jo uorl€rado eql pue Jo
] eFerols eql sllurred qcrq.tr dlrcedec d;oureur e8rel e qlr.l,r
cr-r{$+{}]J-+a paddmbe eq u€c sretnduor I€lFICi 'l.1prder Lre,r peqsqd
-urocc€eq uec senl€^ elercsrp Ä1uoJo uorsr^rp pue uorl
N O I I V U 9 3 T N I -"crldrllnu'uorlcerlqns'uorlrppe
Jo suorl€JadocrleruqlrJ€
eqt 1€r{1 ?ltslratc€rerIr or{1 s"r{ .ralndruoc plfJp eqa
:,relnduroc plrq^q ro '8o1etre 'l"tFlp rer{}re s€ pe5rsselt
1 9 a e g 1 Z a z 3 gl s l g l - : t a ts 77 te{ ./-*r" aq deur s.ralndurocurepo1l 'srelnduo)
ft;2, to uoslloduro)
9r Zy lr \--'- ra
=n ' ' =' t l6 lC
ü ü="' o; 'serlr^rlJe Äzp-o1-Äep
83nnns srq ur esrJe deur Äeql se sanbruqcel uorlnlos /v\au pue
surelqo.rd o1 seqceo.rdde /v\eu eurur€xe ol pelelnurls eq
'paepul 'Älrxeldurocslr
IIIM eq Jo esn€cequrelqo"rdenbrun
s O lE 3 € Jo qsruq Äpq8r1 trou llrlv\ .raeur8ue sql 'uorlecr1dde
I a o= ? a
;o senbruqeelaq] Jo s" IIe.t\ se s;alnduroo;o dlrcedec pue
8 3 l i r ' l d l l v N r v 9H 0 l l {
l;X T;T.z. r-za
l u / z u=x
'a.----+LLr.
."il '! '.{nrl_ r3
231 1?s-131y=9s 6ol"rv
sro+nduro)
( 0 3 0N n 0 8 9 N nu) o l v n N 3 l l . v
| -rX
lu/uu=x T-9-f
l a x=? a -'!i'L ,.-O-'.
Za"----'f l 1" ,^
uota X
( c 3 0 N n o uu9o) l v n N 3 l l v
slueuodurotr ralndurol 6olpuy-l llgyl
...
TEARNABOUTANATOG COMPUTERS Although it is not shown in Table 1, one of the most
importänt components of the computer is the constant
terms of discrete values which can only approximate the voltage DC power supply. Depending upon the design of
exact solution. Furthermore, the computationscan be car- the computer, the power supply is usually either 100 or
ried out at practically any speed desired. This permits 10 volts. This DC voltage is called the reference or ma-
the rapid exämination of a wide range of parameters in chine voltage and its stability will determine to a large
u rpu"ä of time that can be significantly shorter than the degree the äccuracy of the solution that one obtains'
fixed time of calculation on the digital. Analogs do not, The first element in Table 1 is an attenuator com-
however, have the capacity for memory, the capacity for monly called a pot-an abbreviation for potentiometer'
a large number of operations, nor the degree of precision Two types of päts are commonly found in an analog
of the digital. .o-pni".. One is referred to as a grounded pot and the
other an ungrounded pot' The grounded pot enablesthe
Programing an analog does not require knowledge
operator to öbtuin an outPut voltage which is some frac-
of a hlghly specialized language as does the digital. The
ti,on between 0 and 1 of the input voltage' The un-
details äf programing are generally quite similar to classi-
grounded pot performs the function shown in Table 1
cal methods of solution with which the engineer is famil-
änd is generally used in the construction of special func-
iar. Analog computers find their greatest application in
tion circuits. Attenuators are usually adjusted manually
the solution of differential equations, such as occur in the
to set the value of a constant K. This value is determined
dynamic analysis of processing systems,in which the
by measuring the output voltage from an attenuator with
arnount of algebraic and logical operations required is
the voltage measuring device built into the computer'
rather limited.
The next component listed in Table 1 is the high gain
A hybrid computer is the combination of a digital and amplifier. The function of the high gain amplifier is -to
un unälo.g computer in which the solution of a problem
-between muitiply an input voltage by a large constant, generally
is shared the two. By containing both an analog of thä order o1 10s. The high gain amplifier is seldom
and a digital, the solution of a problem on the hybrid used by itself, but usually forms an integral part of -other
computer can incorporate the advantages of both' The .o-*on analog components. The high gain amplifier is
digital computer portion provides a Iarge memory capac- generally designld so that it provides very stable operation
ity" and peimits rapid logical and algebraic operations t th" -ugttitude of the output voltage does not exceed
*trite tfte analog permits continuousintegration' The the comprfter reference voltage. If the output voltage of
transfer of infoÄition from one portion to the other and the high gain amplifier exceeds the machine voltage the
back again extends the versatility of the hybrid beyond operatlo.t"of the iomputer will be non-linear' For this
that of the analog or the digital alone. Hybrid comPuters ,öu.o.,, the output of a high gain amplifier during oPera-
are finding application in the solution of complex prob- tion should not exceed that of the computer reference
lems wherä thä time required for a digital solution is ex- voltage. Most computers have a built in alarm system
cessiveand where analogs do not possess the capacity or whicü warns the operator when the output of an ampli-
the precision required for a solution. fier exceedsmachine voltage. This restriction on the out-
put voltage holds for .o*ponent in which a high
The term analog in "n"ry
Analog Computer Components. gain ampiifier aPPears.
analogiomputer is really a misnomer. That is, the electri- The next component of interest in Table 1 is the sum-
problem is not an
cal ciicuitry used to solve a particular ming ,circuit. e'i its name implies, the- summing circuit
electrical analog in the conventional sensethat electrical adds two or more voltages and gives the resulting su-m'
current is analogous to fluid flow, voltage is analogous to The summing circuit cönsists of a high gain amplifier
pressure, electrical capacitance is analogous to mass, etc' with a feedbaik resistorand three or more input resistors
The electrical analog iomputer is actually a device which The input resistors and the feedback resistor, in a sum-
has been designed to perform certain mathematical opera- ming circuit, are usually fixed not variable- resistors' The
tions, such as addition, multiplication, integration, etc' on inpu't voltage to a summing circuit is multiplied by. the
specified voltages. The use of an analog computer conse- of the feedback resistor
-the re tulio of the resistance
neeati
q.rentty involvis the specification and completion oj the to resistanceof the input resistor. For most comPuters
Äathematical operations required to solve a particular the value of this ratio is usually 0.1, 1 or 10' Since a sum-
set of equations and is entirely independent of the fact mer contains a high gain amplifier, the output from a
that the equations may describe a flow system, a heat summing circuit rnust not exceed the reference voltage'
exchangeproblem, a mass transfer problem, or a reactor One äf t-he most important components of an analog
system. computer is,the integrator. The circuit fo.r an integrator
Since most of the analog comPuters being used today is essentiallya high gäin amplifier equipped with a capaci-
discussion is restricted to this type
are transistorized, this tor in the ieedbäck-line and two or more input resistors
of analog computer' This is not a large restriction because The output from an integrator is -the -negative sum of
most of ihe older vacuum tube computers operate in the the initiäl condition and the integral with respect to com-
samefashion with the exception of a few mechanical com- puter operating time of fixed ratios of the input voltages
ponents. äs indicäted in Table 1. Since the output of the integrator
A list of the components found in a modern electronic is time varying, there is a possibility that this value may
analog computer is given in Table 1. The first column be greater ihaä machine ,röltage at. sometime during the
in
'IiUte 1 contains a symbolic representation of each pro6l"rn. Therefore, special precautions must be taken to
component, the second column contains an electrical rep- insure that this does not happen' These precautions will
,eserrtutio.tof each component and the third column shows be discussedlater in the series.
the mathematical operalion that each component performs' The componentsdiscussedto this point are commonly
6
cailed linear components and the principle of superposi-
:ion holds. The next group of componentsto be discussed
a:'e commonly called non-linear components.
As an example of how an analog computer circuit The final diagram with pot settings is given in Fig' 4'
may be set up we will consider the case of the change Although the arialog diagram aPpearssatisfactory, it may
in üeight of liquid in a tank having a cross-sectionatea A' in<ieed-beunsatisfactory for one or both of two rearsons'
A mass balance on this systemyields: . lffr" output from'a high gain amplifi-er may exceed
the referenc" ,roltage and hence yield a f-aulty solution'
Accumulation- InPut - OutPut (2)
o The output föm a high gain amplifier may be so
(3) in
A (dh/dt) : et- Qz small that inierent limitatiäns-of the analog results-
of u fu.ttty solution or difficulty is encountered in monitoring
We will assume that Qt is constant and that the rate
the
- value.
fto a knowledge of the nature of the numerical solu-
About the qulhors tion, the output, hi fro amplifier 2-has an initial value
Tsnonoen W. ClolraN is assistant pro- of + a final'value of L0 v', and asymptotically uP-
".,
iiito, 'i'n the clzemical engineering d'e-
'partment orou.h"r'th" final value in an exponential fashion' Hence
fOO v. or a 10 v. machine, amplifier 2 will not be
(JniaersitE of Maryland'
'CoUese of i".
Park, Md.' where he teaches " dhf dt
oraduate and undergraduate courses
overloaded. Similarly amplifier 1 with the output
value of 0'6
in oro""r, control, separation processes, rvill not be overloadeäb".ä,,t" it has an initial
Ilowever,.for a 100 v'
mathematical modeling, and computer and decreasesmontonically to 0'
applications. Dr. Cadman also conducts änd' dhldt ate quite small and
machine, the values of. h
rii"arch in process contt^ol, computer the cir-
for a 10 v. machine dhldt is quite-small' Since
applications, and process modeling' Be- the solution will prob-
cuit uses no nonlinear .o*pott"tttt,
fiie receiuing h-ry !h. D-. .desr.ee in difficult'
(J
clwmical engineering f r om Carnegie-MeUon- ni'aersi'tg' he ;üiu il satisfactory although monitoring 1n?y !9
-"i1;;;iJ
with St' JosephLead Co'' Allied
i"A""iiiri-t""r'* Trortrtiont be noied that;if the value of the highest.de-
öi"*liät Corp.,'and Gutf Resert"t"ch and Deuelopment.Co' rivative does not need to be available for monitoring'
of AAAS',ACS, ISA, AICLE,NSPE' Si'sma diagram
i;-;;;-*"*bLr' step 2 aborremay be eliminated' In this case,the
Xi and Tau Beta Pi. chosen example
;h;;' t" nig. 5 is obtained for the
Tuoooone G. Snrrtn i's associ,ate pro- with a net reJuction in the number of analog components
essor in the chemi cal engineering
f'ilepartment required for a solution.
of [JniuersitA of Maryland', be pre-
in the next article of this series,techniques will
Coltege Park, Md' He conducts re- to be obtained
,noril, anil instructs undergraduate ,"nr"a which permit an analog solution
direct solu-
anil grad'uate courses on polEmet' phgs- rrfr"" ,t aforementioned reasons prevent a
ics, ieactor desi'gn, and mass t'r'a'nsfer' " by which satisfactory computer variables
ii.". fn" process
Dr. SrwitlL receiaed B.S. and M'S' de- so that the outputs from amplifiers are main-
orees in chemical engineering from "r"-.nit"ti voltage' is
tlniuersitg and a D"'Sc'
iohns Hopkins Uniueriitg
iolms D.'Sc' i.l.r"a near, but do not "*ce"d the reference
degree ii ehemi,cal enginee'r'ing from
in ch.emical termed amPlitude scaling.
Wäshington (Jniaersitg. He has tt;orked
in tlw-plastics
'and department of E'-!' du-P^2nj de Nemou't's :e1'"*
,-n'TH:,i"ii}:ru-,''*$i1".ä,;19,"'?äJ*lL
:li;tilf,
A Co. 'i's a' member of ACS, AICILE, A A A S and
Sigma XI.
LearnAbout Analog Computers
PART2: AMPLITUDE SCALING-The amplifiers are maintained near, but do not exceed, the
reference voltage. This process is termed amplitude scal-
outputs from high gain amplifiers irg.
**"*"1**ft1
*%"*,Y;
i+r
T A N KI T A N K2
1e$.{ryj
-j:-,;tfl
ärli
iä'
ffi
e lr:
i ..:-
..!
Fig. 6-Unity scaling is applied to an engineering problem before it is solved on the analog computer.
LEARNABOUTANA,IOGCOMPUTERS
. . . variable by its estimated maximum magnitude. Note theie
on the list made in Step 3.
lABl,E 2-Scollng Amplifier Output3 ond Potentiometer Seftlngs Step 6. Insert additional pots in the amplifier input
lines and/or modify the values of the pots which were
Output Estimated
Scaled needed to complete Step 2 so as to compensate for the
Component or Maximum
Value Value
Value change in scale from one amplifier to the next in series.
Again modify the list completed in Step 3.
Amp I hL ht* ha/hJ Step 7. Sketch the completed analog diagram using the
Amp 2 _h h x -hJhax
normalized variables and the new pot settings. Note that
Amp 3 h:z h * h2/h2* this step is easily accomplished by referring to the pre-
Amp 4 -h2 h"* -h2/h2* liminary sketch and to the list rvhich has been completed.
Pot I fr'(o) h'(0) /h'*
If the programer closely follows these rules and system-
Pot 2 q1/Ar qr/Arh:
atically lists each variable and the pot settings during the
Pot 3 l/AaR1 | /aaR1
scaling process, scaling can be readily achieved, errors
Pot 4 h,"(0) h'(0) /h,* easily detected, and any further scaling rapidly completed.
Pot 5 q2/A" q2/42h2* In addition, a completely general analog computer pre
r f r 1 l 1 [ r , r l gram can be obtained by syrnbolically retaining the maxi-
4 L r y- & J
Pot 6 _ t _ I _ l
A2L\' Rz) mum magnitude of the problem variables rather than
Pot 7 -7n
I
(#) (#) computing their numerical values. The following exam-
ple illustrates the use of this seven-step procedure for
Pot B
ZF,
I
(#) (#-) achieving unity scaling.
Fig. 7-This analog circuit representsthe variables in the Fig. 8--Amplitude scaling determinesthe proper setting for
two-tankexampleproblem. the amplifiersand potentiometers.
with initial conditions of /z' (0) and hz (0) respectively. have beenavailableand an outPut of.(h' - hr) I &' - hr)*,
Assuming that gr and g. are positive constants,a pre- where (ä, - hr)* * hrx or /r2*, would have been the
liminary analog diagram for solving Equations ( 10) and scaled output from the summer. Additional Pots, not
found necessary in drarving a prelirninary diagram, would
( 11) for h, and ä2 versus time can be drarvn. Prior to
have been needed in the input lines to compensate for
specification of numerical values, the diagram in Fig. 7
and the list of variables which are outputs from amplifiers the change in scale factors.
and the pot settings in Table 2 are obtained. Choosing the numerical values given in Table 3, a nu-
Letting a superscript* indicate the maximum value of nrerical solution can be obtained on the analog computer
once /i,* and,h2x are estinated. Thcse may be estimated
the output from the amplifiers as indicated in Table 2,
the protess of unity scaling can be symbolically carried by calculating the final valucs of /r, and /2, at equilibrium,
to completion. The results obtained are given in the final noting that these values are larger than the initial condi-
colunn of Table 2. The output fron each amplifier be- tions, and conscrvatively choosing maximums which are
comes the value of the problem variable divided by its about nvice the steady state values to account for the
maximum value and some of the values of the Pots are possibility of oscillation before the final values are ar'
changed to compensate in the change of scale from one tained. The numerical values of the variables and the pot
amplifier to the next. The latter changes perhapl become settings for this numerical example are given in Table 4'
moie evident if the final analog diagram given in Fig. B It should be noted that the analog diagram for the
is considered. foregoing example does not include physical constraints
In readjusting pot values follou'ing the scaling of out- which may be imposed on the actual situation. Conse-
puts from amplifiers, four casesarise: quently, for certain Pararneter values, the analog solu-
Case 1. The pot value must be divided by a maximum
expected value. Examples are pots 1, 2' 4 and 5 wherein Used for the Exompte PrirtRnre
TABLE 3-Porometers
the maximum value is included so that the inputs to the
amplifier contain the same scale factor as do the outputs' Physical Parameter Numerical Value
Case 2. The pot value remains unchanged. Examples ar 2 sq. ft.
are pots 3 and 6 which remain unchanged because the A2 4 sq. ft.
input to the pot contains the same scale factor as is de- Qr B cu. ft.,/min.
sired in the outPut. Qz 6 cu. ft.,/min.
2 min.,/sq. ft. .**q
Case 3. The pot value is changed by a ratio of expected R1
I min.,/sq.ft.
maximum values. An example is pot 7 which is changed R2
by hr* f h1* becausethe input to the pot contains the fac- h1(o) lJ lt.
tor lf h2* whereas the output must contain the factor hr(o) 10ft.
I f hrx. Pot B is another similar example. h, at equilibrium 30 ft.
ftr* - 60 ft' . q €
Case 4. Additional pots must be incorporated into the ft, at equilibrium 14 f.t.
- hz been
circuit. This would arise, for example, had fr' h^x - 30f.t.
-hrf hrx and h2f hrx would
desired.In this caseinputs of
t2
. . .
TEARNABOUT ANALOG COMPUTERS and a oI y ) * : 2 A . (15)
IABIE lt-Seltings Used for the Exomple Problern Three precautions should be exercised when using this
rule.
Component Numerical Value
1. The estimated maximum magnitudes must either
Amp 1 h1/60 continually decrease or increase as the derivatives are
Amp 2 -h1/60 considered in order. If this is not the case, the magni-
Amp 3 h2/30 tudes should be estimated in another manner.
Amp 4 -h2/30 2. The rule assumeszero initial conditions. For a prob-
Pot I 0.2500 lem with non-zero initial conditions, the estimated
Pot 2 (through gain of 0'1) 0.6667 magnitudes should be compared with the initial con-
Pot 3 0.2500 ditions and the larger of the two in each case used
Pot 4 0.3333 for scaling purposes.
Pot 5 (through gain of 0.1) 0.5000 3. The rule assumesthat the problem has a stable so-
Pot 6 0.3750 lution. If the programer is doubtful of his problem's
Pot 7 0 . 12 5 0 stability, the stability is easily examined by using the
Pot B 0.2500 Routh criterion.
situation. In other cases,one must resort to other tech- Using the equal coefficient rule and evaluating using the
niques for estimating maximum and minimum values. numerical values in Table 3:
One method is to obtain these values by a trial com- * : 2 (qrRz*
[är] q r R z* { r R r ) , : 6 0 (17)
puter run. Another technique is that discussed by Jack-
sonl which is called the equal-coefficientrule. For differ- 9 z + Q t _ + Q r
l+1.
'
ential equations of second order or higher, this rule may _AaR\A2 A1A2RL' A:A|R2 _q /rq\^ v /
I I
I -' t r-l
of the dependent variable and of its derivatives. Given LA]R, ArR, ', ArR,J
an n-th order diflerential equation of the form
f a " n , 1 * : - - - q! -" + , 1 t = (19)
l-:l ' +' jA: L-A12 R. 8 75
dNy dN-1y dy L dt, _J A1R1A2 AlA2Rr |
o* o,#+aor-f(t) (12)
ä*at-t#+...+ The magnitudes decrease as the order of the derivative
increases, a check of the initial conditions indicate that
where l(r) is the forcing function, the maximum magni-
tude of y, they are less than the estimates, and the solution is
known to be stable. Hence the indicated values may be
dv dNv used as estimates for the maximum magnitudes.
' '
7l'" dtv In many problems the computer operating time, in
addition to the output voltage from amplifiers, must be
are estimated as follows: scaled. If the physical problem takes hours or days for
completion, the analog solution must be speededup both
Replace f (t) by its maximum magnitude, denoted by for convenience and precision. Likewise if the physical
A. The dependent variable and its derivatives are multi- problem is completed very rapidly, the analog solution
plied and divided by their maximum magnitudes to yield must be slowed down. In the next article of this series,
"'L-*)
-| | | z,Fl,_..+ ."rrt_lffil:a (13) be presented.
l.#l '{ä".[lri,*ä;
f;. tAi]"r
LITDRATURE CITED
conputation, McGraw-Hill Book co. Inc., New
* 'r1*
f anyl l-dn f a y 1-*
" r l r , ) : d , - , l z p =) :, " ' : ' , 1 ; ) A ( r 4 )
REPRINTED FROM HYDROCARBON PROCESSING
LearnAbout AnalogComPuters
l4
I^EARNABOUT ANALOG COMPUTERS For a second order differential equation of the form
Cld'zylde'z)* Cr(dyfde) * C,y: Cqxithenaturalfre-
qrr.ttcy of the equation is ar, : (CtlCr)'h. The natural
fiequency, for a third order differential -equation of the
forrn C,(dsyldes) + C,(d'ylda'z) * Cs(dylde) * Cnv :
Crr; is a,:(CafC2) %. There are no simple approxima-
tions for determining the natural frequency of differen-
tial equations higher than third order. However, if the
coefficients of the low-order terms of an equation are
quite diff erent from the coefficientsof the high-order
terms, a change of tirne scale will probably be required.
Fig. 10-The circuit after time scaling is considered'
Difierentiql Equotions. When it is found that a change
in time scale is desirable for a differential equation, a sub-
stitution must be made. If f is the real time variable, and
r is the computer time variable, the time for solution of
the equation may be changed by substituting rlp f.or t.
If the solution is to be slowed down, B ) 1; and if the
solution is to be speeded uP, ß ( 1. As an example we
will change the time scale of the following equation by
the factor B.
CJd2y/dt2)+C"(dY/dt) { Ce!:C * (20)
Componenl Output Max Value Scaled Value TABLE6-How Circull Components Are Scqled
Amp I a ar a/ at
Amp 3 b o1 b/b1 Physical Parameter NumericalValue ScaledValue
Amp 5 ca c/cr Amp 1 [0.01a] [0.01a]
Pot 9 1 ar/ c, Amp 2 [0.01b] [0.01b]
Pot 10 1 br/c, Amp 5 [0.01e] [0.0I c]
Pot 1 a(0) at a(0)/a, Pot 1 1.00 1.00
Pot 2 b (b,/ ar) k" Pot 2 50 0.50
Pot 3 k1 Pot 3 100 1.00
Pot 4 0 0 Pot 4 0 0
Pot 5 k1 ( a r / b , )k , Pot 5 100 1.00
Pot 6 k4 (cr/br)kn Pot 6 75 0 . 7s
Pot 7 kr*ku (k"+k") Pot 7 200 2.00
Pot B n n/ 6t Pot 8 1.00 1.00
t6
LEARNABOUT ANALOG COMPUTERS sary. It should be uoted that a, b an,J c each must be
magnitude scaled. The reactions are constrained and the
initial conditions are such that the concentrations of
A, B or C at any time may not exceed 100, the initial
concentration of A.
The numerical values of each of the physical param-
eters may now be determined,assumingat: br : cr:
100. These are shown in the second column of Table 6.
The large gains on the integrators indicate that the sys-
tem is a naturally fast one and must be time-scaled to
slow it down. Since it is desirable to have pot settings
beween about 30 and 0.1 we choosethe time scale factor
B to be 100.
Only pots that are on inputs (but not initial condition
pots) to integrators must be scaled. Pots 2, 3, 5, 6 and 7
are adjusted by dividing the numerical value found from
magnitude scaling by ,,B: 100. The magnitude and time
scaled values are listed in the third column of Table 6.
Fig. l5-Summation can be performed in the integrators. The final computer circuit (amplitude and time scaled) is
given in Fig. 16.
If it is desirable to plot the concentration of A, B, or C
as a function of time on an X-Y plotter, a time signal
for the plotter may be supplied by the analog computer.
(o0lo) -(00t0) It is important to note that the analog circuit for the time
l\
signal (a ramp increasein voltage) should be scaled just
as the problem. Suppose the solution is to be recorded in
F secondsof computer operating time. The scaled circuit
for the ramp signal is shown in Fig. 17.
The ramp units are equal to z/fl where r is the com-
puter operating time, and thus will reach a maximum
value equal to that of the reference voltage after F sec-
(0orb)l\-(ooru) onds of operating time.
In order to obtain the relationship between actual
problem time I and the length of computer chart travel,
two additional factors must be known: the time scale fac-
tor, B: rft; and the recorder chart setting G, volts per
inch. The relationship is developedin the following man-
Fig. 16-After amplitude and time scaling is applied. ner:
Ramp : (Ref. Volts; r/d volts
Since z : pü,
Volts Developed : (Ref. Volts) Bt/F.
Volts Developed per Unit Actual Problem Time
: (Ref. Votts) B/F
Fig.17-The time signalfor the plotteris alsoscaled. Using a recorder chart setting of G volts per inch,
Inches of Chart Swept by Ramp per Unit Actual Prob-
lem Time : (Ref. volts) pl@c)
must be programed on the computer. The physical param-
As an example, if the solution is to be recorded for
eters chosen for illustration are siven below.
10 seconds of computer operating time using a l0-volt
,t, : 100sJc-. machine, a recorder set at 1 volt/inch, and p : 10, gives
,t, : 50 sec-r 10 inches of chart : 1 unit actual problem time.
*, - 150sec-r In many problems it is desirable to check to see if the
kn: 75 sec-t chosen time scale is such that the solution is recorded
a ( 0 )- 100
accurately. This may be done by plotting the solution for
ä ( 0 )- 0
c ( 0 )- 0 the chosen time scale and then replot the solution for a
different value of B. If the two solutions do not overlap
We may now draw a preliminary computer circuit to the chosen time scale is a poor one and some of the solu-
solve these simultaneous equations as shown in Fig. 14. tion information is being lost. New time scalesshould be
This circuit may be simplified if we are not interested chosen until two different values of p give solutions
in obtaining da/dt or db/dt by performing the summa- which coincide.
tion in the integrator. The revised circuit is shown in Fig.
15. Indaing Term: Analogs-g, Cbcuits-I0, Computations-4, Computers-$ De-
rcriptiorS, Electricity-IO, Enginecring-4, Programing-l0, Simulation-4.
We now must scale Equations (23), (24) and (26).
We will first magnitude scale and then time scale if neces-
t8
LEARNABOUTANALOGCOMPUTERS
...
Fig. l&-Flexibility must be considered
when setting up an analog computer
simulation circuit.
e\ n\ c
and
dA/dt: - hA (27)
dB/dt: hA - kzB (28)
dC/dt : k, B (2e)
A preliminary analog circuit based on these equations is
given in Fig. 18. An alternate, but equivalent, solution
can be obtained by eliminating the dependent composi-
tion B to yield:
!
"
lcz
\1 dt ff-r,r):o (32)
this series is recommended for completing amplitude
the incorporation of which would require major modifica- scaling. If maximum magnitudes are unknown or may
tion of Fig. 19. Similar difficulties would be encountered change from one run of the final simulation to another,
for other model changes.Thus, although the circuits in amplitude scaling may be completed in a symbolic
both figures represent the same physical system,Fig. 18, manner so that modifications may be readily made.
in which the one-to-one correspondenceis retained, is c. The process of time scaling as described in Part 3 of
significantly more flexible. this seriesis a systematic way of completing this step.
Once the model has been completed, with due consid- It is to be noted that tirne scaling does not affect am-
eration to the simplicity, acclracy, and flexibility of the plitude scaling but that changes in amplitude scaling
analog circuit, the details of the preliminary analog cir- may affect the choice of an appropriate time scalefac-
cuit can be completed and the scaling undertaken. In tor. Thus time scaling should be completed after am-
completing the scaled diagram, it is suggestedthat: plitude scaling.
20
1TARNABOUTANALOG COMPUTERS
...
Pr PH
Thus far in describing the steps of a simulation, atten- c. The choice modification of a simulated system
tion has been directed toward the calculation of the andf or its parameters in order to obtain results possess-
values of known variables for a specified known system ing some characteristic, such as the determination of
with specified known parameter values. This has per- the optimum production rate for a chemical reactor
mitted the sequentialexamination of system.
The development of the mathematical model and The simulations classified under a may be readily
the preliminary analog diagram carried out a^fterthe circuit for the first set of parameters
is prepared for operation by systematically:
c The development of the scaled analog diagram
o Setting new pararneter values, and
a The achievement of accurate patching, and finally
o Recording the results for each set of parameter
o Computer operation leading to the desired results. values.
As such, the presentation describesa rather narrow facet If the parameter changes to be made are small, there is
of analog simulation for frequently the simulation will be generally no need to rescale. Flowever, if the changes are
directed towards: large, the assumedscaling factors should be periodically
a. The calculation of the values of known variables checked and the circuit rescaled if necessary.In the re-
for a specified,known system using numerous param- scaling process, a clear concise circuit diagram together
eter values with tables of all analog variables and pot values, which
presents the results of scaling obtained by symbolically
b. The calculation of the values of known variables retaining maximum values of all amplifier outputs, can
for numerous modifications of the simulated system be of great use. There is generally no need to perform
using numerous parameter values additional static and dynamic checks.
22
LEARN
ABOUT
ANALOGCOMPUTERS.
.. )
DESIGN-StabilitY'
PART5: CIRGUIT
mustbe
andflexibilitY
accuracy
consideredbeforechoosingfrom al'
Fie. 23-An easy circuit to solve an algebraicequation
amongseveralcircuitdesigns ihEuehit may be-unstablein operation'
in error
that the outPut from amplifier 1 is momentarily
u. Following the progress of the error
üu u" t*o.*t
;i;;h ;i'; feedba.k loop In an äppräximate' sequential
in
Theodore W. Godmqn and Theodore G' Smith ;;;;;., the error is seän to resuit in an error of eo
" the first pass' Thus the
Univenity of Maryland, College Park, Md' ,ft" from amplifier 1 after
initial error results in a subsequenterror of the same sigrr'
""ip",
error
DunrNc rHE couRsE of preparing a problem for
analog n-""" t'fr" term positive feedback' Now if a 1 l, the
methods may soon die out. However, if. a ) 1 the error is propa-
solution, several mathematically equivalent will
if the gain
occur to the programer. At this point, he must make a *"i"J t"a the amplifiers soon overload' Thus
äi,r*J""al".k loäp becomes excessive, a ) l, the circuit
judgment as 1o which will best suit his particular pur-
the becomes unstable.
porä. et indicated in the last article of this series'
aocuracy of
rrno,rrr, of required analog equipment, the Positive feedback arises when an even number of
am-
and the flexibility of the To insure
required analog plifiers occur in a loop of an analog circuit'
"o*po,,""t', factors. in this choosing
unulog circuits ä be-significant ;;ilir" in such cases, the total gain around the loop
be given to such considerations'
^ho*"rr".,Attention can
pro..ir. must bä less than unitY.
only after analog stability has been assured' Negative feedback arises when an odd number
of am-
Below attention is drawn to those situations in
which ptifiei occur in a loop of an analog circuit' As indicated
can arise. Particular emphasis is given g*rrn"r," rattier restriotive maximum permissible
analog instability ü'W.
than 8' 2'9
to iß äccurrence during the solution of algebraic equations gilrn urir" in ittis case also' These are less
containing 3, 5 and 7-amplifiers respec-
on the analog. Methods for insuring stability
are discussed änd 2.1 for loops
pr"i"n." of intägrators in- feedback loop'
to indicate the applicability of the ana- tively. The flle
and are extended to the point
calculations. The article is concluded t o*äu"t, märkedly enhance"sthe stability,
Iog in optimization than ever
of transfer functions on the where maximum permissible gains are higher
by'corrsid".ing the synthesis
required.
analog.
To avoid feedback instability it is advisable to:
per- r Avoid, where possible, feedback loops which do not
Feedbqck 3rability. Although the analog computer
solution of algebraic equa- contain at Ieast one integrators
mits the almost instantaneous
of circuits
tions, precautrons must be taken in the design o Avoid summing-amplifier gains in excessof 306
As an example, consider the
to ,irr" algebraic equations' r Double check the occurrence of loops containing
ay * x'a ) 0' it is desired
simple eq"=ationy: -where (i.e. loops with an even number of am-
'ptin".,r" J"edback
positi
to ärr" for y given r. An analog circuit is readily designed and integrators to insure that they are required
for the equation in the form given in Fig' 23' ily the physical sytt"* being examined) '
: - o) '
The sohrtion which might be expected is y x l(1 o Adhere religiously to the maximum permissible feed-
1 and
In actuality, the circuit *ill b"to*" unstable if a ) back gains if loops wiih only amplifiers are used'
from the
the ampliders will readily saturate' This -results in the
with excessive Such precautions are of particular significance
phenomena known as positive Jeedback foi solving algebraic equations'
a"tiä" analog circuits
gui". f" i[ustrate the reason for the instability, assume "i
24
...
TEARNABOUTANALOGCOMPUTERS ticomponent distillation on an analog. In this case, if
Raoult's Law is assumed, it can be showno that the lin-
earized equilibrium relationship is
g -v1.o*i,o
!i,i:ki,r,,,o*jrL
a =1
t 7,,pfu,,p
B=r
where
!i, i:the deviation from its steady-state value of the mole
fraction of component i in the vapor leaving tray j
7j, o:the steady-state value of the mole fraction of component
i in the liquid on traY j
v* , : the deviation from its steady-state value of the mole
fraction of component l' in the liquid on tray j
Vr,'- o: ratio of the vapor pressure of component i evaluated at
the temperature of tray j to the column pressure
R ,, o: the derivative of. K1, l, evaluated at the temperature of
tray j
N : number of components present.
wnere
14- ll t2 t3
an over-all gain of (ar" atr) I (a2 a""), From the last form of Equation 57, it is evident that
o The loop consistingof amplifiers I and 3 which has dElat < 0 is insured if
an over-all gain of (ar" att) I (art a"").
If a stable solution is to be obtained each of the three ah : / öer , öez , dea\
- t (eB)
gains must be less than unity. For the particular case of dt \" art " öxt-r "t a*, )
multicomponent equilibrium, one would generally expect
dxz / let -r' öez ' öes\
ooo) ooi; j+ i,i;which casethe stabiliiy criteria wÄuld ez (5e)
be automatically satisfied. If, however Equations 45,46 \"il axz-r " u' )
and,47 had been solved for x2, x3 a.id x1 respectively, the dxs / öer , öez , deg \
resultingcircuitwouldbeunstable. - ez (60)
ä: (a *t Arr+ " Ao)
Although the direct method indicated above can fre-
quently be_used to.great advantage, care mu-st be taken or in the general case
to insure that maximum permissible feedback gains are
not exceeded. Particularly troublesome are those casesin }eo
dxo - | (61)
which the equations can not be rearranged to make
A: 1u ar'
a a t ) a r , t jj * i , t h e r e i s n o u n i q u e r e l a t i o n s h i p b e t w e e n
the a41'swhich can be used to the programer's advantage,
F"l Equations 54' 55 and 56' Equations 58, 59 and 60
or the a;1.'sare variable. In the latter case,tt" pott i" nig.
reduce to
24 woulä be replaced by multipliers and the analysis of
stability could become unduly complex. Under such cir-
aaf dt : - (arr er * azrez I aatea) (62)
cumstances the programer might be well advised to use a
: - (ane'l-razz
dx2rds ez* anea) (63)
ff1#,$J'H':L:|;:: "il-,ä:iäJT"l;;i,'iJ
il:f'i;
dxt/dt : - (ot er t azae2 ! as es) (6+;
aiternative.
The rnethod of steepest descent results in the synthesis Programing of Equations 62, 63 and 64 as indicated in
of a set of differential equations, the steady-state solution Fig. 26 yields an analog circuit which, after a brief
of which represents the solution of the algebraic equa- period of operation, will produce the desired values
tion(s). The method of synthesisis given below. of Nr, N'2and x3 without the necessity of performing a
Choosing Equations 45, 46 and 47 for example, errors stability analysis.
ey 02 a;fide3are defined in the following manner.
et: -!t * at *r * at, x, * aß xa (54) To speed the approach to the steady-state, the speed
(55) of integration can be increased by using a smaller feed-
e,2:-y2I aztxt* anxz* azsxs
back capacitor. Many analogs permit a P :0.1 or less
ss : -le,* a", x1J- a"" x2* a a sx s (56)
to be selected in this manner. In the application of tlis
If the correct values of xr, x2 and x3 are chosen, er : ez technique, the calculation of the er's is also crucial since
: e3 : 0. ChoosingE : "t2 * e"2,E :0 iS seen a null point is being sought. To enhance accuracy, feed-
"2" *
to result if and only if the correct solution is obtained. back pots may be placed in the feedback line of the am-
Other forms for E may also be used. For example, plifiers which calculate the s4's-Pots Py P2 and Ps in Fig.
€r 4+ e 2 a* e s aE, > 0 . 26. By setting these pots at small values, the errors, e4's,
E: l e r f * l a r l * l e r lo r E :
generated can be amplified by several hundred so that the
It is recognized,that if an analog circuit is designed for
null point is more clearly defined. If this latter approach
which dE/dt < 0 for all ü, the steady-statesolution would
is taken, momentary overloading of the amplifiers may
result in E : 0 and thus solution. Letting )r1, x2 ar,d 2cs
occur prior to the approach to steady-state.It is common
be functions of /, the following is obtained by the rules of
to limit the values of e4's between -f reference, using
differentiation
techniques which are to be described in the next article,
in order to prevent such overloading. After the analog
f i : z( u k o, f, f*+" f f 1 circuit has attained a steady-statefor a set of fixed yr, |z
and ye, variable values of ht lz and le can be handled'
: , ("l#,#* ##* ##l) The values of xy x'2and 13 obtained from the circuit will
essentially be the instantaneous solution of the algebraic
set for the variable ya's.
, löez d4 -r, , öe2 a41
öe2 dx2 -r
The method of steepestdescent as described above for
r a-, dt J
"Lö*, dt öi dt linear algebraic equations has several other applications
and extensions which can be of considerable use to the
, Fäea dx1 , 0q dxz , ?et dxa
r nlö"t dt -r u, dt - art dt t) practicing engineer:
26
LEARN
ABOUTANALOGCOMPUTERS.
..
-€3
Nonlinear Algebraic and Transcendental Equations. Ap- The squared error form, E : X e62,is probably the most
propriate u,;'sare defined so that e4's : 0 corresponds to frequently used.
solution. Equation 61 then forms the basis for the syn-
thesisof differential equationswhose steady-stateresponse Minimization. In light of the application given in the
is the solution. If multiple solutionsexist, the analog solu- preceding method, the extension to the general use of
tion will proceed in the direction in which dEldt < 0for minimization becomes a natural one. The approach re-
all t from the initial conditions to the nearest solution.
mains the same. If E : E e;2 is to be minimized, Equa-
tion 61 details the differential equations to be used. If
When Number of Equations Exceed the Number of
E : I ei is to be minimized, it may be verified that the
Unknowns. For these cases,no change needs to be made
in the approach. The individual ei's will, however, reach equations to be used are:
non-zero values at steady-statecorresponding to a minimi-
zation of. E. Thus the "best" solution can be obtained ac- ax! flöer
(65)
cording to the relationship chosen between E and the e4's. dt:-I-ta,
Maximization. For maximization, the method is re- Fig. 27-y(s)/x(s) : r/(s + a).
ferred to as the method of steepest ascent. Differential
equations for which dE ldt > 0 for all t are used. For E :
t (s)/z (s) : l/(s * l)'? (72)
E elt, this is readily achieved deleting the negative sign in
Equation 61 to yield where 1/(s + l)'is the transfer function relating z to rt.
Upon specifying z(s), y(s) can be determined and y(ü)
dxt F }et recovered using established techniques for invenion. For
: t''E (66)
T or, systemsinvolving a number of coupled equations, the use
of the transform results in considerable saving of time and
the transfer function becomes a very convenient repre-
as the required equations. For E : I ei, the equations sentation of the systemunder consideration.
are
Consequently, the practicing engineer may frequently
be concerned with the analog simulation of processes
dxo F det (67) which have been characterized by transfer functions
- öxJ
dt rather than differential equations. In order to facilitate
I
the analog simulation in such cases,techniques for analog
circuit design directly from the transfer functions have
For further details of these and other steepest descent
been developed.
methods, including optimization with constraints, the
reader is referred to References4 and 5. Consider for example
Letting s be the Laplace transform, l(s), of a function is encountered.By noting that Equations 75 can be equiv-
f (l), where IU) - 0 for ü ( 0, is given by alently expressedas
(s) :
r " at.
t (s)/z (s) : l/(r * o) (76)
Based upon this definition, the Laplace transform of stan- z $)/x (s) : K/(s * bt (77)
dard functions and operations can be derived. A short list
the analog circuit is readily designed as given in Fig. 28.
is given in Table 8.
The usefulnessof the transform is indicated by consider- Adding the acceptable circuit given in Fig 29 for
ing the set of differential equations
TABI,E 8-Loploce Tronsforms
q/dt+t:xt)(0):0 (68)
dx/dt*x:<)r(0):0 (6e) tg ra
A A/s
T | /t'
Upon transforming Equations 68 and 69 they become -al
e r/(s * a)
sy(s)*t(s):x(s) (70) t ' Q ) , J ( O ): 0 s"f(s)
[, /__G)
loJ ( t ) a t J
sx(s)*x(s):a(s) (il)
J Q- ?), Tconstant l"J G)
which can be algebraically manipulated to yield
28
LEARNABOUTANALOGCOMPUTERS.
.. T2(dy/dt) * y: Tt(dx/dt) I x (81)
/ \ t / \ - 7 s
.l \s)/x \s) : e (84)
Fig.3G-y(s)/x(s) - (T,s+ l)/(T"s + 1). the time delay, which is equivalent to the physical situa-
tion in which y(t) : x(t - T). This transfer function
can not be simulated exactly on the analog using only
standard analog components. In order to approximate the
time delay, numerous circuits based on expansions of e-r"
have been developed.n,?,8The most frequently used is the
first-order Pad6 approximation
rrld_r-Ts/2 (85)
x(s)-r*Ts/2
Fig. 3l-y(s),/x(s) : li - (Ts/2)l/U + (Ts/2)1.
ffi:'/(*+f +r), (78) Other approximations used for greater accurary are the
second and fourth Pad6 approximations and the Stubbs-
all transfer functions of the form Single approximation. In Reference B, qpical responses
to step and sinusoidal inputs are given for the various ap-
-r (s) _ K
(7e) proximations. As indicated there, the time delay of a step
x(s)- n l m
J (J tarj
I
30
Some comprex oon-linear circuils are possibleby comhining
several simpler circuils
\l .
vALUE
Anolog Gomputer Dlode Clrcuirs
leol
e1 --# lbl
e,a er+|bl
06
Fig. 33-Minimum selection where x - -min (e,, e, e, S O F TL I M I T E R
eo, er),
lol
eo .'--ib
-em I
H A R DL I M I T E R
€1
* €m
e2
€g
e3
" K 2
en---++-J x^=l9l_l
a lemI
DEADSPACESIMULATION
Fig. 34-Maximum selectionwherex : -max (el, ez,es .
eo, e").
' u : lI nn; 4, I
)'"
These circuits require that the source .and diode for.-
ward resistancebe negligibly small compared with resis-
H Y S T E R E S IO
SR B A C K L A S H
tances R, and .Rr. The voltage eo may be either positive
or negative. If ,Rois equal to .Rr and eo is twice the posi-
tive reference voltage, then e, will be the positive refer-
ence voltage, and the circuit in Fig. 33 will find the mini-
mum of the set of input voltages.When eo is positive in
the circuit in Fig. 33 the circuit is equivalent to the AND
circuit used in digital computers. If eo is negative in the
''ff*
SL0P=
E| eo
an opera*tionalamplifier and two diodes'
-'Tli"',r""
cir-
,r"gtori"d"a potentiometers used in the
sei at ratios of a' and a' ol the
their"wipers
bias
".rir-hurr"
;;i;"g* :r e. The potentiometersare used to obtain
f;; ;fi" diodes. The output of the circuit .is
positive and
I r_rrurrrR
_l | \
e2 and is negative and
of a value a2e wnen " i' l"'s thatr
e'
oi u ,rutr" ire when r
may be variables in a
is greater
Problem'
than ez' Both r and
i"
____H- I f--
J I
F
rvith diodes to
Relays are often used in conjunction
circuits'.'Two types-of
form analog computer sr'vitching S P A C E E A N GB- A N G
DEAD
used' Thev are the differential relav
;;i;;t;.;;**o.,ly
type, having one or more poles and
of the double throw from
Plate-iype relays can usually Fig. 37-Complex non-linearcircuitscan be combined
the sensitive-ptut" s i m o l e rc i r c u i t s .
"iu1''
f" ty the ot'ipt't 9l '" operational amplifier'
"".tgit.a shätu,l in Fig' 36 can be used to produce
ä"-"itäi,
to
the function: A nutnber of simple circuits have been developed
t h a t a n s e s rl]
if x-Y 1 o s i m u l a t et h e d i s c o n t i n u o u s p h e n o m e n a
' = 1\ *a o 9 contains son.re of the
if x-Y ) rn.'"f fnyti.al problems' Tablä
circuits
*ori oit.n .rr.ä diod" circuits' Each of these
T i l e l o l v e r c l i o d e i r r t h e c i r c u i t o-3-rvolts
fFig.36limitsthe has an equivalent relay circuit replacing. the
and pre-
o,rio,ri of the operational amplifrer-.to "r""ffu
ilä;;l wn.., ttt"'diode conducts the relay srvitch
is
from overloading' This circr-rit is a
;;ir-';; closed. Other',vise, the srvitch is open'
"*frlin"t showrr earlier in
variation of the comparison circuit in the lit-
Several other clioclecircuits rnay be found
Fig. 35. {' 8-10
erature.l'
to con-
More Complex Gircuits. It is sometimespossible
combining rel-
,,.rr., *or. complex nonlinear circuits by
simple nonlinear circuits in series or parallel'
;;i;;ir
Fig' 37'
ir"l" ä""*ples of such circuits are shown in
of nonlinear
Scoling Nonlineqr Gircuirs' The scaling
manner as linear
compoients is accomplished-in the same
by changing the gain
comionents. They are scalecl .of
components are affected bv
;;J;;;; itputs.'Nonlinear
,.'uti.tg of their i'rputi and it is not uuusual
;;;iiil; "both
;;-t;;; to scale the input and output' Scaling can
,o*"ii-", be done simply by patching the nonlinear ele-
Fig. 35-A simple comparator circuit using diodes'
Po 0oNTRoLLER
n O
r2 -2
controlled'
Fig. 38-The pressurein the first tank is to be
32
LEARNABOUTANALOGCOMPUTERS.
.. P : Pressule,psi
Q : Flow rate, cfin
In this systern there are several nonlinear elements.
Each resistancemust usually be considered to be non-
linear and the pneumatic control valve is nonlinear. For
Fig. 39-Pneumatic control this example we will linearize each resistancefor small
valve parameters.
departures from the steady state and rvill consider only
the control valve as nonlinear.
A pneurnatic valve is usually designed so that it op-
erates over a pressure range from 3 to 15 psi. If the
xv(t)
valve is normally closed, it rvill begin to open when the
signal from the controller is 3 psi and will be fully open
0 Q r / d AP when the signal reaches 15 psi. A normally open valve
operates in the reverse fashion. The simulation of the
control valve requires the use of a nonlinear element
P2(r) (a limiter) because the controller output signal range
is normally 0 to 20 psi while valve movement is re-
Fig, {gi-g|o.k diagram of the control valve. stricted to the 3 to 15 psi range.
The outflorv Q, from the control valve is a function
of the upstream pr"essure Pr, the dorvnstreampressureP2,
ment on the cornputer and then scale it rvhile forcing
and the vaive stem position Xu (or the degree to which
the element rvith a variable input such as a ramp
the vah-e is open). For this problem we assumethat the
function.
valve is very fast so that valve stem position is linearly
related (no dynamic element) to the pressurePo supplied
A Process Control Exomple. As an example of the use
to the valve bonnet. The relationship betrveen the valve
of a diode circuit in an analog computer problem rve
elementsis shorvn in Fig. 39.
consider the system of Fig. 38 in rvhich it is desired to
c o n t r o l t h e p r e s s u r ei n a t a n k b y e m p l o y i n g a p n e u - The relationship betrn'eenoutflorv, pressure drop and
matic control valve to throttle the florv. For this example valve stem position is given in the follorving linearized
the pressureP, in the first tank is to be controllcd. The eqr-ration:
symbols are defined as follolvs:
C : Capacitance,cf./psi e, (t) : *u A n x, ro + ö19t P,(t)
t"' +
' *
p, (,) (86)
R : Resistance,
psi/cfm Pr'" ?Pt
SET Po(t)
POINT K.Ir+(r/ri)Jot] oQr/öaP
Pr(t)
t aer/at
P3(t)
I aer/at
P{ft)
R E P R I N T E DF R O M H Y D R O C A R B O N P R O C E S S I N G 33
+r0
problem'
Fig. 42-An analog diagram for the example
is used to con'
A proportional plus integral controller
this in turn may be written as ffol the Process
dtl (92)
(87) Pole: K, [l + 0/r) f
Q,(t):ft,x,(t)+#^Pft)
where Po : controlleroutput pressure,,
are con- signal)
In the linearized form the partial derivatives e : errot ft"t poi"t minus feedback
operating conditions' The : integral time
,runir-""ufrruted at the "o"ttäl 7l
40 illusträtes horv Pr' Pz and X' Kc : controller gain
Ljä.[ äi"s.t- of Fig.
affect outflow Qz. very fast leads to a
drop',flow rate' re- The assumption that the valve is
The relationshipsbetween pressure output pressure Po
be developed' In.this fl""r, ,"irri,inrhip b"t*""n controller
,irturrc" and tank capacitancemay and valve stem Position X,'
are pure capacitors
ä"rr"foo*"rr, it is asiumed the tanks (e3)
lines are pure
irro ,ärirtu"ce to flow) and connecting Xo: KoPo
resistances. 93 as it now stands
It should be noted that Equation
it tättti"tion on the extent of
For Tank 1 indicates that there "L
- (BB) valvestemmovement.Therelationshipmustbemodified
the follow-
Q(dPz/dt): Qz Qa i.^.""f"ä to the physical relation' fe have
ing restriction on Equation 93'
For Tank 2
- Qa
(8e) Xo: KoPo 3<P,<15
Cz(dPa/dt): Qa
Xo: O P o1 3
For Resistance2
Po > 15 (94)
(e0) Xo : 15 Ko
Pz - Pz: Qr rRr
94.requ,ire- lT-
The restrictions shown in Equation "
For Resistance 3 before the
it";;;.;;i; to u" pt^t"J in the Pressure line
Pt -
(el) valve bonnet.
Pr: QrRr
34
...
TEARNABOUTANALOGCOMPUTERS TABIE I O-Normql Vqlues of Constqnts
Function Value
sP'z1o 2.56 psi
Kc 0.7 psi,/in.
Ti 36 min.
Po 0.036 psi
WITHOUTLIMITER
KU 0.044 in./psi
aQ/reXa 10 cfm,/in.
aqnLP 0.083 cfm,/psi
R2 45 psi/c[m
.R3 590 psi/cfm
Pn/to 3 psi
cr 0.007 cflpsi
c,2 0.0001 cf,/psi
boundary conditions
seneous equation with the Proper
PART7= FUNCTIONGENERATION- *ltt U" the desired function'
ftrnction y-:
As an example, supposerve required the
Usuallythe bestwaYto generate K;:;;u p."i"" of äproblem' To generate this function
(dt'/dt)*at:o (s6)
solution are
The initial conditionsrvhich give the desired
: O, ! : K, and the analog crrcult is shown in
at t
Fig. 45.
36
TEARNABOUT ANALOG COMPUTERS
...
otz' of 3
a
Flg. 46-fnis differentialequationgeneratorrequiresa deriv-
ative input.
Fig. 49-lntegrators in cascade can be used to generate
powerfunctionsof time.
The solution of this equation may be obtained from the
Fractional powers of functions can readily be generated
circuit shown in Fig. 46, where yo : Ke-o,o. The genera-
if the function is not complex or imaginary and if multi-
tion of this function requiresthe {erivative input.i.
pliers and logarithmic elements are available. For example
the square root, cube root and 213 power of x may be
Sine and cosine functions of dependent variables are
used quite frequently and may be generated as solutions generatedby the circuits shown in Figs. 50, 51 and 52.
of differential equations or by a special device called a
resolver. If we wish to generate functions of the form
fu*o:o (r04)
be obtained by using logarithmic function generators. The
function may be generated using the circuit in Fig. 53.
A computer circuit for this function contains multipliers The logarithm of a function may be generated. by ttre
in place of potentiometersas shown in Fig. 48. use of function generators as well as by the solution of
When g_eneratinginteger powers of t, multipliers may differential equation. For example to obtain z : lnx we
.
be used. Ifowever, a very convenient method involves in_ take the time derivative of both sides.
tegrators in cascade with the initial condition on each
integrator set at zero. These are shown in Fie. 49. dz/dt: (r/x)dx/dt
by the differential equation technique. The equation to The taps on most servo motor driven potentiometers
be solved is are usually located a fixed distance apart. This feature
may limit the usefulnessof this device for a comple" f":"-
<' (l0e) tion which has several points of inflection which are close
<: -- \/ -\ /
z(t+a) together.
Servo motor driven potentiometers wound with resist-
and the computercircuit is givenin Fig. 55. ance wire so that resistanceis a function of the potentiom-
eter shaft position are usually found only in analog systems
that require repeated use of a particular complex func-
tion. This type of device can not be varied and must be
3(t+o) carefully choien and scaled for the particular application'
Modified X-Y recorders are often used as a curve fol-
lower type of function generator. The function of inter-
est is piotted using conducting ink and by using a special
signal pickup on the X-Y recorder the functional relation
may bä retriced and the resulting signal used in the ana-
log computation.
Th" ,tt" of function generating devicesemploying
Fie. 55-Proper scaling must be used when generating a func- servo motors requires careful time scaling of the analog
tio-n with the' differential equation technique.
38
LEARNABOUTANALOG COMPUTERS
O------1
!
'l :- e 0
e2
el
Fig.57-A diode function generator is faster than servo-
motor-driven potentiometers.
R3 o"1
SRq
)
o ö €l
'r5 r
? 3 4 5 r0
I
Fig,58-The simplestdiodefunctiongenerators
do not have Fig. 50-The relationbetweenviscosityand shear rate is di.
a meansof varyingthe breakpoints. vided into segments.
c.ircuit.sothat the problem is scaled within the frequency seriesit \\'as shown that if the plate voltage of a diode is
limitations of the servo motors. This means that fist re- larger than the cathode voltage the diode pemrits a cur-
petitive operation employing oscilloscopedisplay can not rent to pass through it. If the reverse is true, the diode
be used, and the solution must usually be recorded with acts as an infinite resistance and no current can pass
a servo recorder. through it. Fig. 57 illustrates the properties of a diode.
In the circuit of Fig. 57 the diode acts as a switch.
Diode Function Generqfion. Much of the servo motor
When the applied voltage a1 is less than the bias voltaee
driv-en
_function generation equipment has been replaced eo, the diode acts as an infinite resistance.At values äf
by diode function generatorsin modern analog computers.
applied voltage equal to or greater than the bias voltage
Although diode function generato.s t"p."s"rrt functions
the diode acts as if it has zero resistanceand current may
by straight line segmentsjust as servo motor driven po-
flow. The entire voltage drop e. due to the flow of current
tentiometer function generators, the diode function eln-
though the diode appearsacrossthe resistanceR as
erator is an all electronic system and is therefore m-uch
faster than seryomotor driven systems.
e2:(e1-ro)/R (l l0)
These are two general types of diode function genera-
tors: fixed and variable breakpoint generators.Thi volt- The solid curve in Fig. 57 illustrates how e, would vary
age at which two line segment approximations intersect with an increasein e, if the bias voltage were set at eo for
(a breakpoint) is fixed by the computer manufacturer in an ideal diode. Actual diodes have characteristicsmore
a fixed breakpoint function generator, while the ensineer closely representedby the dashed curve. The value of.e,
solving the problem may decide, within certain ti*itu- of bias voitage in Fig. 57 is called a breakpoint. The grad-
tions, the location of breakpointsfor a variable breakpoint ual.rather than sharp change of slope at the breakf,oinr
function generator. The diode has such an impoitant of the diode characteristiccrrue petmits a more accurate
function in diode function generation, that it is essential representationof many functions which do not have sharp
its operation be understood. In a previous article in this changesin slope.
+ref
- Log
Function Simulqtion. The processof simulating a furrc- contain an adjustment which permits the operator to
tion with diode function generators is one which involves translate the entire function l(x) between the two ex-
setting the slopes and breakpoints of the various line seg- tremes of machine voltage.
ments which make up the approximation. It is clear that A fixed breakpoint diode function generator should only
by setting e, in Equation 110 a breakpoint may be fixed, be used for functions that do not have sharp or closely
and by setting.R a slope may be fixed. The simplestdiode spaced changes in slope. The variable breakpoint diode
function generators do not have a means of varying the function generator has greater utility than the fixed break-
location of breakpoints. Fig. 58 shows a simplified fixed point type becauseof the easewith which breakpoints may
breakpoint diode function generator. be adjusted to conform to the variations of the function.
A 1O-volt computer with a l0-diode fixed breakpoint The procedure for the adjustment of breakpoints and
function generator would normally have breakpoints at slopes may vary depending upon the construction of the
l-volt intervals. The slopesof each of the Iine segments computer. Before attempting to generate a function by
are set by adjusting the values of resistors in the circuits this technique the computer manual should be consulted
shown in Fig. 58. Fig. 59 shows how the slopesof the first for procedural details.
four line segments might be set to obtain the composite A simple method of setting up and checking a function
function shown as a dashed line. is to follow these steps:
It should be noted that as the independent variable x Step 1. PIot the desired function on a sheet of graph
increasesfrom xo the dependent variable l(r) follows the paper to a large scale.
curve starting at ilo until r reaches the value rt. At this
Step 2. Locate breakpoints. If a fixed breakpoint func-
point the next diode circuit conducts and the value of
tion generator is used, the breakpoints are usually located
l1r; ir the sum of the curve starting at xo and the curve at even voltages. If a variable breakpoint function gener-
at 11. As * continues to increase each line segment adds
in as the value of the breakpoint voltage corresponding ator is used, convenient points to locate breakpoints are
to the function is reached. Most diode function generators at points of inflection. A great deal of judgment is needed
40
LEARNABOUTANALOGCOMPUTERS.
.. Step 3. A table should be prepared listing breakpoint
voltage and the value of the function at each breakpoint.
It is some times helpful to approximate the function on
the graph with straight lines between the breakpoints to
obtain an idea of the shape of the function to be gen-
erated.
Step 4. Potentiometers may now be adjusted to obtain
the proper value of f(*) corresponding to each break-
point voltage x. This may be conveniently done in the
operate mode.
Step 5. A time ramp may now be applied as the inde-
pendent variable and the generated function plotted with
an x-y plotter. The plotted curve may be compared with
the desired function and potentiometers adjusted until
the two curves conform as well as possible.
42
LEARNABOUT ANALOG COMPUTERS. . . is called the logical complement of I and is obtained
from an "Inversion" of a logical signal.
The simulation of this system using components from It is worth noting that the bang-bang circuit given in
each of the four component categories resolves into the _
Part 6 of this series can be used us ä comprätor by
following: making the proper choice for the lower and ,rpper volt-
o Designing an analog circuit assuming the proper age levels.
value of P6p is available. Using an electronic comparator, the desired signal Z,
o O_btaining a logical signal (21) which is TRUE if can be obtained by letting dt : hy az : -T, and,L : Lr.
ht) h and FALSE otherwise, and a logical signal (l*) Similarly L, can be obtained by letting et : - F , az : p t,
which is TRUE if F > P, and FALSE otherwise. andT:L".
o Obtaining a logical signal (Q) which is TRUE if The logical signal Z3 is obtained by operation on I,1
Lr: TRUE and,L: TRUE but FALSB otherwise. and,L2 using a component known as the AND gate. The
o Choosing between F - P, and 0 for the value of.psp symbol and characteristics are given in Fig. 67 and l, and
depending on the value of Zr. /2 are input logical signals and l, and Z are output logical
signals. I is referred to as the AND signal and Z as the
A preliminary circuit for the first step, assuming F is
NAND (not AND) signal.
available, is completed as shown in Fig. 65.
I F THEN
I 0" L L
I
THEN
o i ,4
looo
o,r-T
L L
F THEN
D, I2 L L
The OR Operations
Fig.74--This is a track and store unit.
In the same manner in which components operating
with only analog signals can be connected in a variety Track and Store Units. There are numerous ways in
of ways in order to achieve a desired operation, so may which the track and store units can be used to great
the components which operate with only logical signals. advantage during the solution of engineering problems on
Two examples are the use of AND gates and logical in- the analog. Several of these are outlined below.
verten in order to achieve the OR operation (Fig. 72)
and the exclusive OR operation (Fig. 73). The validity Use of a Single Track and Store Unit. It is frequently
of the circuits may be readily verified by sequentially the case during the completion of process designs that a
following the possible sets of input values through the particular value of a particular analog variable is of pri-
circuit. mary concern. This may be the maximum value of the
44
TEARNABOUT ANALOG COMPUTERS
. . .
_ P _ A P P
t - P _ A P P +AP
2 _ P _ A P - A P p + ^ P
a
_ P - A P _ A P i'iÄF+ap
- P _ A P - A P _ A P P+AP+AP
bucket bri' Two Parameter Sweep. Consider next the case indi-
Fig. 8l-A
gade for a single Param' cated schematically in Fig. 85 for which it- is desired to
a. eter. automatically determine A as a function of P1, in incre-
ments of aPr, and P2, in increments AP'2,by operating the
Assuming that the analog circuit is in the. repetitive- op-
eration moäe with un op"Ät" period of at least Z, wh\ch
is in turn greater than ihe time require! fo-r Z to change
pro-
to FALSE] it is desired to automate the following
Fig. 8+-A two parametersweepcircuit'
cedure:
46
LEARNABOUT ANALOG COMPUTERS
. . TABLE t 5-The Sequence of Events qs P,
qnd P, Undergoe Cycling
It is to be noted I,u is the result of an exclusive OR Speclal Symbols Used Wtth Varlables
* superscript to indicate expected maximum value of a variable,
operation on ,L' and Zn. Thus Is is TRUE if only Z3 is - overscore to indicate an initial or a desired value.
.dot over variable to indicate the first derivative with respect to exchanget
TRUE or if only Zn is TRUE and I,s is FALSE other- length.
UA
, c ! 9
- y T H /( T H ')z
89'
Fig. 9G-The preliminarycircuit for the reactor of Fig'
in
considering the simulation of an exothermic reaction
the chosen
a double-p"ipeheat exchanger.A schematic of
;;; is'gi,re., in Fig. gd taute 16 summarizes the
notation used.
e\o-
If it is assumedthat component c undergoesan
fi rst-order, temperature- dependent re-
thermic, irreversible,
;.;i; ;t it flows through the reactor, the differential
presentedin Täble 17 may be derived in order Fig. 92 Presentsa scaled analog diagram which was
prepäred fär numerical illustration of this technique'
to describä the system.The preliminary analog diagram
"f"",i""t
used in order to scale
bots Gr, Gr, G", ar.rdGa have been
given
' in Fig. 90 maY be designed' multipliers so
are assumed the outputs from the divider, squarer, and
S i n " " t h ä v a l u e o f 'T u a i d c a t x : 0
: x : 0,i, and'y are known' llowever' that errors are minimized in the serial use of these non-
known,7,, Trt at
at tc : X' In order to solve the linear components.The values of the required pots are
the vaiue of ?" is known
T" at x:0' summarized in Table 18.
problem it is necessaryto assumea value of
at x: X, and to repeat the cal- Using the numerical values given in-Table 19, the plot
to see if ?" :7" T" zt x: 0 to
"h."k using a new value if an appreciable error exists' eiven ii Fig. 93 was obtained by guessing
culation and then cycling the computer between
x : X is not B" ,..o iniiially,
fh" ,"lutio.rrlip b"t*""tt T" at x: 0 and at
48
TEARNABOUT ANALOG COMPUTERS
. . .
Pot Yalue
00 T"*/(To - Ts)*
01 U A (Tc - Ts)*/(F" cp Tc* B)
o2 ao/lu* 500
Qg fqx/(f6 - Ts)*
r.0
04 U A (T" - fg)+/(Fs ce Ta*)
0b tE*/(TH* B)
oa l,1yx
Gl Such that (G1 Ts*/Tn)-.,
= REF
G2 Such that [(G1 Tu*,/Tr)z (y/v\ Q/G2)I-",3 REF
G3 Such ttrat liCr Tn*/^la)2 (v/v\ Q/G2) (iu/iu*) (1/c3)l-- e REF
OZ E G2 c3 i'*/(c12 TE+2 B)
o8 E7c*
G4 Such that [y c/(y* c* G4)]-.' = REF
09 [AH] y*.* G4l(vn cp fu*)
10 y* G4l(vs B)
x,ft.
480
. Direction in which a correction must be made when
an erlor exists.
320 o Approximate magnitude of the required correction.
(ic- Tc),.
R The latter is generally determined by trial and error on
the value of K.
t60
By placing the computer in the repetitive operation
mode, the required trial and error is generally completed
so rapidly that the time for a solution is not noticed. Due
0
to the high speed, parameter changes may be made and,
for all practical purposes, the instäntaneous solution is
- 160
continuously shown on the output oscilloscope.
In the next article of this series, more sophisticated
applications and capabilities will be qualitatively discussed
- 320
and a summary will be given of this series in retrospect
to the developments in the field of digital simulation.
bdexing Term: Analogs-g, - Circuits-I0, Computations-4, Computers-g, De-
scriptions.-8, . Electricity-10, Engineering-4, Läiic-g,- Mä-"ri"-.-'g,--i'..ärari-
-480 ing-10, Simulatim4,
0 x,ft. X
Fig. 93-The computer was cycled between reset and operate.
50
An engineer should evaluate allernatives when deciding
fo use the analog compuler
to be achieved fairly easily and allorv the programer terminals are now commonly used at installations
to make full use of the flexibility of the digital computer. equipped with a time sharing digital facility. These sys-
tems are very convenient for digital "debugging" of
Programing the digital computer is, however, quite de- procedure-oriented based programs and on-the-spot so-
tailed and care must be taken to insure the proper logical lution of small to moderate sized problems. The future
sequenceof operations required for solution. Moreover, may very well witness the extensivedevelopment of user-
a program, once developed,tends to be specific in nature oriented languages for use with teletype terminals,
and not readily subject to extensive modification. If one thereby significantly enhancing even more user-machine
neglects the comparison of the time required for solution
communication.
and choosesa problem which is amenable to both analog
Other oossibilitiesfor the future revolve aronrndactive
and digital solution, the balance between the time spent
work being done to communicate with the digital com-
on analog setup and the time devoted to the development
puter by means of the written word. Here the objective
of a digital program using a procedure oriented language
is to develop means by which the user can write his
heavily depends on the past experience of the poten-
problem on specially prepared paper and have the digital
tial user.
computer formulate his problem in detailed terms, con-
The state of affairs with regard to easeof digital pro-
struct a program, run the program, and finally deliver
graming has, however, been markedly changing during
the results to the user. Communication by means of voice
the past few years. First, there has been a proliferation of is also being seriously considered.
F user-orientedlanguagesaimed at reducing to a minimum
Of particular interest to design engineersis work
I the complexity of programing certain specific classesof
currently being carried out with the objective of com-
problems. Particularly to be noted in conjunction with
municating with the digital by means of large oscilli-
D this seriesare a group of languages(MIDAS, MIMIC,
scopes.Technology has probably advanced to the point
Ir DSL/90, just to name three of several dozen) referred
r where today it would be possible to develop a system
to as digital simulation languages. Some of their im-
a which would require the following sequenceof steps for
portant characteristicsare:
D an engineer to calculate the dynamic responseof a plant
o They are specificallyoriented towards the numerical complex:
solution of ordinary differential equations and are prin-
o Specify the equipment and interconnections in the
cipally aimed at simulating analog operation on the digi-
plant complex, using a light pen, on an oscilliscope.
tal exclusive of the scaling and storage restrictions of the
analog. o Specify, either on the scope or by means of cards
o Very little previous exposure is required for their and/or tapes, the parameters required, the conditions to
D be imposed, and the results desired.
I comprehensive use.
o Pressa button and have the digital solve the problem
t o The burden of prescribing solution procedure as well
and display the desired dynamic response(s) on the
as the error check of numerical procedures such as inte-
) gration are removed from the user.
scope.
Although such a facility would be inherently complex
r Their use permits a very significant reduction in the in its makeup, the user could be completely unaware of
time spent in problem preparation over that required for the complexity and be directly concerned only with
a comparable solution on the analog or with a procedure- drawing lines using a light pen, specifying operating con-
oriented digital language.ll ditions, and interpreting results.
I Digital simulation languages are becoming ever in-
)t creasingly more popular. Developments will undoubtedly Fotentiol of Anolog Compulers. With the digital
continue and potentially can very markedly reduce the simulation languages now available and with the possi-
time required for programing a digital to provide numeri- bility of reducing communication with the digital com-
cal solutions to partial as well as ordinary differential puter to typewritten material, oscilliscope displays, the
1
equations, boundary value problems, and classesof itera- hand written word, or even voice, is the use of the analog
: tive calculations. The use of one of these languages re- in engineering to become pass6in the forseeablefuture?
quires a large computer because of the storage needed The answer is an emphatic NO!
for the program.
In contrast to an analog computer, a digital computer
1 The second marked change in digital inplementation is slow. The digital operates on discrete values in a se-
rvhich can be observedis the continual increasing in the quential manner in order to obtain a solution. For a given
degree of user-machine communication. Remote teletype problem and procedure the number of discrete values
52
LEARNABOUTANALOG COMPUTERS
".. the output would be initialized to FALSE. During the
operate period, the reset input would be made FAISE
equations,to be achieved auton"raticallyand continuously. and the set input would be the output from an electronic
'fhe
user might also automatically change the time con- comparator governed by the desired analog comparison.
stant so as to operatevery rapidly during a trial and error The first time the output from the comparator becomes
solution and then slow down the solution rate after con- TRUE, the flip-flop will set on the falling part of the
vergence so that a permanent record may be made on the next spike and the output will remain TRUE for the
logically controlled x-y recordersavailable. The possibility remainder of the run regardlessof the subsequentresult
of multispeed calculations and automatic selectionof the of the analog comparison.
mode of individual integrator's operation thus open a
whole new realm of analog applications which can be Further insight into the opera:ion of a flip-flop can be
used to great advantage in engineering. obtained by examining a schematic representation of a
component known as the differentiator such as is given
in Fig. 97. In this case, assumethe enable input to the
Potchoble Digifol logic. A necessaryprerequisite for
flip-flop remains TRUE.
the automatic control of integrators, electronic switches,
and track and store units is the synthesisof the appropri-
ate logical signals. The electronic comparator provides
one means for generating logical signals but is dependent
on an instantaneouscomp irison of analos sisnals. The OUTPUT
I
AND gate provides mea,,,sfor continuouily "comparing
logical signals. These, together rvith others to be trieflv INPUT OUTPUT
3
discussedlater, constitute what is referred to as the oatch-
able digital logic capability of the modern analoe and
OUTPUT
2
provide the user with the capability of logical signal syn-
thesis.
The flip-flop is an essential element in patchable digital Fig. 97-Schematic representationof a differentiator.
logic. In order to insure proper logical operations for cir-
cuits using flip-flops, their operation on many analog
To understand how the differentiator works, assume
computers is governed by carefully timed pulses (106 per
that input and output are initially FALSE. The sequence
second, 105 per second, etc.). Essentiallyeach pulse can
of events which occurs when the input becomes TRUE
be thought of as a spike. During the rising part of the
for a length of time and then becoÄes FALSE aqain is
spike the flip-flop "decides" what it is to do. Durins the shown in Table 20.
falling part of spike, the action is completed. The iche-
matic and characteristicsof operation of a flip-flop are
given in Fig. 96. TABLE 2o-Sequence for a Differentiator
0utput
qFT_-- J
I
54
. ..
A B O U TA N A L O GC O M P U T E R S
TEARN and petroleum engineering are applications such as model
building for large systems,2sthe simulation of the wet end
"subroutines." The digital's operation sequentially chooses of a paper manufacturing machine,2eand the simulation
a continuous stirred tank reactor or distillation tray of a chemical plant in order to train operators and opti-
mize design.so' 31'32
for analog simulation, scales the analog circuit, specifies
the initial conditions and period of analog simulation, The modern electronic analog cornPuter is an in-
permits the analog run to be completed, stores the re- herently powerful and rapid computational tool. The
quired information from the analog simulation, performs
addition of analog memory and patchable digital logic
the required intermediate calculation, and proceeds on to
capabilities establishesthe analog as an important tool in
the next reactor stage or tray. Problems of this type re-
engineering for the forseeable future. Despite the ad-
quire a prohibitive amount of time if only the digital is vances in digital software, the marked speed advantage
used and are too large for simulation if only the analog of the analog permits very large and complex engineering
is used. oroblems to be simulated in a reasonable amount of time'
Other hybrid applications have been primarily oriented The combination of the analog and the digital, with
towards the solution of mathematical relationships and appropriate interfacing, yields the most powerful and
application of mathematical techniques commonly en- näxiUte computational tool available today, the hybrid
countered in engineering. Solution of boundary value computer.
problems,ts application of the state variable conceptle and
Monte Carlo method,2o'21statistical studies and ACKNOWLEDGMENTS
The authors wish to thank Dr. F. J. -\4unno- IJniversity.of Maryland,
correlation, 22'2s and implementation of optimization and Mr. W. H. Kelly, Electronic Assciates, lnc., lor therr- sugges-trons
durine the preparation of this series. The analog time used-lor problem
schemes15,2a,25 are examplesof the wide range of applica- i"äs'*ade available bv Electronic Assocätes, Inc. and the Depart-
"."Ä.i"tio"
inerit of Chemical Engineering, University of Maryland.
tion and applicability of the hybrid in engineering. In all
CITED
casesthe use of the hybrid is highly recommended over LITERATURE
1Jackson, A. S., Analog Computation, McGraw-Hill Book Co., Inc', New
either analog or digital computation alone. Y o r k , 1 9 6 0 ,p . ' 1 0 1 .
, Smith, .I. M., Chemical Engineering Kinetics, McGraw-Hill Book Co', Inc.,
Of particular interest to the chemical and petroleum New York, 1956, pp. 150'158.
3 llouqen. O. A.,'anä Watson, K. M., Chemical Praess Principles, Vol. 3,
engineer is the use of the hybrid in solving partial differ- .Iohn"Wiley & Sons, Inc., 1947, P. 8'16.
t "Rq"rs. A. E.. and Connollv, T. W., Analog Computation in Engineering
ential equations.Certain of theseequationscan be reduced Deiien. McG.äw-Hill Book Co., Inc., New York. 196{J.
5 Korn. G. A.. and Korn. T, M', Electronic Analog and Hybrid Computers,
to ordinary differential equations by an appropriate McGiaw-Hill Book Co.. Inc., New York, 1964.
6 Cadman, T. W., Ph.D. Thesis, Carnegie Institute of Technology, 1965.
change of variable such as the familiar Blausius equation 7 Couehanowr. D. R., and Koppel, L' 8., Praess Systems Analysis and
describing the laminar boundary layer on a flat plate2s Conirol. McGraw-Hiil Bok Co., Inc., New York, 1965'
s Handbmk of Analog Computation, Dlectronic Associates, Inc., 1965.
and numerous physical systems involving diffusion.26'27 o Levine. Leon. Methods for Solving Engineering Problems Using Analog
Cmouters. McGraw-Hill Bok Co', Inc., New York, 1964.
Frequently a boundary value problem which is amena- '" fnuoduction to Analos Computation' John Wilev & Sons'
f,iLlii*ulto"rt,
ble to analog solution is obtained. In other casesthe sepa- tl Fm;Ls, 'Proerams.t'
R, G. E., and Schiesser, W. E., "The Ev-qlgtjon- -of Digital Sim.u-
presented at 59th Annual AIChE Meeting, Detroit'
ration of variables technique which has direct analog lation
Mich.. Decl +8. 1966.
t2 llannauer. G,. and Crcca. W., "Automatic Plotting of Stability Areas and
utility may be used.n Launch Windows," EAI Alplications Reference Librarv, l.3.ll a/h, March
The solutions of other partial differential equations on 1967.
ts Walli- C. R.. "Ouantizins and Sampling Errors in Hybrid Computation,"
the analog require finite differencing of all but one inde- AFIPS Conlirenie Proeecdinss,Vol' 26, Fall l9&, p. 545.
ra Frank. A.. and Lapidus. L., "Hvbrid Simulation of Chemical Engineering
pendent variablea since only one independent variable can Systenis," Chemicai Engineering Progress, Vol. 62, No. 6, June 1966.p. 6ti.
ü Anon. Hvbrid Simulation Gains in Prces Design," Chemical E Engineer'
be continuous simulated on the analog. The use of this ing News', Vol. 45, No. 23, trlay 29, 1967, p. 48.
16Witsenhausen, H., "Hybrid Simulation of a Tubulr Reactor," EAI Bulletin
technique with only the analog can require an excessive No. ALHC 6252-25, 1962,
lt Ruszkav- R.. and Mitchelt. E. E.. "Hvbrid Simulation oI a Reacting Dis'
amount of equipment although two independent variables tilfatiori'Column," AFIPS Conleience Proceedings, Vol. 28, Spring 1966,
can be simulateds. The altemative of digital simulation o. 389.
18Fadden- E. I.. "Hvbrid Computer Solution of Two Point Boundary Value
can lead to excessive computer time due to the large Problenis," fresented at 55th National AIChE Meeting, February 1965.
10llammond. I. L.. Ir.. "Application of State Variable 'Techniques in Analog
number of discrete values which must be handled in solv- Computer'Piogrammihg,"- Simulation. Vol. 4. No. 5, Mav 1965. p. 316.
, Korn. G. A.. "Hvbrid Computer Monte Carlo Techniques," Simulation,
ing these equations. Vol. 4, No. 4, April 1965, p.234.
?1Soudack. A. C.. and Little. W. D.. "Economical Hybridizing Scheme for
The hybrid affords the opportunity of time sharing the Applvini Monte Carlo Meihods to Solution of Partial Differential Equa-
tiöis," Simulation, Vol. 5, No. I, July 1965, p. 9.
solution of partial differential equations resulting in a sig- 22Mqer. I. H.. and Cuoit. C. R., "Resolving RTD's with Hybrid Comput'
ers," Ciemicäl EngineZöhg Progiess, Vol. 62. No. 6. Juoe 1966, p. 60.
nificant savings in analog equipment and time. The simu- 23Hmpton. R. L. T.. "A Hvbrid Analog-Dieital Pseudo Random Noise
lation of reactorspreviously notedla'16representsone such Generator," AFIPS Conlereacc Proceedings, Vol. 25, Sprins 1964' p. 287.
s Kim. M.. "On Optimum Control o[ Discrete Systems: II. Hybrid Computer
use. The time required to simulate a two dimensional Implementation,"-lSA Transaetiozs,, Vol. 5, No. 2, April 1966, p. 184.
s Mitchell, B. A., -Ir., "A Hvbrid Analog-Digital Parameter Opt-imizer for
equation on a hybrid is substantially reduced over the time Astrac II." AFIPS Conlerence Proceedings, Vol. 25, Spring 1964, p- 271.
6 Bennett. C, O., and Myers, J. E., Momentum, Heat and Mus Transfer,
required for digital simulation since finite differencing McGra*-Hill Bmk Co., Inc., 1962.
, Bird. R. 8., Stewart, W. E., and Lightfoot, E. N., Transport Phenomena.
need be used for only one independent variable. For third Johri Wiley & Sons, Inc., 1962.
a "A Hvbrid Solution to a Chenical Problem," Beckman Bulletin 1005, 1966.
order equations, one independent variable can be repre- F Harter. M. D., and Smittr. G. R., "The Hybrid Computer . . Powerful
sented by analog operating time, a second by finite differ- Tml for Process Problems," Paper Trade Jownal, April 25, 1966, p. 76.
o Fox. D. N., Parsons; J. R., and Dutcher, D. L., "Hybrid Simulator Spurs
ence approximations on the analog such as is done when Conirol Studies," Cottrol Engineeing, January 1966, p. 71.
31Panons. -I. R.. "Hvbrid Simulator Helps Develop Computer Control,"
solving second order equations on the analog,8 and the C h c m i c a l E n g i n e c r i n g ,V o l . 7 3 , N o . 1 7 , A u g . t 5 , 1 9 6 6 , p . 1 4 l .
?2Panons. J. R.. "Hvbrid Simulator Aids Computer Control Systen Develop-
third by finite difference calculations on the digital. ment," plesenied at 3lst Midyear Meeting, Division of Refining, American
Petroleum Irotitute, Houton, May 9-11, 1966.
The hybrid possessesthe power which permits the
Indexins Terru: Analoqs-9, Circuits-10, Computations-4, Computers-9, De-
solution of very large systems.Of importance to chemical scriptiois-8, Digital-9, Electiicity-10, Engineering-4, Simulation-4.
-4r-B+AB (1)
Fon ueNv TTIGHLYexothermic reactions, it is necessary
to dilute the feed to a commercial reactor with a recycle Ke
stream for ternperature control. Often, means for dr:ter- AB* B+ABz (2)
mining the initial composition. to the reactor are nÜt
Ks
availa6le; thus, an initial condition problem is encoun- (3)
A+C
tered in the mathematical description of the process' In
other words, the composition of the fresh feed is known' The rate of heat transfer is sufficiently high so that the
but the unknown composition of the recycle-fresh feed reactor is essentially isothermal and the reaction rate is
mixture is needed for the solution of the simulation equa-
tions. The schematic diagram in Fig' 1 illustrates this type
of reactor.
After the startup of a Gulf processing unit incorporat-
ing a recirculation reactor, it was found that an unusually PRODUCT
loi conversion was being obtained' It was suggestedthat
the reacto, design was unsottnd since the feed was quickly
circulated to tle exit point and could "short-circuit" FEED
throueh ',he reactor. Ilowever, low reaction rate due to
"asitation
noor could also give low conversion' It was
ä".id"i to check the soundnessof the reactor design by Fig. l-Shows schematic diagram of jacketed recirculation
ri*"tuti"g it, assuming that the agitation was good' If reactor.
56
-too
ll.b^ [cocJr
to
-K3
ro [soaJpo +[cocJ
- xr[soallroaJ roo
r-o
IsocJ1
[roa]r
-[roaJ
500
* [roaJ
IroeJro
-roo too
*[soaa
*[roeJ I sERyo
-T I
*[5oeszl
r-o
-
[soreJs loorezJr
.id%lP - [soeaJ
r-o
r l r -
+ [5o ABJ + [5o ^BJI
Fig.2-Shows analogcircuit diagramfor the simulationof a recirculationreactor.
-
!# : kt (A)(B)+ h (A) (4)
- Fie. 3-Shows that part of the analog circuit.to halt simula'
3y : k, (A)(B)+ k, (AB)(B) (5)
iibl wnen the outlei of the reactor was reached'
- d(AB)
-dt- : - k,(A)(B) + k2(AB)(B) (6)
- d(1!; -
- k, (AB)(B) (i)
dt
- d(c)
-T : - k"(A) (B)
58
F E E DF L O Wq,
( F E E DC o M P 0 S t T l 0 N )
sY,
SEVERIT
R E J U V E N A T ITOI M
NE
RN
PR0DUCTIO A T EP
'
production.
Fig. l-The simulation program shows the regenerationpolicy which will maximize total
Cycle
FindOptimumCat Regeneration
60
FIND OPTIMUMCAT REGENERATION
CYCLE byproduct C. Assuming no B in the feed stream, the steady
state solution to the differential eouations:
d A l d t : ( e / a )1 t o , - a ) - k J (2)
d B / d .-t ( o / u )1 - B ) + k l A - k z B (g)
INCREAS
NIG
Jno SEVERITY
F "."
(ts is given by:
z
F A: t?ffi (4)
o " "
-OO 9: qB (10)
o
Thus, using Equations (5), (7) and (10), the best level
of severity may be chosen once production rate is fixed
03 or vice versa.
o2
ol Case 2. Now consider the following kinetic equationr
0
0 2 4 6 8 l0 12 t4 16 t8 20 22 24 26 28 30 32 ,q:, a (tl)
D A Y SI N T OF U T U R Eo.
Fig. 3-The No. 2 reactor is rejuvenatedafter 12 days. The steady-statesolution to the equation
dAldt: ela(Ao^- A) - kl (12)
is given by
B:A4n-A:,lu:r,o^- (13)
( 1+ f t o )
rf; Differentiating the production rate of .B u'ith respect to
flow and substituting for O gives
of opera-
Degenerolion. Decay is governed by severity
productiä., tui", the latter term correspond-
ffi;;Jnet
ins to a poisonousinfluence'
"'itlr--u"t"*ed
that catalyst activity decay rate is
is also proPor-
increasedby higher operating severitiesand
tional to the Production rate'
daldt--ClsalPl
(15)
of carbon
This latter term corresponds to the deposition Fig. S-Catalyst activity increases sharply after rejuvenation'
for exanrple or gradual catalyst poisoning'
takesplace
Reiuvenction. It is assumedthat rejuvenation
uif n""a rate from the activity
a standard
level at
"clean"
which
level'
rejuvena-
qo-i r_.1
tion was begun up to I
lo'n"rol l
s1--*l
I F-*;l
Nor
place at low levels' the re- I I
,"irrrolnutiott närmally takes I
j;;";;il; interval f" chuttg"s little if r is fixed' Thus I
in- qz-!
ih;',it"" i;. r'hich productioir is lost is approximately I
r' During the-regeneration' begun tz ---l
;;;r;it-pt.portionaito
level at that
at iÄe't,., catalyst activity is raised from its
rrp to some'maximuÄ level ao at ttre rate r' The time
il
i" regenerate 2", depends on the time regeneration
;i.;.
to the maxi-
is started in a(t,), t".taing assymptotically Fig. 6-The simulation of three reactors in parallel'
mum aof r as regeneration is postponed'
PROCESS OBJECTIVE
search
The approach to the maximization is a simple
(i,i) for each value of a, q' and s' The
orr"r th"'rpuce
p."di"tive time extending from now to the
;;;iJi" :
"'it Note that the base t o
end of the planning interval 7'
is moving. The objective function then is:
maxtr (r) -
(i, i,
f - + T
I / ! i { o l s ( ' ) , s ( " ) ,t r ' 6 f ' s ( ' ) ' q ( ' ) ' r ' t ' ) d o
n J, ,l7)
with r{t,,o1rlT
and o 4t I'T
t,: - 0 , 1 ,2 ,' ' ' , 3 2
i(T/32),j
over the
The accumulated product from each reactor
interval for-one particular regeneration policy is
oturrnine
'.tto*"
ii nig. 3. Note the plateau in-the-curve for reactor
äe improrrement in production due to fresh-
No. Z u"a
ened catalyst. These cuwes are for uniform operating con- Fig. 7-The analog circuit for reactor No' 2'
interval T and between re-
Jitiorrs bolh throrrghout the
u"tors b,rt for diffäent initial catalyst activities results
ity is not felt, and an actual decrease in n2 (T)
(ao: a o ,a t : 2 a o / 3 , a": a o / 3 )' from the shut-down.
is
and fall' The influence of regeneration time on production
As regeneration time is varied n, (T) will rise each reactorln Fig. 4' The decreasing effect on
- the benefit from improved activ- shown for
It C T,) < t, {7,
62
CYCLE. . .
FINDOPTIMUMCAT REGENERATION
-t.
la)
3l
l'. Ino] 34
v . 'vr 64
f..
9l
v
39
35
?'lt/^'rg! 69
95 laln
_ f. 99
N [SoVo 44 -lsv1
36
[z-x srV,l to 74
eel
t3ä'l xtz
se, )- [szvz]
lo ro4
-[tu]
- tpl 49 ro [q 4
;3s
1 4
-tql 79 to
r09 4''l
-J99
Fig. &-Production from all three reactorsis estimatedto determinethe best regenerationpolicy.
the right is due to the shut-down loss just mentioned. For policy P (2, 12) and uniform operating conditions,
Other points to be noticed are that the peaks for each Fig. 5 shows the three catalyst activities, initially ao : eo,
reactor occur at different times and that A, the production at: 2a013 and a2 : ao13.The slopesof decay at any time
improvement-the difference between ,r (t,) and, (T)- depend on operating conditions. For example, by reducing
is greatest for reactor No. 2. The rightmost points no (T) the severity on a reactor, it is possible to prolong the cata-
represent production if regeneration is not made within lyst "life." The area under the curyes, a rough measure
the planning interval. By inspection, the best policy in of production, is easily seen to be affected by a change in
this case is P (2, 4). severity, throughput, and regeneration. For constant oper-
ating conditions and a given regeneration policy, the pas-
sage of time represents a continuous re-definition of the
About the quthor origin of o.
rro(i )
Tij+ Ti Tii>T i I
rij + ri
rr, (j)
r,20)
STORE OVERALL
REACTOR NUMSER. i BEST 2,0 , FOR
REGENERATIONTIME : J iIh REACToR tr-
L*
T _
L -
64
FIND OPTIMUMCAT REGENERATION
CYCTE. base case of otherwise uniform operating conditions with
an initial catalyst activities distribution of ao : (tro,dt:
2aof3 and az: aol3. As expected, the net effect is to
If Mo: (Lo) (L") : 1 then 56 best.
decreasethe maximum yield from each reactor. The best
lf M,: (f') (I.) : 1 then Sr best. policy is still obvious although time of regeneration is ad-
vanced. One useful application of this information might
Tf M": (Lr) (L1) : I then S, best. be to provide quantitative data for a regeneration de-
The rösults from decreasing the regeneration rate 25/o bottleneck study.
are shown in Table 1. A comparison is made against the The coresponding data for a change to severity are
shown in Table 2. The best policy under the new condi-
TABLE l-Sensitivity to Regenerotion Rate
tions (increased severity) is to regenerate reactor No. 2
but postpone for two days the time of regeneration. On
the other hand the recommendations with decreased se-
verity is to advance the regeneration time two day. The
change is production for this reactor is (0.4 * 0.1) and
0.80
0.56
(0.4 - 0.12) respectively.
o.32 Sensitivities to throughput changes are shown in Table
3. The alteration in production for reactor No. 2-still the
TABIE 2-Sensitivity to Severity one to regenerate-is the same as in the case for severity
changes,but in the reduced flow instance, an advancement
of 4 days in regeneration time is necessary.
Not shown in the table is the influence ort zrtra: ft is,
respectively, 0.06; 0.04; and 0.02. This indicates that
when reactor No. 2 is shut down, the available feed should
be diverted unequally to reactors Nos. 0 and 1, with re-
actor No. 0 having the larger share. Whether all the avail-
able feed should be used and in what precise ratio it ought
to be split must be determined at the time when reactor
No. 2 is about to be shut down.
TABLE 3-gensitivity to Thruput
All these sensitivity studies were made at a particular
value of time z. By increasing the number of runs at each
BASE z from 96 to 480, the sensitivity trials could be automated.
q=1.5 q:1.0 a:0,5
The conditions fed to high-speed circuit would be:
I ti
a (r), r, s (r), q (z)
0 7.14 0.78 1.0 0.80 0.76 o.72
I 0.96 0.56 0.86 o.56 0.66 o.44 a (r) , r, s (r), * Äs, q (z)
Best
Pollcy
0.82 0.28 o.72 0.32 0.60 0.28
a (r) , r, r ("), - As, q (r) Fora,i,i
2 z a (r), r, s (r), Q(r) * Aq
l
I
a(r),rrs(r),e(') -AS
t 72 10 6
to each
their use equilibrium and enthalpy data specific
This rigorousm.elhod incorporotes unäertaken must be pto"iata as part tl-:,,:^
ft.ui"- by :l
computtng
put data. Plate temPeratures are found
the Chqo-Seoder correlqlion ond the feed'
t"üfi" t"i"as below änd dew points above the
Redlich-Kwong equotion of stqte to
ap-
Clcrss ll-Programs based on the Thiele-Geddes22
compule rhe equilibrio ond entholpies
pr"".fr i" *rti?n feed, reflux, -quantity,u".d.tol" t^t|,"
are specl-
for multicomPonenl colculqtions äistillate, and the number of theoretical plates
are made using
fied. Plate to plate computations -mol
found by a
fraction ratios, änd product compositionsare
Wendell W. Wotermqn' Vem E' Alden Co', Chicago' meshing
-- comPutation'
;;-;;.;-P. Frszier, Natural Gas Pipeline Co' of Pt"d"*. reported by Lyster Mills'1? Newman'8
"J-'l,tn
II type' Again, enthalpy
America, Chicago u.,a-riunroo'" are of tlh" blu*t
u.J .q"lliUtium data must be supplied for each problem'
Tne rnocReu here presentedis useful for hydrocarbon io*"'J.., in these progru*t ä"t"ttttination of plate
mixtures over a wide range of compositions,temperatures i"-p".u,.tr"s is more aific"tt' It is accomplished by an
and pressures.It combines two important features: iterative process, a temperature profile being
assumed
Various special techniques have
. Rigorous computation of the number of theoretical t. i"iti"t" comprrtutions.
been usedto facilitate convergence'
pla"tesrequired for a specified separation'
o Incorporation of the Chao-Seader correlation and Closs tll-Programs based on solving the
simultaneous
the Redtich-Kwong equation of state enabling com- equations which express
-Th" plate and column material and
putation of reliablJ and enthalpy values it"*s specified are the same as
"q'ilibti"* balances.
*ithout introcluction of dutu specificto each problem "rithulpy
for tnä'Thiele-Geddes plate to plate method'
--
belongs
considered. fn" program developedby Amundson et al1'2'3
Matrix i,t""rsiott is used for solution of the
in Class II-I.
PREVIOUS COMPUTERPROGRA'VIS equations. Plate temperatures are deter-
,i*r;ttun"o"t
Computer programs developed to solve multicom- mined by iteration as in the Class II Programs'
be
ponent distillaiioriproblems by iigorous methods may change in
classedas follows: €lqss lV-Programs in which the progressive
proäuct compositions from start-up to the
Dlate and
ap- äf ,t"ady statä are computed bv a relaxatio'
Gloss l-Programs based on the Lervis-Mathesonl3 ;;;;i;;;
successfully
prou"n in wliich feed, reflux and the separation method. Rose, Swelny and Schroät" have
-desired
äre specified. The required number of theoretical plates developedthis aPProach'
material
is foäa by plate to plate calculation, component meshing All these programs are rigorous with respect to
b"alance computing techniques' but fail to
at the feeä ptut. t"ittg obtained by over-all iteration' u.rJ
".tthutpy of equi-
tuf" into u..o.rttt the composition dependency
Programs developed by Mclntire,lu Maddox and Er- tedious process of
In fifrrir* values. Further, alt'hough the
barts änd Bonner2' are examplesof the Class I group'
66
The melhod includes a rigorous computalion oI the numher
oI iheorelical plales needed
making plate to plate calculationsmanually is eliminated, the first reported attempt to incorporate a generalized
there remains the time consuming work of selecting and equation of state into a distillation program. It is under-
preparing equilibrium and enthalpy data for individual stood that difficulty in attaining convergence has been
problems and the attendant possibility of human error. experiencedwith this program.
Greenstadt, et alll described a program of the Class After an investigation of alternate methods for pre-
I type in which Newton's method is utilized to solve dicting K values in hydrocarbon mixtures, Cavett?
plate and overall balances as well as the equations for adopted the Chao-Seader correlations together with the
adjusting product distribution to accomplish feed plate Redlich-Kwong equation of statele (or alternately for
meshing. fn one method of applying the program the low pressure ranges the Black equation of stateo) as the
Benedict, Webb, Rubin equations is used in a subroutine preferred method for obtaining thermodynamic proper-
to compute equilibrium and enthalpy values. This was ties for use in distillation computations. Cavett developed
two programs,8 one of the Class II and the other of the
TABLE l-Acceptoble Componentj Class III type in which it is necessary to specify the
components in the feed, the reflux and the operating
Non-Hydrocarbons Diolefins
pressure.Equilibrium and enthalpy values are computed
Hydrogen Propadiene as required.
Nitrogen 1, 2-Butadiene There is an important difference in utility between the
Carbon Dioxide 1, 3-Butadiene
Carbon Monoxide Class I and the Class II and III programs. The latter
Hydrogen Sulfide Cyclop,ara.ffins are analytic; that is, they compute performance in a
Sulfur Dioxide Cyclopentane
Oxygen Methyl Cyclopentane
column having a fixed number of theoretical plates. They
Ethyl Cyclopentane are well adapted to studies on existing installations, but
Paraffins Cyclohexane are inconvenient for design applications since the number
Shaight Chain-C, through Cr Methyl Cyclohexane
iso-Butane Ethyl Cyclohexane of plates required to effect a given separation can be
iso-Pentane determined only by submitting a series of problems and
neo-Pentane Aromatics
interpolating among the results. The Class I programs
2-Methyl Pentane Benzene
3-Methyl Pentane Toluene are advantageous for design work since they compute
2, 2-Dimethyl Butane Xylenes-Ortho, Meta and directly the required number of plates" They also de-
2, 3-Dimethyl Butane Para
Ethyl Benzene termine automatically the optimum point for feed intro-
OIefins duction; whereas, with Class II and III programs this
Straight Chain-Double Bond must be found by trial and error.
in I Position-C" through C,
2-Butene-Cis and Trans Further the Class II and III programs usually have
2-Pentene-Cis and Trans a maxirnum number of plates that can be considered.
2-Methyl-1-Butene
2-Methyl-2-Butene
With Class I programs the number of theoretical plates
3-Methyl-1-Butene is unlimited.
x
I
68
The occuracyol ihe resulls depends upon hhe accurocyol
Ihe Chao-Seadercorrelalion
ffiil#
*i*trt"t
en$.
. Over its range of applicability deviations from ex-
perimental
'seadet data of K values computed by the thug-
method are of the same order of magnitude
as discrepancies among independent determinations
by different exPerimenters.
W;rii*+ptr
Fig. 2-The progrun is limited to ranges of conditions over
which the Chao.Seadercorrelation is valid'
Of the several K value correlation techniques that
have been devised some are not easily adapted to com-
puter applications. Some do not permit determination of
äquilibrium ratio as a function of compositio'n as well as Experience. The program has been used to design dis-
temperature and pressure. For general -use none aPpeax tillation columns of the following tyPes:
more accurate or versatile than the Chao'Seader cor-
relation. o Natural gasoline stabilizer
o Deethanizer
Progrcrm Limitcrtions. The program is limited to ranges
of önditions over which the Chao-seader correlation is o Propane separation from heavier hydrocarbons
valid.e Limitations with respect to acceptable components
and the number of cornponents handled in a specific o Separation of iso and n butanes
problem are described under an earlier section, "Char-
r Separation of iso and n Pentanes
acteristics of This Program."
70
DISTILTATION
PROGRAMGENERATES
OWN DATA W moles of bottom product
tci mole fraction of component i in liquid
the quthors
About
LITERATURE CITED
WnNour,r, W. WtronulN i.s director ol
chemical engineering f or Vern E. Ald,en ;f si:'tu*;t:tl;]ä4tß:li#x!,"'i:tf
r^4rar1p^d_s^o3, !"ä,;"ll,,lo1'?,5'??
_ N. R., A. J. Fontinen and J. W.',fi"iti"y, Ä.i.Cn.i'jäilia
Co., Chicago, IlI. He holds a B.S. de- 5, 295 (1959).
gree from Iowa State Uniaersitg and aBachelor,
J 8., Petroleum Refiner 36, No. 6, 16l (1957).
N., c. B. Webb and L. C Rubin,'Chem)-Ei'g'.'progress 47,419
M.S. and, Ph.D. degrees in chemical en- 'o95BlT"di".;
gi.neeri,ng from Uni,uersi,ty of Iotoa. 0 Bfack, C., Ind. Ens. Chem. 47.1685 (1955).
H., Proc. Americän Pet,roteum Institute, Vol 42, Sec. 3.
Before joiwing Ald,en Co. he 'tDas a 351-3ä"iib,6r}:
process engi,neet'f or Esso Eeseatch and a Cavitt, R. H., Personal Communication fune 24. 1965.
e . Q h 1 o ,K . C . a n d
Engineering Co, and,in subsequent con- 'o E4!.^"-!, A- J. D. Seader,A.I.C|.E. Jouriat 7, No.4,598 (1961).
J.,M.-S. Thesis in'Chemical Engineerinq Uni".iriiv-oi-öilä-
necti.ons he has superuised, pr.ocess t.r"nut".,']r'"
Chemical
Ensineer's
H;ndb@k-P13-42.
4rhEditio;.
desi,gn f or plants in petrochem;icals, in- tnlt,llj-Urllr,
organic chemi,cals and metallut"gy. Dr. Waterman is a, rnen1,-
ber of ACS, AICLE and the National Societg of Profes-
,.1ff;:l'.'l$.rir,,Ir"-11tiÄlHr:e':1."#;.'J,ftr,bt
iltl,""li
Separation Processes, Reinhold Publishing Company,'New ior:k
lföÄ!i:.*"
sional Engineet's. rrLewis, W. K., and C. L. Matheson,Ind. Ens. Chem.24.49+ (lg}2\.
il Lyster,-!V. N.,-S. D. S. Biltingsley and C.'D. tlbllanä.
^ ,!.5ulliv1g^Jr.,
rerfoleum freFnef Jö, o, zzl (lvJy).
.R N. and J. H. Erbar, Oil and, Gas Journal 57, April 27, il6
';tr$idd."
JAMES P. FRAZIER i.s a process desi,gn R L., Chem. Eng. Symposium Serier No. 21, Vol. 55, 59
engineer for National Gas Pipeline Co. 'ö;äf:I"ri.",
'
1?Mills. A. K., Chem. Eng. Progress 55, No. 7, 93
0959).
of America, Chi.cago, III. He receiaed, .
tl
N"yIog^1, l. 5., Hydroeärbon Processing and Petroleum Refiner 42, No.
a B,S. degree 'in chemi,cal engineering 4, r41 (1963).
lsRedlich, O. and
J. N. W. Kwong, Chem. Reaieu M.233 (19+9\.
from lowa State Uniuersity and is a ,.2oRiedel, J.,.C..(R_eport on work of John S. Bonner) Oil and Gai lournal
member of the Nati,onal SocietE of Pro- JJ, May O, rUJ ( lyJ /.) .
4., R. F. Sweeny and V. N. Schrodt, Ind. Eng. Chem. 50, 731
fessional Eng'ineers. ,äff.",
zzThiele, E, W. and R. L. Geddes, Ind. Eng. Chem
25,289 (1939).
ACKNOWLEDGMENT
The authon acknowledge the advice and csistance of Mr. R. A. Leach.
of the Engineering Computer Depiltment of NarrEal ca;
b,i3:itl:"äg:",
72
COMPUTEBESTDISTILLATION
FEEDPOINT. . .
proper key ratio match point. Further, the proportions It has the advantage, since key designationsare dropped
äf h.uuy components on the feed plate are found to after tlvo iterations, of meshing all components at the
be excessivesince the amounts in the distillate u'ere feed plate and thus more accurately represents actual
set for meshing on the mth plate and computations tou'er performance than does the Mode I computation.
were continued past this point' The N * 2 assumptions
are then computed on the premise that meshing must Mode tV. Problems in n'hich the optimum feed point is
be accomplished on the m * lth plate. This calls for displaced from the plate on rvhich the ratio of key com-
reduced amounts of heavy non-key components in the porr".rtt matches that of the feed have been encountet-ed
distillate, the amounts being similar to those assumed ivith sufficient frequency to \varrant development of a
in the Nth iteration. Upon running the N * 2 itera- feed point optimizing technique.
tion m plates above the feed are again found and the For binary systems, the compositions of vapor and
computed quantities of heavy components on the feed liquid leaving the optimum feed plate are directly related
platÄ are too low. Oscillation in the number of plates to the feed composition. As a first assumption, in multi-
ivill continue until an assigned limiting number of component systems, the ratio of key components in the
meshing iterations has been reached. feedvapor (or liquid) may be taken as equal to the key
It might äpp"ut that this oscillation could be avoided by component ratio in the vapor (or liquid) leaving the feed
results in
slightiy altering the specified split. However, in cases -rnu.ry This criterion gives reasonably good
plati.
It has been observed to represent roughly
en-countered so far, this was not possible. Increased "tt"t.
the optimum feed point when the relative volatility be-
stringency in the separation merely shifted the oscillation
point so that, for example, instead of osoillating between tween the key components is significantly closer to unity
m and m * | plates, the computation oscillates between than either the relative volatility between the light key
m*landm*2Plates' and the next lighter component or the relative volatility
Integer plate number oscillation occurring in this between the healy key and the next heavier component'
fashion is äue to relative variation in the K values of When only one or neither of these conditions exists, the
key components stemming from composition changes' It key ratio on the optimum feed plate frequently has been
conclusively demonstrates the necessity in truly rigorous found to be appreiiably displaced from the ratio of keys
computations of using composition dependent equilibrium in the feed. Other factors may also affect the location
values. of the optimum feed Point.
Difficulties arising from plate number oscillation might Plates computed from the top down and from the
be avoided by determining the fraction of a theoretical bottom rlp t." plotted in Fig. 1 on independent ordinates
plate required for key ratio matching at the feed point' against the key component ratio in th9 llui-d for - a
iTo*"rr"., another method of coping with the PtoqgT tlpical multicomponent problem in which the feed, the
has been devised and incorporated in Mode III' This täwer pressu." u.rd an operating reflux ratio well above
Mode may be used if oscillation is discovered in a prob- the minimum are specified. In the case illustrated, the
lem submitted under Mode I. optimum feed poini falls at the point -where the key
In Mode III two iterations are mn under Mode I rätio in the vapor from the plate matches that in the
and the larger number of plates found, above, and also feed vapor (oi the vapor portion of the feed, or the
below, the feed is accepted and fixed, and operation is
',rapor in wiih a iiquid feed). It will be noted
"qrrilibti.r- plates from the top down the key
thät in cÄputing
transferred to Mode II. This leads, in some cases, to a
slightly more comPlete separation than -was originally ratio falls räpidly at first, then at a diminishing rate'
späcifiäd with respect to key comPonents but provides a Finally when carried far beyond the key ratio of the
practicable and rigorous solution to the problem. feed, ihe key ratio asymptotically approaches a limit rep-
Mode III can, of course, be used directly for problems resenting a very large number of plates, and has an ex-
in which no oscillation is either discovered or suspected' tremely small change from plate to plate.
74
COMPUTEBESTDISTILLATION
FEEDPOINT. . . With the optimizing technique the best solution, repre-
senting minimum plates for the separation specified and
A similar, but inverted curve results in the cornputa- the reflux ratio considered,is readily found.
tion of plates from the bottom up, a large change in An example is presented in Fig. 5 in which a hook
key ratio is experienced at the beginning and an ex- existsin the top plate curve making it impossibleto reach
tremely small change as an asymptotic limit is ap- the feed composition key ratio. The bottom plate curve,
proached.. The spread between the asymptotic lirnits however, continues well beyond the key ratio of the feed.
represents the range of possible solutions to a specific In executing the program this condition is detected by a
problem. With hydrocarbon mixtures such a range typi- slope inversion test. When inversion is found top plate
cally exists for any problem in which the operating reflux computations are stopped, bottom plate computations are
ratio is rvell above the minimum. The optimum solution initiated and the point of parailel slopes is found in the
is that which gives the minimurn number of total theo- usual manner. A problem of this type cannot be solvecl
retical plates. by direct convergenceon the key ratio of the feecl but a
When the key ratio on the optimum feed plate matches range of soiutions neverthelessexists and the ootimum is
that of the feed, as illustrated in Fig. 1, the tangent lines determinable by the program. Fig. 6 illustrates the same
to the curves at this key ratio will be parallel. It is type of problem but with the hook occurrinq in the
apparent that the minimum number of plates for any bottom plate curve at a key ratio belor.r,.
that of ihe feed.
particular problem will be found at the key ratio where A low reflux ratio problem is illustrated in Fis. 7
the slopes of the tangent lines to the curves are equal, which exhibits another variation in the interrelations
rvhether or not this key ratio coincides rvith that of the among plates, meshing and possible feed points. When
feed. the problem is introduced in the usual manner n,ith very
These slopesare designatedS,opr and ,SDorrin Figs. 1, small amounts assumed for heavies in the distillate and
2 and 3. Fig. 2 illustrates the case where the key ratio lights in the bottoms the first meshine iteration eives an
at the optimum feed point is appreciably below that of indicated optimum of the type shown in Fig. i. Ho*-
the feed. Fig. 3 shows the optimum feed point falling ever, with new assumptions respecting the distribution
at a key ratio higher than that of the feed. It will be of non-key components, in the seconcliteration the re-
noted that in Figs. 2 and3, the slopesof the tangent lines sults show hooks in both the top and bottom plate curves
at the key ratio of the feed Stp, and Söp are unequal and with no overlap of key ratios.
that the sum of top and bottom theoretical plates is In this particular problem, any attempt to distribute
greater than at the optimum feed point. In Mode IV, non keys to secure meshing at the key ratio where this
the program computes as much of each plate vs. key ratio first set of equal slopesis encountered will result in dou-
curve as may be required to find the point where the ble hooks with no common key ratio. Despite indications
slopesare equal and the number of theoretical plates is at this point, possiblesolutions may exist, as shown, the
minimum. entire range falling between the inversion points of the
The use of this optimizing technique is facilitated by top and bottom curves respectively.
entering, for the first iteration, small numbers for the The optimum feed point is found at the point where
assumed mole fractions of heavy components tJre slopes of the inverted curves become equal. The zone
^bottomsin the dis- of possible solutions may be appreciably less than the
tillate and light components in the since this
permits optimization for the first iteration to proceed full span between low and high invert points since it is
with minimal chance of overbuildins. However. if needed. limited to points at which complete meshing of all com-
the overbuilding routines will operate to enable the ponents is possible.
computation. Location of the optimum feed plate, as Problems of this typ€ are rarely encountered. When
found in the first iteration, is subject to redetermination they are, practical considerations usually call for higher
in subsequent iterations, but such redeterminations are reflux ratios. Consequently no attempt has been made
not apt to cause a shift of more than one theoretical to include in the program the capability of optimizing
plate. After each computation the key ratio on the opti- and meshing in this special situation. If such a- problem
mum feed plate is stored and used as the match point is submitted for solution, the existence of non-overlap-
for the next meshing iteration. ping double hooks is detected and the computation tä-
At reflux ratios close to but above the minimum a minated. The type of problem presented in Fig. 7 can
number of variations in shape of the plate vs. key ratio be resolved, however, by use of the program. Mode II
curyes may occur. At the true minimum reflux ratio only will compute the meshing of all components in this case
one point of feed introduction gives a solution and the with key ratios following the curves as shown. However,
ratio of keys on the feed plate may not coincide with the number of total plates required and the optimum
the feed key ratio. Consequently it is possible to en- feed point for a specified separation and given reflux
counter problems which, over certain ranges of reflux ratio must be determined by trial and error in successive
ratio, have real solutions but which p."r"ni no solution entries of the problem.
based on feeding the column on the plate where the key Typical curves of plates vs. key ratio at minimum
ratio matches that of the feed. reflux are illustrated in Fig. B. In this case only a single
mesh point is possibleat one particular key ratiq which,
A problem is illustrated in Fig. 4 in which a normal as indicated above, may be either above or below the key
curve exists for top plates but the curve for bottom ratio of the feed.
plates approaches the feed key ratio very slowly. In this
case an extremely large number of plates below the feed Computofion of Golumn Pressure. In the original
is indicated when closing on the key ratio of the feed. prograrn, column pressure was specified and the con-
76
COMPUTEBESTDISTILLATION
FEEDPOINT. . . practice to enter 1 X 10-'o for hydrogen in the bottoms,
particularly if a Iairly large number of plates below the
in order of decreasing volatilitn healy components in the feed is anticipated.
distillate and light components in the bottoms may be A convenient problem input form has been devised
specified as zero or left blank on the problem input cards. consisting of a single B% X ll-inch sheet on which all
If meshing is called for, the progiutmsrig.rs to such necessaryinformation may be entered. The form includes
components the starting value I X 10.-ro The split of all required instructions to the key punch operator so
key components must be specified by entering an arnounr that the time of technical personnel required to set up
for the heavy key in the distillate, and for the light key in and enter a problem is minimal. Once defined, a prob-
the bottoms. If for either key the other amount is omiited lem can ordinarily be entered in 15 to 30 minutes.
the program computes it. A starting assumption with
respect to split must be provided for distributed com- Computer Time. Core time has been considerablv re-
ponents falling between the keys. duced by numerous minor improvements since the de-
velopment of the original program. Over the last year,
The input formats permit entry of any desired set of
many problems have been run on IBM 7090194 and
starting assumptions with respect to distribution of non-
Control Data 3400 computers. In this application the
key components in the distillate and bottoms. Thus. if
speeds of these machines are about equal. A 17-compo-
reliable approximations are available in advance tirev
nent Hydrotreater Effluent Stabilizer problem with com-
may be used to minimize the number of meshing itera-
ponents ranging from hydrogen to decane has been
tions. Further, if a problem has been rr.n for say three
found to require about 2t/z minutes when run through
meshing iterations, the computed distributions may be
five meshing iterations. Satisfactory meshing of all com-
entered as input in re-submitting the problem, in which
ponents was achieved. In the IBM 7094 computer the
case, meshing iterations are continued without repeti-
core requirement including that of the processor pro-
tion.
gram and library subroutines is approximately 18 K, 36-
In the case of hydrogen in the bottoms the initial
bit words.
assumption of I X 10-'90has been found too high in
many instances. Any desired value may be entered and NOMENCLATURE
this will override the arbitrary assignment. It is the usual D molesof top product
Ad adjustment to amount of component i in W for new
assumption in meshing iterations
Ki equilibriuitr ratio for component i, (ya/x6)
About the authors
kt moles of light key component in liquid
WuNpnr,r, W. WarsBMaN is aice pre:si,-
dent of Pl ocess Dioision of The Katz-
rl moles of heary key component in liquid
number of theoretical plate above feed counting from
m.an Co., Youngstoutn, Oluio. He holds
top down
a B.S. degree from Iowa State (Jnüser-
sitg and, M.S. and, Ph.D. ilegrees in rv number of meshing iterations
chenvical engineering from (Iniaersitg süt slope of key ratio versus plates below feed curve at
of loua. He has been a process engi- key ratio of feed liquid
neer for Esso Eesearch and, Engineör- sr, slope of key ratio versus plates above feed curve at
ing Co. and in ntbsequent connections key ratio of feed liquid
he has -superui,sed process design for Sbo", slope of key ratio versus plates below feed curve al
plants in p et,ro c herndc als, inorgaruic
optimum feed point
chernicalsand, metallut'gg. Dr. Waterman is a membär of
ACS, AICLE anil the National Societg of professi,onäl Sto* slope of key ratio versus plates above feed curve at
Engineers. optimum feed point
tv moles of bottom product
tci mole fraction at component i in liquid
J.q.Mns P. Fnazrrn i,s a process design xIa,t mole fraction of component i in feed plate liquor
engi.neer for National Gas pipel,i.ne Co.
calculated from top down
of America, Chicago. He rbceiaecl a
B.S. degree i,n chemical engineeri,ng xl.,i mole fraction of component d in feed plate liquor
frorn lowa State Uni,aersitg and. i,s ä calculated from bottom up
membet' ,of the National Society of pro. yi mole fraction of component i in vapor
fessional Engineets.
LITERATUR"ECITED
P', HldrocarbonPtoccssing aut Pctro-
I"Y;"f"f:;,Yi.S;: SloträY,r"ä,1
,J.
llCyil, W. K., ud Matherci,G.'L., Iad. Eng. Chcm.24, 494([ßZ\.
8Thield,E. w., md Gedde,*.. Ira. E"i:eh";:ü: Is6-iröä5i1,'
R. H., Prc. Anerican_t.,
peircte,- rfütitu'tä,-vä.'+1,d*-.-31'esr_66
ir"näü:,,
'"hr:;:;.Ja*;,1"äur:e.tlrl',,Iöibfltttt'gsrev,
D. s., md Hoilad,c D.,
Gnoncp M. BRowN is prof essor of chem- 0Mills,A. K., Chcm.'Ens.progicss
i,cal engineering at Nmthwestern (Jwi- 55,No. 7. 95 0959).
-procissin| -
J. t., Hydrocärbon ini' Fi,äii,i ncfinn 42, No. e,
oersitg, Eaanston, Ill. Of lvis chemical lä"ifdäi.
EHanson, D. N.; Dufüa,
engineeri,ng d.egrees, hi,s B.S. ilegree was J. H., -and,_Someroille,G. F.. Cmputation of
lvlrltistage Sepaaiion Prcäeöes,.nqinnold-fülishir's
eEdison, Cr., N;'*Väk*-fidit2Y:
fr_om Uniaersitg of Michigan, S.M. from ^A. J- M. S. Tbeis in C.hlnicat e"ii"."k"J,'uä;1ü;i'öää:
" -E'ä:;
Ma,ssachusetts Institute of Technotogg chemicalnnsineeßHan-d6äaFi#äl 4rü
tf,E."$3$1
loRiedel.
1,.r,t;s I".'y'"
and, Ph.D. from Northwestern Uniaäi- J, C.. (Reoort on wok of Joha S, Bonaer), Oit and Cat lottttd,
5 5 , M a v 6 . 1 0 3 ' (i 9 5 7 ) .
sitA. Dr. Brousn has done process d,esi.gn
and, conffiIting i,n separation procesies ä3*"il;t#;,",;"1:r**"?;1ir6l$$i\l;,{..,,T;,*.l*'9
NGPA Convention, March 20I
.:'.f#ffi
and, thermodynnmics. He is i member
of AICLE, ACS ünd ASEE.
ff rix#''".1*1r5f
lndexjns_Tem.: B-onn_er-0, Chao_Sea_der-0, Columns/process/-4, Commsi_
,",rff
[ff1,.1ffi*#f*:o].f;[#ffi r'ä],ffi.
REPRINTED FROM HYDROCARBON PROCESSING 77
.*,
AnalogSimulationSpellsSafeStarfups
new refinery on an analog computer. The goal is to dem-
Analog simulators have long been used
onstrate major process interactions and their effect on
in operator training for normal plant operations.
operating conditions. But training for Plant type instruments tied into the computer serv'eas
the "contrbl house." This training program has proved
vital startup phase has been largely highly successful, as this refinery has been brought on-
ignored. Here is how the Humble-Esso stream smöothly. Thus, more training sessionsof this kind
are planned.
organization bridged the gap
Startup simulation is especially useful to an oPerator
o., a ,r"*, complex, highly heat integrated plant. He has
a unique problem becausehis plant has never before been
R. E. lieber, EssoResearch & Engineering Co., Florham started up and is inherently more difficult to operate
Park, N.J., and T. R. Herndon, Humble Oil & Refining than older plants.
Co., Baytown, Texas
There is benefit in extending such a simulation to cover
IN eN EFFoRT to provide experience before stattwp, startup-that is, bringing the plant up to design operat-
the techniques of dynamic, realistic simulation have been ing conditions. This would be of value not only in terms
used in a Esso Research & Engineering Co. training pro- oii-pto't eci operator training, but also in setting opti-
gram for experienced operators. These techniques involve mum startup procedures.
simulating operations of a highly integrated section of a There are big differences between a model developed
78
ANATOG SIMULATIONSPELLS
SAFESTARTUPS.
..
LEGEND
f l : Reoclor
t - D i s t i l l o t i oC no l u m n
P r i m o r yT e m p e r o l u r e
C o n l r ol el r s ( c o n t r o l l e r s
o n dr e c o r d e rosn
c o n l r o cl o n s o l e )
I n d i c o t eUs n i tH e o t
I nt e gr o l i o n
FIGURE l-This sectionof the refinerywas simulatedby the tains no intermediate tankaqe. Not shown: Four lS-minute
computer and_model (Six distillation towers, with präheat holdups (flow control on outEow); All other associatedequip-
exchangersand reboilers,and two reactors).Notes: Unit con- ment ( condensers,coolers,etc.) I Pressure,flow, level controlleis.
for normal plant operation and one that simulates start- o<:cupyingfour standard consolesof analog computing
up ope:"ations.After describing the normal operation equipment, was then connectedto a control console,
model. we will point out these differences.We will con- The control console has standard olant recorders and
sider what is involved in a plant startup and how this electronic controllers. Their input and output voltages
affectsthe computer rrrodel. are compatible with the computer.
The analog model and cäntrol consoletogether simu-
Anolog Model Adds Reolism. The section of the re- Iated the operator's normal plant environment. It also
finery simulated for operator training is shown in Figure gave him accessto similar records and indications that he
1. Included are six distillation towers. with r;reheat ex- would have in an actual control house. To use time most
changersand reboilers,and two reactors.For oi.er-all heat efficiently in the 10-day program, simulation was time-
economy, 11 exchangers transfer heat from one process scaled,speedingup the processby a factor of 10.
stream to another, rather than to coolins watei. The Recorder speedswere stepped up by the same factor.
plant has no intermediatetankaee.The onlv storasecon- Thus, recorder charts appeared familiar to operators. By
sistsof four l5-minute holdup diums with ihe out'flo* of taking advantage of the analog computer's tinre-scaling
eachon flow control. feature, many dynamic control situations could be inves-
tigated in a short time.
A mathematical model was developed for each major
Trainees were all experienced operators. So emphasis
piece of process equipment, describing its steacly-state
was on familiarizing them with operating and control
and transient behavior. The complete plant was pro-
characteristicspeculiar to the unit. The model save di-
gramed on an analog computer cöntaining about 400 am- rectional indications of systembehavior under upiet con-
plifiers and 400 potentiometers. This plant simulation, ditions. It also illustrated the time relationships involved.
amount
Slmulcting Process ltnlts. Even with the large
equipment used, certajn simplifications in
;i;;pJ"g
Since operator
the ma'themätical model were required'
on
,rui"i"g was the primary Purpo6e' .o.perability limits
some of the equipment were not consrctereo' FIGURE 3{omputer model f.or kettle reboiler simulation
For example, it was assumed that the fractionating (constant shell side temPerature)'
towers can hydraulically handle the design rates or -any
applied deviaiions. Therefore, the model did not include
con-
dumping and flooding correlations' Since pressurie
trol is rilatively fast and usually good, constant pressure The model was to represent daily opgr-alions' So, ex-
This
at the design value was assumed for the model' treme excursions from disign points would be rare'
allowed certain simplifying assumptions, such as linear-
Control console instruments were mainly recorder-con-
trollers for temperature and differential temPerature-
at ized perturbation models, tower control temPeratures as
the fiactionator towers, pre-heaters, and else- i"*,io"t of composition only and not of pressure, and
poirrts in
pseudobinary tower feeds.
iuhere in the heat integration circuit' Such local instru- '
ments as level indicator--controllers on low holdup vessels To simulate major items of refinery equipment we em-
were not simulated. Other recorders (i'e', flow) were sim- ployed existing ,igo.ot, digital models, from which
ulated by 8-channel analog recorders' iinearized transfer functions, were derived' These trans-
fer functions were then approximated with simpler ones'
In deciding how to simulate the dynamic behavior of
The simplifying assumptions were basld on previous op
refinery equifment, the purpose of simulation is very im-
erating öxperience with similar units. In some cases, ac-
portani. nät tttis piogtu*, th"-qai1 colcern was to build
tual pTant^data from a similar unit was used as a check'
irrto th" simulatiön äirectional indications that familiar-
Becauseheat integration affected the reboilers and con-
ize the operators with the idiosyncracies of -particular
-F 10 percent were tolerated' densers, distillation columns were isolated from the over-
units. Errärs uP to
80
ANATOG SIMULATIONSPETLSSAFE STARTUPS
... nored. These responses can be assumed instantar:eous
relative to the important control aspects of the unit-
head and bottom equipment in the calculations.Dynamic temperature control and liquid level in 15-minute holdup
plate-to-plate calculations were made digitally, assuming drums.
pressure to be constant (but not necessarily uniform
across the tower). Stortup Operotion. Shortly after the training program
The feed was broken down into pseudobinary compo- was completed, the refinery was brought onstream. Al-
nents. though some of the usual mechanical problems occurred,
Linearized transfer functions were obtained to relate the plant was smoothly lined out. No problems arose in
product compositions, overhead and bottoms tempera- areas the simulation covered.
tures, and control tray temperatures to such inputs as Similar analog simulations for operator training are
preheat, reboiler heat, reflux and feed rates, and product
now underway, no.: only for highly integrated plants,
withdrau'al rates. In simplifving these transfer functions,
but also for new processeswhere no previous operating
a balance was struck between program realism and com-
puter capacity. Depending on the type of column and experience exists.
its complexity, the analog model, Figure 2, contained 7 As plants become more complex with higher heat and
to 20 amplifiers and 10 to 25 potentiometers. process integration, the startup phase becomes more criti-
cal. With the successof the program just described, we
investigated how to simulate actual startup. There are
Reolistic Reboiler Simulotion. Simple heat exchangers several distinct phases after construction of a new plant
such as kettle reboilers with constant shellside tempera- is completed. These involve the pre-startup activities such
ture and no phase change in the tubes were simulated as commissioning utilities, running in of pumps and com-
rigorously. They required 7 amplifiers and 10 potenti- pressors, flushing lines, drying furnace refractory and
ometers. For the more complex exchangers, the dynamics steaming out towers.
were first derived digitally, then approximated by sim- After the unit is ready to go, circulation is started by
plifi ed transfer functions. pumping streamsinto various pieces of equipment.
_ Each heat exchanger was broken down by tube sec- Levels in drums are established.Heat is added to warm
tions. For each section, heat and material baiances were up the equipment and process streams. Inert gases are
programed to solve for tubeside inlet temperature to the vented and pressuresare built up. The startup progresses
next section 07, and shellside vapor generated V. as additional equipment is brought into operation in
order and the unit begins making product. These streams
_ The analog model for a kettle reboiler, Figure 3, solves
the equations: are initially recycled or sent to slop. The final stage of
startup consistsof lining out the operating units to bring
da- the products on specification.
-ucrur : hr Ar (0r- e*) * hs As (as- ar)
.il
Tr\-vL
-,",#+]
d P , - R Ie " - h ( * ) " , * - r * ) holds B.S. degrees in mechanical and'
electrical engi,neeri,ng f rom Southern
Methodi.st Uniaersi,ty. A regi,stered pro-
fessi,onal engineet'in Teras, Mq'. Hern-
*-+:-+[,",#t]
don has been with Humble 17 Aears.
He is a past director of Simulation
Counci.ls, Inc.
Prr : Pt- P,
82
TrainPowerStationOperatorsby Analog
N O . 3P A NA M N O I. P A NA M
STATION S T A TOI N
5 EorLERS 8 BOILERS
300,000LBs/HR. 200,000LBs/HR.
t,250L8SPSIG 4OOLBSPSIG
84
BY ANALOG. . .
TRAINPOWERSTATIONOPERATORS
* J .
passed through three turbines to produce 400-psi steam, few large processing units. So, the likelihood of sizeable
and through one turbine to produce 100-psi steam. Three swings in demand is much greater.
small turbines in the No. 1 Station (only one is shown
in the figure) use 100-psi steam and provide added oper- Byproduct Power. Power produced by the turbines
ating flexibility in special situations. when a process requires their exhaust steam is called
Waste heat recovery types of steam generating units "byproduct power." In this case the cost of the fuel
are located on refinery process units and beyond the con- represented by the latent heat in the steam is charged to
trol of the power station operators. These provide half the process.
the total refinery demand for 400-psi process steam. Such byproduct power requires about 31600 Btu per
Pressure reducing and desuperheating units parallel kilowatt-hour. But if condensing or atrnospheric exhaust
each of the two main sets of turbines. They are auto- turbine drives are substituted for the usual electric
matically controlled to maintain a set pressure in the drives on process units, the "non-byproduct power" thus
steam system downstream of the reducer. The desuper- produced could require as much as 45,000 Btu for every
heaters normally absorb the steam flow variations caused kilowatt-hour.
by changes in steam and electric demands. One can see that minimizing the production of non-
The line widths in Figure 1 show the relative mag- byproduct power will save money. Clearly, when the
nitude of steam flows in the various parts of the system. stearn-power demand ratio is tight, the reward for opti-
mum day-to-day operation is great. During such a period,
Opercting Problems. The station operators face two rnuch skill is required to handle some types of sudden
basic classes of problems. One class involves unusual demand changes, because minimum stearn flow is desir-
demand changes and equipment failures, when operating able through the desuperheaters.
conditions are tight and flexibility is limited.
The other involves ability to minimize production of Development of q Trqiner. Several years ago, an
"non-byproduct power" during day-by-day normal opera- intensive operator training program was undertaken by
tion. American Oil to improve power station operating tech-
In the past, process steam demand well exceeded that niques. It included both classroom and on-the-job instruc-
produced by turbines in meeting the electric demand. tion. The results were mostly satisfactory. But new opera-
And, enough steam flowed through the desuperheater tors required something more than the usual forms of
units to accommodate any sudden change in demand for instruction because they lacked the years of experience
either steam or electricity. Individual refinery processing of older operators. Simulation seemed a good answer.
units were small enough that sudden large changes in
system load were unlikely. Compufer Plus Gontrol Boqrd Equols Troining. An
Today, steam demand barely exceeds that produced analog computer facility designed to simulate refinery
by the turbines alone, and the refinery consists of a process equipment is located in a laboratory near the
operators.
'Th" STEAN
po*"r-station trainer features a graphic- instru- C0NTROLRATE D R A F TF A N
-""i-aitpfuy which follows, as closely as possible, the MODE SPEED
tvä,rt of equipment in the two stations' Record-
"irG.tf steam
ä* u"a indicators are provided for all important
änd electric power demands' Fig-
flows, steam Pressures,
tfto** tire electronic analog computer used to simu-
rr" Z
late the system. HEATER
R E C O RNDGI TION
CONTROL LE R
a
Gontrol Boqrd Setup. Figure 3 demonstrates how
and-a desuperheater are FIGURE 3-Here is how a boiler, trubine generator and de'
boiler, a turbine-generator,
t"!r"J""t"a on the"control board. The trainee has control r"p"tn*t"" are rePresentedon thä simulatoi's control board'
the fi"i""o can start.up, shut'down, and trim ouq)uts of each
koloUt t"t starting up, shutting down, and trimming
Switches adjacent to these boiler.
outputs of indivicluil boilers.
knobs permit the operator to switch from automatic to
**l.rui control of boiler firing. He can also choose low
or hieh speed draft fan oPeration. groups, change the dernands for steam, and cause the
Wf,et"rr"r possible, the simulator has been tailored to failure of selected station and refinery equipment'
produce svstem responses which "feel" real to the experi-
änc"d opätttor' The knobs controlling t'urbine governor The Trqining Progrom. -in Training on the analog simu'
adjustrnent, for changing the relative .loadings- -ott--lh" lator is condlucted two phases. Both are under the
tuibine-generatory prJsent the same action and "feel" as direction of an experienced shift supervisor'
their counterparts in the actual stations. In the filst phase, one boiler room oPerator and one
Even systäm frequency has been made to respond turbine room operator are together familiarized with the
realistically to load changes and machine adjustrnents' simulator panel. Then they ire gr,t"l- u standard set of
Behind thl board, compressed air is vented noisily when- typical operating problems to solve. Situations they must
ever header Pressures exceed the setting's on pressure ttä"at" include- l. E*".g"ttcy loss of a boiler under
relief valves. several conditions. 2. Losi of a turbo-generator' 3' Un-
A small panel out of sight of the trainee is used e:gected shutCown of a major refinery process- unit'
by the insti:ctor to vary the electric loads on feeder e. Uini-iang of non-byproduct Power generation during
balance'
-This of tight steamPower
a period
session lasts eight houn. The second phase is
much like the first. Höwever, the operators bring their
About the outhors own problems, questions, and experiences to the training
sessioir. All current trainees have completed phase one'
Lownr,r, G. Wnrrssntt' is a research
superaisor in the Resea,rchand Deael- and several have completed phase two'
oiment Department, Ameri,can Oi,l Co.;
Whiting, ina. I' grad'uate of Purilu'e
Uniueriitg, uritll B.S. and M.S. ilegyees- Resulrs qnd Gonclusions. Because the program is still
in electri,calengineering, he 'loineil rlrrd", -tn the full results of this training technique have
Ameri,conön 1950.He is engogeil in y* been evaluated. However, there are many gratify-
solai.ngengineering problems inaoloöng "oi
ing indications that the program is working.out well'
anatog sim,ulntöon,anil also consults on ä"ufog simulator tttiit-ttS does not eliminate the need
eleetioni'c asp ect s of instrumentation it
and control. He is a member of AIEE, for classioom and on-theljob instruction, nor does-
the value of many years of experience' But
Eta Koppa Nu, Tau Beto Pö ond Sigma Xi' detract from
a
it does provide a quicker, clearer means of geting
problem across to an inexperienced
E. II. Bowlns 'is o superintenilent at
Ameri.can Oit's Wlvi'ti,ng refnerg, where "or.rpt"*'operating
operator.
he has spent 29 years 'i'n the Uti.l:i.ties
Diaisi.on. He also has setten Aedrs e8-
The performance of the trainees and the questions
perience with Ingersoll-Rand 'ön sales th"r;k' reveal the weaknesses and gaps in their under-
bngineeri.ng. Mr. Bowles is in charge standing and previous training that were not apparent
con-
of the boi,ler house, ai'r cornpressor before. These'weaknesses point .wfti"tt up areas in the
hwse and, utilitA ilistribution at t'he need to be reinforced'
plant, and han(lles statistöcal stud:ies
ventional training Prograrn
are
inaoloing pouter station operations an-d Finally, turbine room and boiler room operators
equipment,, He has a B.S. d,egree 'in becoming more conscious of one another's problems'.It
mechanieal engi,neering from New York it- that a better informed, more thoroughly
Uni.aersitg. "pp"t?"t ##
rorrnd"d-orrt crew is being developed'
86
MinimumRefluxFiguredan EasierWay
Dovid [. Ripps
(e")'- (e)"
American Cyanamid Co., Wayne, N.J. (o),+r: (e), + (6)
| - @e,/de)
To cer-cur,erB minimum reflux ratio by the Under- For efficient solution we u'ill use the previous two
wood method, it is necessaryto solve the following equa- in order to evaluatethe derivatives.
c1'cles
tionl for O:
4 ! q " _( e r ) , , - ( e . ) , , -' tt i. -- _ '(' ) r?' " ' /\ '7t \
dg- (e), - (O)"-t
s-{
o ' r ' ,- ( l p ) : o (l)
u
" @
\ - /\ : )',
4at_O
+ To start the procedure n'e rvill take O as the mid-point
2 - 1 between a7 afid. ara.
where ai : relati\re volatility of component i
(O)t : 0.5 (a, * o" + r) (8)
xr : mole fraction of component i rn the feed
.y' : thermal condition of the feed \\Ie assumed originally that the root in the interval be-
/ : number of components
tween a/ and a"*r is closer to a' than to 47a1. Evaluating
Equation 1 is highly non-linear wit}t a singularity at the function at (e)1 provides a test of this assumption.
each value of e : cl. The roots of interest lie in the open If U(O)' is positive the assumption is correct and rve
interval between adjacent pairs of c's. Because of the may proceed. If that function is negative, however, the
strongly non-linear character of Equation 1 a direct ap- root is closer to a711ärld we should replace r by r-ll in
plication of the standard methods of iteration (viz', New- subsequent parts of the iteration.
ton, false-position,etc.) often fails to convergeor requires To continue we use Equation 5 as a direct iteration to
an excessivenumber of cycles to reach a root. We rvill give the second starting point. Thus
now develop a specific method for solving Equation 1 in
a reasonably efficient manner. (o),: (o.)'
Thereafter rve rvill use the modified Newton given by
In this discussion it will be convenient to number the
Equations 6 and 7.
components so that the c's are in ascending order.
where
0, (e) : -e!'r-^
d " - w
- U (e) @)
Example 2. Four component sl'stem.
Equation 3 may be solved for O by treating {r &$a con-
stant. Since {, is actually a function of O, the solution ar : 0.3 ar : 1.0 az : 1.3 at: 3.3
must be iterative. Then p: -r
(o")o : ', (, -,-) (s) I)esired: Root in interval 1.0 < e < 1.3
ToleranceonO * 0.00001
u'here
(e)" : guessedvalue of O at cycle z of the iteration
(e,)" : computed value of O at cycle n of the iteration
o.75 0.10
Expanding Equation 5 in first-order Taylor seriesgives o.20 0.05
88
TABTEl-lnput Dqta
o
T S - t l 2 9 L l 3 3 l r
r zq 4 I 26 I o 0 0 0 0 0
T6-...-trF;örTZ---a;3--
-ä
+-#ffi
o 2 | o I t o ll I 35 334l.to 4 o'o
t+ 16
1 Z Z
i i
,: ,"t".g9 i
t z 2 i 4
-, 4 2 - 5 t t 2 z o 2 t -T"'4,
f f i
i
".", o o, , , r, , +
i c ii rr o o o o o o o o
'i
.r""ää )ä z" to rr 32 ro 22 42 te
,; 'e .b .1 2
+ ,rj,-'zrl ß :t
DesignBestCoolingWater SYstem
diameter of pipe for required flow and maximurn
This computerprogramhelpsdesig_nimum velocity. The system thus designed is the best design if it
an econoiricalcbolingwatersystem.lt can ruiirfy the third requirement of available head, but
eivesthe overallinstalledcost of the usually it does not. Therefore' some lines in the system are
Fipine and can be used for sizing tovelocity)
be increased in size (within the range of minimum
such that available head loss conditions are satis-
the pump fied.
Which lines should be increased? The answer can make
the difference between the best and the worst design. If
Jogindro P. Kohli*
the lines that are increased cost the least, the design is
Union Carbide Cotp., Tonawanda, N.Y.
the most economical design. It is not an easy job to find
Trrrs coupurER pRoGRAMis based on the installed cost which lines cost the least to increase (especially in a large
of the piping for a cooling water system at a given loca- system). For example, a line of a relatively smaller diam-
tion. For given allowable pump head, pipe routing and eter with two valves may cost more to increase in size
maximum and minimum velocity limits; it designs the than a larger diameter line of some length but without
most economical piping system. The program can also be the valves. This is where the computer can be used to the
used for sizing the PumP, calculating head loasesin an greatest advantage.
existing system,and computing the cost of a given system. The design of a cooling water system (or a water sup-
Typical results from the computer prograrn are presented' ply system! involves a trial and erro-r procedure. In the
normal pröcedure, the lines are sized approximately for
Optimum Design. What is a most economical design? the given flow and the friction head loss is calculated'
It is the design which satisfies all the required conditions This friction loss is added to the various other head losses
and costs the least. (equipment head loss, cooling tower head, etc') - and
For a cooling water systef,n,the required conditions to checkid against the available pump head. If the head loss
be satisfied are (1) flow, (2) maximum velocity, (3) calculated for the system is more than available pump
available head. Flow through a pipe is equal 'to its area head, some line sizes'in the system are increased and fric-
times'the maximum velocity and therefore two of the tion head loss recalculated. The procedure is repeated
three requirements can be satisfied by calculating the min- until the total head loss through the system becomes less
iil..ot-"adt.o: Statc University of New York at Bufiato, Amherst, N.Y.
than the available head.
90
TABIE 2-Recults of Cool:ng Woter Plplng Derign progrom
.o Computing the cost of the system thus designed and
printing out of the results.
CIRCUITNO. HEADLOSS(FI.
2 87.O Initial Sizing of the System. Initially all the lines are
sized to satisfy flow requirements with maximum velocity.
5 18.'l
6 8?i6
1
8
62.3
86.2 Area : Flo{ Velocity
9 65.3
l0
tl
82.3
?1.3
rD2/4 : (r44 (0.r337)Q)/90AMV)
_ 12 8I.5
I L L o H A B L E P t J f r PH E A D = 1 0 8 . 0 C 0 0 L l N G T o H E R H €A t ) = 2 0 . 0 0 where
INSTALLED COST 0F THE HH0LE SYSTTM lN DOLLARS E 290?6.93
D : pipe diameter, in.
HE PROGRAM
Q : flow through line, gpm
iAXIßUü AILOWABLE O E S I G NV €L O C I T YF L O H 4 5 G P I A N t , l O R E = I 2 . 0 0 F T . / S E C .
X A X I i l U i lA L L O W A B LDEE S I G NV E L 0 C I T YF L O I L E S S T H A N4 5 G P M = 1 0 . 0 0 F T . / S E C .
AMV : allowable maximum velocitn ft./sec.
A V €R A G € L A B 0 R R A T E F o R I N S T A L L E Dc O s T c A L c U L A i I o N s = 7.Oo $/HANHoUR
Then
! N t J A L L E D C O I T U N D E RG R o u N pP I P t N c l N O o L t - A R S : 21366.98 --.--
D : r2\/o.tß7 aF.1 ^* (l)
I N S I A T L E D C O S T A ß o v E G R 0 U N DP I P T N G I N D O L L A R S= 7-öts
I X S ' A I L E D C O S TO F T H E W H O L €S Y S T E MI N D O L L A R S= 290?6.93 -,- The lines thus sized are the minimum sizes for given
maximum velocity conditions.
?ABLE 3-Heod log Through rhe Circultr Head Loss Through the Circuits. After lines are sized,
aRANCh FlOr Ora xtloross
head loss through each line is calculated by means of the
Nn. @
G,p.il. ItcHES FEET ootLAks c o s _ l ! _ l D o t L l R 5ct o s T l n l L L ^ R 5 ) following two equations:
---
Nze
For length of pipe (Hazen Williams Formula),
HL: (0.002083)
(Qr) (Q4'"u11pDc1n.tuu, (2)
Where
_+#_ HL : friction head loss, ft.
QZ : length of pipe, ft.
QF': flow through pipe, gpm
!E
19
055.89 6.000 12.01 9{8.73 -- 2A?.30 t6l.4f PDC : pipe diameter, in.
381,90 4.000 4.a4 z::.is lom 123,68
?0 l?!.99 4.000 2.?4 3 ? ? . 9 ! -------- ztl& 144.80
21 630.50 6.000 !0.44 e8a.l- .9r.25
For fittings in the pipe,
ii -.Iä..H+*+f--- HF : (0.0025e)(r) (QJO(QFf/(pDC)n (g)
--#;#- Where
---s:F- -""+.\+-
_ "aIü"_ HF : friction head loss, ft.
_+# _ K : constant for different fittings
drops in the loop mains. This l. Collect Data. Pipe diameters, pipe equivalent lengths
(i.e. including length for fittings.), system configuration
method reduces the calculation time and average fluid properties'
from days to minutes 2. Estimate Flows in the Network. The flow rate of
fluid in all the pipes in the system is estimated. The only
criterion for this estimate is that the chosen flows must
B. Wesl and A. J. Newton be consistent with the over-all mass balance. It is pref-
Polymer Corp., Ltd., Sarnia, Canada erable, however, to make reasonable estimates, as this will
reduce the length of the calculation.
IJrrr-rrv DTsTRIBUTIoN sysrEMS for refineries and chern- 3. Calculate the Pressure Drop in the Pipes' Any
ical plants are becoming larger and more complex. Usually standard pressure drop calculation is used. fn the com-
these systemsare looped, so that the utility supply to any puter program, the Polymer standard procedure is used'
unit is not dependent on one path through the system.
As a result, complicated piping networks are built up.
A major plant expansion will normally include a study
of the utility distribution systems,to determine the efiect
of the load increase, and to indicate where expansion is
required. These studies when performed by hand are
very time consuming and prone to error because of the
large number of repetitive steps involved.
The repetitive nature of the calculations naturally leads -r:
K = l " x s c'
to the use of a computer method of evaluation. Com-
puter prograrns 2, 3, 6, 7, B have been written to solve
this tlpe of problem.
The following is a description of the network analysis
program developed at Polymer Corp.
94
SIMPTIFIED TOOPBALANCING. . .
UTILITY
!," - ?,,- kcr(cF)(vol t (fu) O (4)
/ ' \
FY - \ Y / +2.50(InW) -0.76 (InR) -0.207(tnR)2
1.03
E:
d(FY)
dv (8)
96
SIMPLIFIED LOOP BALANCING. . .
UTILITY The service water distribution system consists of 41
mains and 10 loops.
2 , 0 0 0U . Sg. p m A typical computing time is 15 minutes.
No specific times of calculation by hand are available,
u.s
1,000 3,000U.S.gpm but several days of continuous work were required to pro-
duce results of comparable accuracy.
A recent paper8 gives an excellent description of a much
more rigorous method of solution to complex pipe system
oroblems. The method has obvious attractions for use
in systemswhere the fluid conditions and properties are
1Q000U.S.gPm 4 , 0 0 0U
r'vvv g. p m
. Sg!'rll
v'r' subject to considerable change. The simple procedure
presented here, appears sufficiently accurate for fluids at
Fig. 2-Sample problem. steadyconditions.
ACKNOWLEDGMENT
TABTE l-l6pul Dctq for rhe Scmple Problem The Pol>mer Corp. Managementis duly acknou,ledgedfor the permission
to publish this article.
MainNumber Diam. (in.) Eq.Lgth. (ft.) Est.FlowUSgpm
NOMENCLATURE
I t2 1000 5000
2 10 600 -5000 AS Algebraic sum of pressure drops in a loop
2
t0 500 4000 CF Compressibility factor of fluid
+ r0 1200 -5000 D Diameter of pipe, inches
10 1500 -3000 f
Weisbach-Darcy friction factor
6 10 200 4000 f1 Fanning friction factor
7 10 i000 1000 FY Friction factor test variable
Newtons-law conversion factor. ft.lbs. massllb.
force-sec.2
TABLE2-Computer €alculoted Vclues of Flow cnd G Mass velocity, lbs. mass/ft.' sec.
Pressure in Each Loop lloin H Pressure drop in pipe, lbs. f.orce/in.z (lbs.-forcezl
in.a) (units for compressible fluids)
Main Number Balanced FIow, USgpm Outlet Pressure,psia k,, Constant for noncompressible fluids, lbs. force in.5
min.2/ft.3 USG2
I 5934 68.2 k c Constant for compressible fluids, lb. force2 in.5
2 -4065 80.7 oR
hr.z/ft.8 lb. mass
4934 J.t.o
^ K Constant for system and fluid
-4065 34.6 KL Absolute roughness magnitude
5 -2235 16.4 Le Equivalent length of pipe, ft.
6 4764 24.1 M Total number of loops
7 1764 16.4 N Total number of pipes
o . fnlet pressure, lbs. force/ft.2
v'2 Outlet pressure, lbs. force/ft.2
calculation would take at least one hour. The advantases o Volume or mass flow rate, USgpm or (lbs.
of the computer are greatly increased as the size of ihe mass/hr.)
system is increased. One hour's computing time replaces R e Reynolds no.
two weeks or more required for hand calculations on the s.G. Specific gravity of fluid at flow temperature
big utility systems. T Absolute temperature, ("R)
VO Specific volume of gas at standard temp. and press.
(ft.a/lb. mass)
Somple Problem. Determine the florv rates and pressures p Fluid density at flow temperature, lb6. mass,ftt.3
in the network shown on Fie. 2. LQ Flow correction factor, USgpm or (lbs. mass/hr.)
Table 1 gives the main numbers in the loop and tabu- E Friction factor correction factor
lates the diameter of the main, the equivalent lensth, and
the estimated flow in USgpm.
SUBSCRIPTS
Table 2 is the computer calculated flow in each main
and the outlet pressure. i Pipe number
The time taken to complete the problem was as follows: j Loop number
?n Over-all iteration number
Function Time I Friction factor iteration
number
jn Number of pipes in loop j.
Prepare data 10 minutes
Punch data 2 minutes LITERATURE CITED
Computer solution 3 minutes I Pipes, Louis 4., Apptied Mathematics for Engineers and Physicists.
,IBM Public Utility Depdtment, Gö Network Program.
3 Bartholet, C., IBM Library Program.
15 minutes aDukler, A. E., "An Analytical Expresoion for Friction Factor," AICLE
1.,
Vol. 7. No. 4 1961.
6 Hady, Crcs, Bull. 286, Univ. Ill. Eng. Expr. Sta. Urbana 1936.
Hqnd vs Compuler Cqlculotion Time. A steam dis- 6 Kneibes,'f'low
Wilmn. "Dicital Cmputer Solution of Gc Dirtribution Svstenr
Network Prbblems-," CEP, Symposium Series No. 3l Vol, 56, lS0.
tribution system consists of 29 mains and 7 loops. I Hunn, Mclntyre, Auitin, l'Evaluation and Design of Pipeline Networks,"
CEP Svmposiuü Series No. 37. Vol. 56. 1960.
A typical computing time for this system based on a 6 Daniel, P. T., "The Analyiis of Compresible and Incmpresible Fluid
reasonably accurate initial estimate is 30 minures. NetworLs," Tranrctions
No. 2. March 1966.
of the Institutio-n of Chemiel Dngineen Vol,44,
:_
C"Z -1- C n o
__
c*16 -Y4 TY- D
^ - /, { | - t(KDGü/r")9,) (l-rÄl-if *- ' )r\r ,
+Cre , I
98
OPTIMUMABSORPTION
DESIGN. . .
When a centrifugal compressoris The efficiency and head coeflicient* are defined by
We
specified for nonideal gases E _ Q.0ol2B54) (4)
H"- H,
and the manufacturer's data is based *_2(60)2e.W, (5)
rT2D2N2
on air, how do you predict design
where the polytropic head, Wp, is defined
and off-design performance?
we_Q44) (Polytropicpath) (o)
This calculation procedure solves l+
the problem The properties appearing in Equations (1) and (3)
are to be averaged over the compressionpath. The tem-
perature equation, Equation ( 1), requires an efficiency
Stephen A. Shoin which is an appropriate average over the path of the
Shell DeveJopment Co., Emeryville, Calif. compression,but the pressure equation is only valid for
a constant efficiency path.
Trrr, spr,crrrcATroN of a centrifugal compressor and
prediction of design and off-design performance are parts Dimensionsless Numbers. Dimensional analysis of the
of a problem that continually arise in process design compressiblefluid flow equations?indicate those variables
studies.To make these calculations less difficult, a com- and groups of variables which are important in specify-
puter program was written which handles very general ing the performance of a centrifugal compressionstage.
situations including multistage compression, inter- and Applicable equations are the equation of of continuity,
intra-stage leakage, multicomponent gas streams, depar- the momentum equation, the energy balance, and the
tures from perfect gas behavior, intercooling when nec- equation of state. These equations must be solved subject
essary, and allowances for manufacturers' data on the to specifiedboundary conditions.
basis of air to permit design for hydrocarbons and other The solution to the equations is controlled by the
vapors. magnitude of the following dimensionlessgroups:
The equations used for the individual compression
The Grashof number
stages of the compressor are summarized in the sections
^I - 9ßLTL
which follow. The assumotions and limitations imolied (7)
in the general efficiency and head coefficient correlaions
used, are discussed.The equations given are more general The Reynolds number
than those usually appearing in the literature.l,4'6'a,e,12 UL
'^t t -
Re:
-t- (8)
(3)
zRfr
MC ,L; .(n:),1 *The definition of the head coefficient, Equation (5), involves a factor oI
2 which is omittcd in some sources.
Two compression stages will have the same solution
in terms of appropriate dimensionless variables provided
certain side conditions are satisfied. First of all, the
boundary conditions in the two situations must be iden-
tical. Second, both compression stages must be geometri-
cally similar, that is, the ratios of all dimensions must
E
be identical, and third, the equation of state for the
a two gasesundergoing compression must be identical.
,^,w
r -U
":4 (l2) and the formal relations for the efficiency and head
coefficient and the flow coefficients are:
The relations of thermodynamics may be used to flows, but correlations in this form are not always avail-
transform this equation to a form rvhich allows direct able from the manufacturers. The example given is
use of the equation of state: based on inlet volumetric flows.
For impellers which are geometrically similar, the
, ,
|
t . , , , ,
/ ;" r" \I area (A) is uniquely related to the impeller diameter
"":{r**e"n\ae), (r2b) (D).However, the value of ,{ has been left free so that
it may be adjusted to improve the efficiency and head
The final relationship fpr the speed of sound in a coefficient correlations or to put them in a convenient
nonideal sas is: form, particularly if geometric similitude is not main-
tained. For example, the flow coefficient may be inter-
"
t/ _ 1 |+Ag;TZFT-
(r?c) preted as the ratio of the volumetric flow to the manu-
!/, [,_(##),]
facturers' design volumetric flow if a nominal area (A) is
taken to be
O (designpoint)
The calculation makes use of Equation (12c) evalu- , _ (14)
nDN
ated at the inlet of a compressionstage.
This form is often convenient since the manufacturers'
__ln the computer program, the characteristic velocity correlations are commonly in terms of percentage flows
(U) is taken to be the impeller tip speed. In addition relative to the designpoint, i.e., Equation (14) fixes f : I
to the above groups, the equation of state and the at the design point. A similar definition of a nominal
boundary conditions are implicit in determining the diameter (D) rnay be obtained from Equation (5) for
solution. V : I at the designpoint.
l02
Miscellaneous specifications-including the rotational
speed for each frame, and several boundJ and increments
which must be specified.
The output from the program is a reiteration of the
input data and a statement of the calculated properties
for the inlet stream, i.e., the density and the volumetric
fow. A listing of pararreters and properties for the out_
Fig. 3-Flow diagram for a tbree-stage compressor. let stream of each compression rtage is presented. These
include the stream number, the frame number and im-
peller number, the nominal diameter and area for the im-
3. Calculate physical properties for the suction.
peller, the rotational speed, the temperature, pressure,
4. Initialize stage discharge physical properties. density, molecular weight, the volumetric flow, and
arithmetic average value of the specific heat ratio, the
5. Calculate appropriate average flow coefficient. value
,of the exponent which defines the compression
path (given by Equation 3), the value of the äverage
6. Calculate head coefficient and efficiency. flow coefficient, the Mach number, Relmolds numbi,
7. Calculate new discharge temperature and pressure. head coefficient, efficiency, the head and horsepower.
When a stage consists of intercooling, the temperature
B. Compare current and previous discharge ternper_ and pressure of the stream is presented, and, if phar"
ature. If closure to within a specified tolerance is separation occurs, molar flows of both the vapoi and
not achieved, return to (5). If closure is achieved, the liquid stream are presented.
calculate the next stage starting with (a). Aftei In addition to the calculated properties, a warning
the last stage, continue with (9). flag. FLAG, is indicated for each compression stage.
Values of FLAG are used to indicate the course of the
9. Print out final results.
computation and signal any difficulties which may have
Figure 2 is a simplified diagram of the subroutine been encountered during the computation. The warn-
which calculates the performance of a single smge. ings which may be indicated by FLAG are:
o The Mach number exceedsthe critical Mach number.
Use the Computer Program. The input data re-
9f
quired by the program can be divided into the following o The average value of the flow coefficient is less than a
categories: minimum value.
Component data---consisting of a specification for o The average value of the flow coefficient is larger than
vapor components in the form appropriate to the sub- a maximum value.
routines which calculate the compressibility f.actot, Z,
o The number of iterations for the calculation at this
and other physical properties required for the compu_
tations. stage has exceeded the maximum number of iterations,
i.e., the calculatibn has failed to converge.
The equation of state used in the example is based
on a modified version of the Martin and H-ou equation o The calculated head coefficient is negative (may result
of states'10 from inappropriate extrapolation of correlation).
RT Ar* B"T + Cre-rrtr"
fl - l-=-
-f- o The calculated efficiency is negative (as in head coef-
\ a - D ) (a- 612
ficient above).
A"*BrT+C"e-xrtr" An B^T
, . fntercooling has resulted in condensation.
(a_b1+ '
- I
\u_o) (u_b|a
o The temperature has exceeded the maximum allow-
frnpeller correlation data-specifying a correlation of able temperature.
efficiency and head coefficient of the impeller or impel_
lers with the volumetric flow coefficient. (The corräla_
tion may be based on inlet flow, outlet flow, or anv Exomple Calculofion. An example is presented on the
average of inlet and outlet flows from a compression application of the above procedure using a large scale
stage. Variations in the efficiency and head coäfiicient digital computer. Calculations are made for a three-stage
with Mach number and Reynolds number may be in_ propylene refrigeration compressor. The shape of the
cluded, if desired.) head-capacity and the efficiency-capacity relations was
taken to be the same for all three wheels, with the flow
Compressor sizing data-consisting of a list of impeller coefficient based on inlet flows. The head-capacity rela-
scale parameters or physical
- dimeÄions to be used in tion for the second stage was used as supplied by the
the correlations for efficiency and head coefficient. manufacturer. The shape of the curves for the first and
(These usually are a charactäristic impeller area third stages differ by 2-5 percent at the extremes in the
and
diameter, but other variables which refläct the capacity range of 60-140 percent design capacity. An efficiency-
of the impeller may be used.) capacity relation was not available for the wheels in
this machine. The manufacturer was able to supply an
Stream data-consisting of a specification of the inlet
efficiency-capacity curve for a similar machine at a
stream temperature, pressure, and molar flows for
each specific speed which .was close to the design condition
component.
at the second stage. This curve was used to calculate
REPRINTED FROM HYDROCARBON PROCESSING
r03
. .
PERFORMANCE
BESTAPPROACHTO COMPRESSOR
Ll00 Percent Propylene, b-For feed flow, temperature, and pressure. c-1.02 x Compression hp'
ACKNOWLEDGMENTS
an efficiency-capacity relation for the range in capacity
;Ir-ilö-1öG;;;"t;",r'"
of trom bU-I'IU percent ol the cleslgn poi,,-.'.
J"'iä"polnts. Agaln, tnä suppii"a
Äguii,rne ,"$\i"o'1'1""0ff*oi*"öi'bn-*i.ö:1'ifiil""di
the programs {nd _correlations used H-"Sip*'.!'":fi1ri{i
p . o p e r t i e s . -ä r d o i M r s . D . J . A l b e r t a n d Mr'
for estimation of ,..Plv1Yl
M r . E' L . Necechea
E . L' w h o pro
N e c e c h e a who pro
r | - -r:ff--- L - ^---^--:-^r^r-, R -^^-^^n+ ^+ +I ^
wheels may differ by approximrt"iy 5 percen; a! tl-re iin:t-',"t";{."",1.Yi:'r3 J,;*Hffil
extremes of the range. The flows in the recycle (leak-
age) streamswere taken as 1.5 percent at each stage and NOMENCLATURE
6.5 percentfrom the dischargeof stage3 to the suctionof
stagl L Figure 3 is a scheäaticdiigram i.,ai*ti"g th" i" ;#täii1!}'.ro..lj.,""r Btu/lb.mass-oR
pressure,
interstage flow, stream 12, and the intrastage flows,
;o nominal diameter, ft.
streams 8. 9. and 10. g local acceleration of gravity, ft./sec. sq.
The results of the computations are summarized in ;5 c eravitational constant,32.16 lb. masr-fi.7lb.force-sec.sq.
D"
Table 1 and are compared with a performance test. H enthalpy, Btu/lb. mass
The discharge pressures of 312-276 psia and power k heat capacity ntio, Cr/C"
requirements of 2949-3107 hp are somewhat above the L characteristicdimension,ft.
Zli-ZOO psia and 2,400-2800 itp tu.tg.t from the original exponent' defined bv Equation (3)
?- mass/lb' mole
the äalculated valueJ are M
design specificationr. Ho*"rr"., i:l:::]iT ::ltl'il
.r*ä-i.',rt"r" ir"- ,rt" testsfor this {, lo^t,]^t::T]j|"^:d',T:
"ctual'performance '^"r'o
Nrn lcrsrL'u'ucr' defined by
Eckert number, ua''Eu u] Equation (10)
!eua!rv. \rvl
machine. The calculated discharge pressures (third
I -r rr-^ ^^+,.^1-, !';T"
\ cr Grashof number, defined by Equation (7)
stage) are within 1-3 percent of the pressures.actually Machnumber,definedbyEquation (12)
achieved with even closer agreement for the brake horse- N:: prandtl number, defined by Equation (9)
power. The agreement is excellent considering that man- N'/" Reynoldsnumber, definedby Equation (B)
ufacturers' head and efficiency data for air has been P pressure,psia
used to calculate performance for propylene under con- a volumetric flow, cmf
1.987 Btu/lb. mole-oR, or 10'73
ditions where it is a nonideal gas. R ga"s constant, equal to
psia-cu. ft./b. mole-'R
mass-"R
why This Approoch? The approachpresented,,l",'""]: i ä,li3J;,3,lYrb'
intended as a thermodynamically lor
consistent method n characteristicvelocity, ft./sec.
performance calculations when nonideal gas phases are v velocity of sound, ft)sec.
"
important. It is hoped that the presentation -will .en- i" polytropic head, ft.Jb. force/lb. mass
courage publication of further experimental results which Z' compressibilityfactor
will allow evaluation of this approach and outline more reek svmbols
clearly its limitations. The urtho, would like to empha-
p coefficientof thermal expansion'vol fract/oR
size to those concerned with design, manufacture,'and
e polytropic eficiency' defined bv Equation (4)
performance of such compressors that methods and com- thermal conductivitv' Btu/sec'-ft'-'R
I
puter capabilities for evaLuation or tn".-oaiiä;i;pd
ärtiesor nuidsare becoming J';iäi"';ä
generally t ä""?'l';J?i,äTliji.ii't',./,.".
should extend the usefulness of many design methods' n constant,equal to 3.14159
The physical properties programslo used here have been p density, ib. mass/c,r. ft.
made availabl,e tä the Fhysical Properties Project Sub- A volumetricflowcoefficient,definedbvEquation (13)
committee of the American Institute of Chemiial Engi- t polytropic head coefEcient,definedby Equation (5)
neers. Subscripts
1 refers to compressionstage inlet
2 refers to compre'ssionstage outlet
About the quthor