Professional Documents
Culture Documents
ll ion
( :OiUlIlIh ;lIld (:tlt'lIli{ ;1 1 R e; lct()r ....
BOUrJll·IIHllllh. L"f.\. I~};-';Ii
Abstrac t
Starting from famil iar assumptions for dynamic analysis , a transfer-fun c ti on-matrix
(T .F.M) model for long, symmetrical, binary distillation columns is derived
completely analytically. The model relates twin output composition changes to
perturbations in the internal liquid and vapour flow rates. The model is intended for
use in controller design and, being expressed in terms of plant parameters and operating
cond itions should provide useful common ground between con tr ol engineer and plant
designer . Though derived precisely for a symmetrical plant , it is expected that the
model should apply more ge nerally as a good approximation. Experimental data is given
supporting this claim. Inverse Nyquis t loci are generated in precise and approximate
form and compare favou r ably with computed step responses . They accord broadly with
previous, part - empirical, part-numerical research results. A second companion paper
is devoted to packed-co lumn analysis .
Keywords. Chemical variab l es control , Distillation, Fluid composition contr ol, Freq.~esp
~, Modelling , ttultivariab le system~,Nyquist criterion, fartial differential equations.
189
190 J. B. Ec!l"ards a nd M. H. :"I. Tabrizi
and if v and t are kept sufficiently small, steady- normaising to: aI TaY(O) } aYI (23)
s tate values may be substituted for all the capital =E( l-Y (o ) }+-
a, oh h=o
variables in (7) and (8) thus yielding linear rel-
ationships between y,x' and v, t . a {TbX' (o) }
cons train ts
where Ta = Ha / H£O h and Tb = Hb / H6h
Solution of pde's (7) and (8) is practical only if
they and their boundary condi tions are rendered Ha and Hb being t~e molar capacitances of the acc-
symmetrical by appropriate choice of plant para- umulator and r eboller.
meters and quiescent operating condit ions. \~e ther-
2.6.2 Feed-trays. A mass balance on the rectifier
efore make L' and oh the same for the rectifier and
stripping section and set feed tray yields
Eq. (10), (13) and (14) we deduce that Eqs. (32) - (36). Their symmetry allows the ready
diagonalisation of the system if we adopt the out-
F ~ E V (27)
put and input vectors:
so that substituting for F and z in (26) gives,
)l(h,C) - x' (h, c)] (37) and
after normalisation q(h, , )
[ y(h,c) + x' (h,,)
~(L) _2_ + x' (L)+ \l-Y(L) }- -aY\
(28)
a, Cl+l h=L oh rv(,) + £(,)l
tiC ,) = Q (3B)
and, similar treatment of the stripping section's - V lv(,) £ (,)J
feed tray yields
Lapla~e transforming the pde's in s w.r.t. h and in
,.z,~. :JL)= _2__ X' (L)- {l-Y(L)} + oX' \ (29) p w.r.t. , gives
h Cl+1 ar;- h=L
The results are necessary for substitution in the and inversion back to the h,p domain, and setting
solution to our linearised small-signal model (lB) h L therefore yields
and (19) but have a strong practical appeal in them-
selves . The constancy of G, leading to even tray ilL) - Ql(L)i(o) - Q2(L)~ = 0 (42)
loading, has already been anticipated and discussed where Ql(L) = {ipcosh/p L + (E+Tp)sinh 'pL}!/lp(43)
but equation (31) represents another important att-
ribute of the special case we have adopted. It
indicates that, from a 50/50 mixture of components an d Q2 (L) =
cosh/p L-I
-"-"-::":":"p-'--"-'::'
- E 0l (44)
(z + Z = 1.0), a symmetrical plant and operating [ o a+lJ
regime produces top and bottom products that are Now the feed Eqs. (36) and (37) may be expressed:
equally pure, nominally. In cases where the feed
o
composition were richer (weaker) then, to avoid
pinch effects, this would be entered nearer to the
+ -
oh i(h) Ih=L = ~i!. (45)
accumulator (reboiler) so approaching a single-
stage situation, capable of easier analysis (e.g.
Armstrong and Wood (1961». ~ereQ3 = ~P+2
o
'J
,p
Q = [0.5E
and ~
0 -0.5(30+.1)
(46) 0J
2.B Small signal boundary conditions. allowing unknown q(L) tobe eliminated between (42)
and (45) giving
These are derived from the large signal boundary
equations (20), (21) and (26) (and a similar equa- ~(o) = Q(o,p)~ (47)
tion for the stripper feed tray) and the assumed G(o,p) relating and compositions to flow rates
(symmetrical) operating conditions. For the acc-
umulator for instance we get, on implicit different-
iation:
a 0 t oh h' =0 r oh h' =0 The matrices of Eq. (4Y) are all diagonal so that
0J
~o
the first-term of the R.H.S. being zero in steady ll(O'P)
state so that normalising we get simply Q(o,p) = (49)
g22(o,P)
T oy(o) = _ EY(O) + Oy\ (32) and knowing Ql ... Q4' we quickly obtain expressions
a o't oh 11=0 for the,transfer fu~ctions gll(o,P) and g22(o,P),
these beIng, respectIvely:
whilst for the stripping section we obtain
h ' (0) _ d (p+2) (cosh/pL-l) /p+(sinh/pL) IIp+0.5} (50)
Tb -0-,-- = - EX' (0)+ OX'
oh
Ih=o (33) {(1+T)p+2+E}cosh/pL+{(p+2) (E+Tp)+p)(sinh/pL)/lp
(0+1) (cosh/p L-l)+(a+l) (sinh/pL)llp+0.5(30+l)
From feed tray Eq. (26) and noting, from the steady (51)
state solution, that: (p(l+T) + EJcosh/p L + Ip(l + E + Tp)sinh/p L
Y' (L) - yeLl = -G(cx+l)/2 (34) the limiting values for p = 0 are:
2
we deduce gll(o,o) = E(L + L + 0.5)/(2EL+o+ 1) (52 )
- - - v + Cl£) - ~I
~2'(L) _ G (- ( Cl+1)
+x' (L)-y(L) (35) and g22(o,o) = -{(o+l)L + 0.5(30 +l)}/E (53)
dT V 2 oh h=L
r
4. PREDICTION OF COLUMN BEHAVIOUR
Similarly, the stripper feed equation is found to
be: A completely analytic T.F.I!. model relating the
~(L) behaviour of y(6,,), x' (0,,) to V(T), q,) has thus
-G {- a v+ - - 1. }- -OX'I
( a+l) +y(L)-x' (L) (36) been derived in terms of normalised complex fre-
dT Vr 2 oh h=L
quency p and normalised column length L, both read-
ily converted to real frequency p' and real length
3. SOLVING FOR THE T. F.11. L' by the formulae
The small signal behaviour of the column given the
flow disturbances v(t), l et) is now completely spe- p' = P L/(H£ oh) (54) and L' Loh (54)
cified by p.d.e's (lB) and (19) and boundary
J. B, Ech,'ards and ~1. H, :\, Tahrizi
The real time constant T' of the end vessels is ob - 4.2 Effect of terminal capacitance
tainable from its normalised value thus:
Since T drops out of the high- frequency analysis
(55) above it follows that only the final portion of the
step - response is influenced by changes in T. This
In terms of y,x , v and ~ , the model may be expressed:
is indeed confirmed by the simulation result of
ry(o,p) - x' (o,pll -Jg ll (o,p) 0JrV(Pl+£(pn Fig.5 fo r E = 0.1, L = 10 with T = 1 , 20 and 50 .
~(o,P)+~ ' (O'P)J =C Vr lo g22(o,P) l;:;(p) -~ (pJ 5. APPROXIMATE ~ruLTIVARIABLE FIRST - ORDER LAC MODEL
(56) One of the motiva t ions for deriving C(o,p) has been
where composition gradient C readily obta~ned from to validate simpler models. Because-the system is
a and L via Eq. (30) . Some aspects of dynamic be - class - O and tends to an integrating process the
haviour are now obvious whilst others require an multivariable first - order lag approximant, CA(O,P)
examination of the formu l ae for gll and g22. Of of Owens (1975) would seem to be applicable-where
the former, we note that, over the entire frequency -1
range, composition - tilt, y(h,T) - X'(h,T) is driven ~A (o,p) = ~l + ~ P (63 )
noninteractively by average circulating flow {v (T) - 1
+ ~ (T) }/2 whilst composition - total y(h,T) + x' (h, T) where ~l = Lim ~ Lim {p- lC-l(o,p)}
' (o,p) and ~
is driven purely by take off rate VeT) - ~ (T). We p-+o Ip I-+<» -
note also that the tilt gain gll(O ,0) is positive l
Here we should strictly replace Ip l-+<» by L- «
whilst g22 0 ,o)is negative and that for low- relative
IpO.s l « 1.0 but the approximant should still app -
volatility mixtures (E « 1.0) requiring long col-
ly for controller design provided excessive gains
umns (L »l.O - see steady state solution), then
are avoided. Now ~l and ~ are readily derived
Igll(o,o) 1«lg22(o,o) I· These findings accord well
analytically. ~l merely involves solution of the
with simulation experience (Rosenbrock, 1966).
small - signal spatial differential equations obtai -
Rademaker ' s deductions (1975) and Shinskey ' s empi -
ned by setting a / aT = 0 in (18) and (19) yielding
rical data (1963) . They, reported that separation
s~tic gain formulae (5 2 ) and (5~) directly. ~ can
is influenced predominantly by Ve T) but our conclu -
be estimated even more simply f rom (18) and (19) by
sions are in close accordance for low- relative
ignoring after transformation all but the p- depend -
volatiles where large reflux ratios are needed , so
ent coefficients of the dependent variables giving
allowing I(v + n l » Iv - £ 1.
I-E, 01 ~, r -1 ~
4.1 Inverse Nyquis t Loci - 1£ ,
Typical loci of gli(O,j w) and g2~(o,j w ) computed
- p -'l -
lo , a +1J
A
--Q
- lo -1 1
- ( a +1) ~
(64)
A TFM model rel a tin g output compos ition chan ges y Accumulator
and x ' to chan ges v and Z in vapour and liquid
flow - rate has been found analytically. It s para-
V - L
meters are derived-functions of plant parameters r r
and opera tin g conditions . Wh il st its purpose is Top
Product
con tr o ller design, it should al l ow dial ogue between
con tr ol engineer and plant designer at an ea rl y Y(6h , t)
s t age in a project. Our assumptions are those fre - h'=o
quent l y made in dynamics studies , (usually implici-
tl y) . These a r e : a piecewise - lin ear equilibrium
curve , spatially independent flow rat es , cons t ant
tray l oadin g and holdup , zero vapou r capacitance ,
tray e quil i brium and symme tr y . The model shou l d
Y(h' , t)
Rectifier
X(h' , t)
I
l'
h'
n
s hows t ha t first - orde r lag behaviour is not always
a safe assumption . Terminal ca pa ci t ance is proved
to a ffec t only the low- frequency end of th e system ' s
f r equency r esponse . :Iany more predictions could be ~ L'
Fig.2. Accurate and 1st-order lag loci for Fig.4 Unit step responses
gJl-l (o,ju)
10j r
I
I
w = 0.91
Accurate
8u
8
\ (o,p)
- 50
I
I
- .... ... 1st Order la g
!!'w = 0.2
11
L - 10, T '" 1
-100
8j 1 0.72 Im
I -1 50
I 8
22
(O,P)
I 0 •• ) 0 , 039
\
6j
0 .5 4 8
22
0.5
2j
r-
:0/ I
-,
I
0.2 0.' Re
S\l
1
0 I
L _ _ -, 4
2.·0
Im
-0.01
o 400 800 1200 1600
Re
Y(tray 10)
w = 0.01 O.Olj
I 0.525
1
0.032 I 0.02j 0 . 520
I
+ 0 . 515
I KEY
0.03)
1 measured
I y
- predicted
step applied
X'ltra /
I
L .04j
0.080
I ,
0.085 0.05j
0.010 --------
50 loo , 200
t (min)