Indirect

10 Composition
Measurements

10.1 INTRODUCTION

i n recent years on-line analyzers have vastly improved in sensitivity, selectivity,

speed of response, stability, and reliability. Consequently it is increasingly feasible
to measure the composition of column product streams directly and use these
measurements for column control.' But analyzers are expensive and are somewhat
more trouble-prone than the simple devices used for measuring temperature,
pressure, and flow. Consequently, particularly for less critical applications, con-
siderable ingenuity has been expended on the use of these simpler instruments
in various combinations to deduce compositions. For the most part, the techniques
to be discussed have been demonstrated only for binary and almost-binary
systems. For close control of multicomponent systems, analyzers are virtually
mandatory.
This chapter discusses a number of indirect composition-estimationtechniques.
They vary from the measurement of a single temperature on a tray somewhere
in the column to the measurement of a number of variables that are fed into
an on-line computer that performs rigorous tray-to-tray calculations.

10.2 SINGLE-TRAY TEMPERATURE

Probably the most commonly used composition-estimation technique is the
measurement of a temperature on a single tray in the column. This tray is
normally located in the recntjrlng section if distillate composition is more important,
or in the stripping section if bottom composition is more important. On some
columns two temperatures are controlled, one in each section.
The selection of the location of the control tray has been the subject of
numerous papers over the last 20 to 30 years. Rademaker e t d.'present a well-
229
230 Indirect Cumposition Measwements

organized summary of the various ideas and criteria generated over the years.
The normal procedure is to plot the steady-state temperature profile and then
select a tray that is somewhere in the region where the temperature is changing
fairly rapidly from tray to tray.
Theoretically, the temperature at the end of the column should be controlled
in a binary constant-pressure system to maintain constant product composition.
However, the temperature changes at the ends of the column are quite small
in moderate- to high-purity columns. Therefore, small changes in pressure or
the presence of other lighter or heavier components can affect temperature
much more than composition of the key component.
Another consideration is nonlinearity. The relationship between tray tem-
perature and the manipulated variable can be quite nonlinear if a tray near the
end of the column is used. For example, suppose we have a system in which
reflux is on flow control and a temperature near the bottom of the stripping
section is controlled by heat input to the reboiler. An increase in heat input
will drive more light components up the column, but the control tray temperature
will increase only very slightly because it is essentially pure high boiler already.
However, a decrease in heat input will drop light component down in the
column and the control tray temperature can change very drastically. This
nonlinear response presents difficultcontroller tuning problems when conventional
linear controllers are used. Thus a control tray should be selected that is as
close to the end of the column as possible but not so close that it gives a highly
nonlinear response. Nor should it be so close that it is too sensitive to changes
in pressure and to light or heavy nonkey components in the feed.
For good speed of response, temperature probes should be installed in the
active part of a tray, not in downcomers.

10.3 DIFFERENTIAL TEMPERATURE

For a binary distillation, if the pressure is fixed, a single temperature mea-
surement will provide a reliable p d e to composition. But pressure is often
not very constant. Today it is sometimes deliberately allowed to float to minim&
energy consumption (see Chapter 8). Even if one fixes pressure at one end of
a column, it will vary at the other end as a function of boilup rate. An early
approach to compensating for pressure variations was to use two temperature
measurements, one usually near one end of the column and the other at an
intermediate tray. One was then subtracted from the other. For binary or almost-
binary distillations, this works fairly well as long as boilup does not change
much. If, however, boilup does change significantly, and if the two temperature
measurements are separated by a substantial number of trays, one may encounter
a nonmonotonic relationship between boilup and AT. As pointed out by Boyd3
an increase in boilup tends to decrease A T due to increased purity but to
increase A T due to increased pressure drop. The composite effect is that A T
has a minimum at a particular boilup. This can be very confusing to the operator
10.4 Dzfmentrtial Vapm Pressure 231

and may cause instability in the AT control loop, and may either flood the
column or shut it down.
This nonmonotonic relationship between AT and product composition occurs
even when boilup changes do not affect Al’. This is easily visualized if one
considers two extreme situations. If the profile is so high in the column that
the section between the two temperature points is filled with high boiler, the
AT signal will be small. The AT signal will also be small if the profile is so
low in the column that it is filled with low boilers. These two conditions will
give similar 4T signals but vastly different product compositions.
Efforts inevitably were made to stretch the AT technique to fit multicomponent
distillations. Some of the d16culties encountered and some guidelines for successll
application are presented by web be^.^
In another vintage paper, Vermilion5presents a somewhat different application
of dfferential temperature control. Here the intent apparently was not to com-
pensate for pressure variations, but to control the temperature profile in the
middle trays of the column, and thereby control terminal composition. For a
56-tray deisobutanizer, temperature was measured on trays 22 and 42 (measuring
from top down). Use of the AT permitted successful control of isobutane in
the bottom product.
Successful application to a CJC, splitter column was reported by Bonilla.6

10.4 DIFFERENTIAL VAPOR PRESSURE

Another early effort to compensate for pressure variations in binary distillations
resulted in the development of the differential vapor pressure cell.’ The version
made by Foxboro is known as the “DVP Cell.” As shown in Figure 10.1, a
bulb filled with liquid whose composition is the same as that desired on a
particular tray is installed on that tray. It is connected to one side of a differential-
pressure transmitter. The other side of the AI’ transmitter is connected directly
to the same tray. When the liquid on the tray has the same composition as
the liquid in the bulb, pressure in the bulb will be the same as pressure on the
tray. For that condition the AP transmitter is normally set to read midscale.
Deviations in the tray composition from that in the bulb are then reflected by
AP transmitter output signals above or below midscale.
Where applicable, this instrument is capable of great sensitivity. There are,
however, some practical problems:
1. Liquid of the desired composition for 6 h g the bulb may not be chemically
stable over a long period of time, thus leading to measurement errors.
2. Bulb filling is a job for the factory; reusing an instrument originally
bought for another job may not be easy.
3. Response to pressure changes is much faster than the response to com-
position changes, whch causes erroneous readings and sometimes control prob-
lems. Some users, therefore, have installed snubbers in the pressure impulse
line.
232 Indirect Composition Measur-B

Although this instrument is primarily intended for binary systems, Tivy7 has
shown that with a bit of patience and cleverness, it can be used on some
multicomponent systems.
To determine the required span of the DVP Cell in inches of water cor-
responding to a desired composition span, it is necessary to write and solve a
few equations. If the reference bulb has been properly filled with a liquid of
known composition, then at a reference temperature T R pressure in the bulb
is:
PR = XRpLyLR + (1 - XR) PHYHR (10.1)
where
PR = total pressure, atm abs, in bulb
xR = mol fraction low boiler in bulb
PL = vapor pressure, atm, of low boiler at T R

FIGURE 10.1
DVP cell schematic
10.4 Dtfferential Vapor Pressure 233

PH = vapor pressure, a m , of high boiler at TR

ym = low boiler activity coefficient in bulb
ym = high boiler activity coefficient in bulb
At the same temperature, TR, the process pressure is:
PI-= ~ + (1 - X L ) PHYH
L ~ L Y L (10.2)
where
PT = total pressure in process, atm
xL = mol fraction low boiler in process
PL = vapor pressure, atm, of low boiler at TR
PH = vapor pressure, a m , of high boiler at TR
yL = low boiler activity coefficient
yH = high boiler activity coefficient
Now the DVP Cell actually measures PT - PR and is normally calibrated
in inches of water for such spans as 20-0-20 or 50-0-50. From equations
(10.1) and (10.2):
PT - PR = ~ +
L ~ L Y L(1 - X L ) PHYH (10.3)
- [xRpLYLR + ( l - xR) pHYHR]

which reduces to:

PT = ~ +
L ~ L Y L (1 - X L ) PHYH (10.3a)
which is the same as equation (10.2).
We can now solve for xL:

(10.3b)

Since PR “washes out?’ of equation (10.3), one might well ask, ‘Why do
we have it?”The answer is that there is only one instrument to calibrate rather
than two, and the need for very accurately calibrated temperature and pressure
measurements is eliminated.
The activity coefficients may be found from:

(10.4)

(10.5)
234 Indirect Gmpositiun Measuremcna

where An, BE are Van Laar coefficients. For other ways of finding activity
coefficients, see the various texts on distillation and vapor-liquid equilibrium.
The vapor pressures of the pure components may be found by the Antoine
equation:
pL -- ,+AL+BL/(T+CL)I (10.6)
PH = eFH +EN/(T + CH)] (10.7)
where
, ,CL = Antoine coefficients for the low boiler
A L BL

A H , BH, CH = Antoine coefficients for the high boiler

T = temperature, "C
For the two-parameter Antoine equation:
C H = CL = 273.16
If, as we have postulated, the bulb contains the same chemical species as
the process, PT - PR will always be zero when the process has the same
composition as the bulb. Ifthe chemical species are the same but the compositions
are different, PT - PR will vary somewhat with pressure (and., therefore, tem-
perature). That is to say, for xL - xR constant, PT - PR will change slightly
as PT changes. To facilitate calculations, an HP-41C program has been written
that is valid if the process and bulb contain the same chemical species.
For systems that are not strictly binary but where the nonkey components
are present in either relatively constant or very small concentrations, the DVP
Cell may ofien be used by calibrating its output against laboratory analyses.
When the desired composition is such that the components would react with
one another over a period of lime, one may substitute another fill material (if
one can be found) that has the same PR as desired at TR.In this event the
equations are slightly changed.

10.5 PRESSURE-COMPENSATED TEMPERATURE

To get around or away fi-om the shortcomings of the previous methods,
composition can be computed fi-om temperature and pressure measurements.
One technique is to measure pressure deviations from flowsheet pressure, PFs,
calculate temperature changes at constant composition due to these pressure
deviations, and subtract these temperature changes fi-om the actual temperature.
In effect, the actual temperature is compensated back to what it would be at
flowsheet pressure. A particular hardware concept (see Figure 10.2) we have
used assumes that the P-T curve at constant composition is a straight line for
small deviations. E, then, pressure deviations from PFsare large, there will be
some error.
10.5 Presure-Compmated Temperature 23 5
c,
c
E
E3
ul
m
E
2
H
e
9)
9)
c,
1c
0
-+
E
0
3
5
p8
P,
r
2;
z1
sk
236 Indirect CmposittiOn Measuremenn

The compensator has the equation:

epc = Om, - Kpc 0, + Ob (10.8)
where
6, = compensator output signal
8, = pressure transmitter output signal
,8 = temperature transmitter output signal
Kpc= compensator gain
Ob = compensator bias
T h s h c t i o n is readily performed by commercial proportioning relays or some
proportional-only controllers.

Theory for Binary Distillation

With the above equation in mind, we can see that if there is a pressure
change AP from PFs that causes a temperature change AT, we want no change
in Opc ifcomposition has not changed. In other words:
AOpc = AO,,,, - KpcAO, = 0 (10.9)
or
Kpc= (10.10)
'8, constant composition

AO- = K,, AT (10.11)

AO, = K,AP (10.12)
where
K,, = temperature transmitter gain

(10.13)

T,, - T- = temperature transmitter input span

(AO,),, = transmitter output span (whether
pneumatic or electronic is
immaterial)
Kmp= pressure transmitter gain
-
- )ma
P- - P-
P,, - P- = transmitter input span
(AO,),, = transmitter output span
 -
K
m
ar/aT
P AP
cornposiuon
constam
“i 1
-- -
K~~ ar/aT
1

FIGURE 10.3
Composition vs. t and p
238 Indirect Cornposh M e a s u r m a

fiaction, and activity coefficient for each component and is defined:

n
PT = yixi pi
i=1

where n is the number of components and Pi is the vapor pressure at T for

the ith component.
pi = e[Az+BJ(T+273.16)]
so
n
pT = yixi c k + B i / ( T + 2 7 3 . 1 6 ) ]
i= 1

Differentiating, we get:

n
-1
-
(T + 273.16)2i = l E YiXiBiPi

and
aT - 1
aPT aP,/aT
Some Practical Considerations
1 . The output of the compensator,, ,e is a measure of the pressure-com-
pensated temperature. The process operator must know what this relationship
is in order to adjust the set point for the controller properly. It is common
practice to adjust the compensator bias so that it reads midscale when T =
TFsand P = PFs,or more generally when:
,8 - ICpc,8 = 0 [see equation ( l O . l ) ]
Therefore:

2. Some operators have found it confusing to have both temperature and

pressure-compensated temperature. We therefore have taken Figure 10.3 and
plotted C versus T (both read at constant PFs)on a new plot. The operator's
control station can then be calibrated directly in terms of composition.
3 . In some cases the transmitted pressure signal is so noisy that a snubber
may be required. For electronics an RC filter may be used.
If a computer is available, it is possible to take a considerably more accurate
approach.
10.6 Multimnpiment Comparitimc fi.om Temperature and Pressure Measurements 239

For a binary system, the basic equation is:

PT = ~ L ~ L Y L + (1 -~ L ) ~ H Y H (10.17)
where
yL = activity coefficient of low boiler
yH = activity coefficient, high boiler
Equation (10.17) may be solved for xL:

(10.18)

The individual vapor pressures may be found by the Antoine equations; see
equations ( 10.6) and (10.7).The activity coefficients may be found by equations
(10.4) and (10.5).
In looking at equations (10.18), (10.4), and (lO.S), we can see that xL is
really an implicit function. By knowing T we can readily calculate PL and PH,
but fiom there on we must find xL by trial and error. This is easily done on
any computer.

10.6 MULTICOMPONENT COMPOSITIONS COMPUTED FROM

TEMPERATURE AND PRESSURE MEASUREMENTS
If an on-line digital computer is available, it is possible to use a more
sophisticated computing scheme than otherwise would be practical. To treat
this subject in a very general way is not feasible. To show what is possible,
however, let us consider as an example a three-component system where the
ratio of the concentration of two components is essentially constant:
_- R
x1 - (10.19)
x2

3
P = -&yipi (10.20)
i= 1

P,= exp A, +
(
T + 273.16
(10.21)

It is desired to find x1 fiom measurements of PT and T, taken at the same

tray in the column. We may rewrite equation (10.20):

x1 =
p - Y3P3 (10.22)
YlP1
Y2
+zP2 - (1 + +) y3P3

Knowing Ai, B j , one may find Pi from equation (10.21). Then knowing
Piand y i , one may find x1 from equation (10.22). The activity coefficients
240 Indirect Composition Measurements

(yls) are sometimes fairly constant (if‘ the range of composition changes is
small). More generally, the y’s would be functions of composition, so an
iterative, trial-and-error solution would be required.

10.7 DOUBLE-DIFFERENTIAL TEMPERATURE

Strictly speaking one temperature and one pressure at a point in a column
can be used to predict composition accurately only if the system is binary. The
thought has occurred to a number of people that perhaps additional temperature
measurements could be used to calculate composition in multicomponent systems.
Two papers that illustrate two Werent approaches have appeared in the literature.
Luyben’ made a computer study of a 25-tray deisobutanizer with temperature
measurements on trays 5, 10, and 15 (numbering up from the bottom). He
defined:
A2T = (TlO - 7-15) - (7-5 - TlO) (10.23)

Plots of A2T versus X, (mol fraction E,)showed that it was affected very little
by disturbances, but they did a show a maximum at about xB = 0.02. Since
most deisobutanizers operate in a range of xB = 0.03 - 0.10, this might well
be useful for control.
Boyd’ took a somewhat different approach. He located two temperature
measurements in the rectifying section of the column and two more in the
stripping section. He then defined:
A2T = ATR - ATs (10.24)
where
ATR = temperature dfference in rectification section
AT, = temperature difference in stripping section
Plant tests on two columns showed great sensitivity. In each case A2T was used
to control top product; each system was calibrated by gas chromatographs.
The computed and measured data checked well.
Boyd‘s technique partially fixed the composition-versus-tray profile in the
vicinity of the feed tray and on both sides of it. Luyben’s method partially
fixed the profile in either end of the column. Both approaches effectively cancel
out the influence of boilup rate and pressure on temperature measurements.
Therefore, they both may be regarded as improvements on simple AT mea-
surements, and are primarily usell for binary or almost-binary systems.
Bonilla6 reported that Boyd’s double-differential-temperature technique did
not work well for a CJC, splitter. The signal was found to be dependent on
feed composition, and also was nonmonotonic.
10.9 Composition Estimm 241

10.8 AVERAGE TEMPERATURE

The use of multiple temperature measurements was proposed by Grote'
and independently (but much later) by Luyben." This technique is useful for
distillation columns in which a large temperature change occurs over a few
trays (a sharp temperature profile). This is found in columns where the components
have widely differing boiling points (peanut butter and hydrogen would be an
extreme example).
Control problems are frequently encountered in these columns because the
process gain is very high (a small change in heat input makes a drastic change
in the temperature on any single control tray). In addition, the system saturates
easily: the temperature drops to the boiling point of the low boiling component
as the profile drops below the control point but shows no change thereafter
as the light component works down the column.
The multiple-temperature-control scheme uses a number (four or five) of
temperature sensors located around the normal temperature break. An average
temperature signal is computed from these multiple temperatures (this can be
inexpensively implemented in conventional pneumatic or electronic analog
hardware).
As the profile moves up or down, this average temperature changes gradually.
The average temperature signal is fed into a conventional feedback controller.
This technique has been successfully applied to many industrial columns.

10.9 COMPOSITION ESTIMATORS

Brosilow and co-worker~"-'~studied the use of several temperature and
flow-rate measurements to estimate product composition. They used the term
"inferential control" for this type of composition estimation and control using
secondary process measurements.
The Brosilow estimator employs a linear combination of selected tray tem-
peratures and steam and reflux flow rates to estimate product compositions.
For example, distillate composition is estimated using an equation of the form:
xQ = ajlTl + aj,T2 + aj3T3+ a j z s + aj5L, (10.25)
where
Fs = steam flow rate
Lo = external reflux flow rate
The a coefficients are constants that are determined experimentally or by
calculations.
The estimator uses steady-state relationships, handles multicomponent systems,
and is linear. Both simulation studies and experience on an industrial column
have been reported. Shah14 had good success with the Brosilow estimator as
long as the column was operated within a linear region. However, when nonlinear
242 Indirect compmitimr Measwemm

phenomena occur, the linear estimator performs poorly; that is, poor estimates
of product compositions are made.
Shah14mended Bmilow‘s work, proposing a nonlinear composition estimator
for binary systems. A fundamental nonlinear mathematical madel of the column
is used, involving tray-to-tray material and energy balances, vapor-liquid equi-
librium and physical property data, known process parameters (tray efficiencies
and heat losses), and measured variables (feed, distillate, and steam flow rates
and one or more tray temperatures).
Suppose, for example, that rdux and distrllate flow rates and a tray temperature
in the rectlfjrlngsection are known,and one wants to calculate distillate composition
x,. The rigorous estimator makes a guess of xD and calculates tray-by-tray down
the column until the tray is reached on which the temperature is measured. If
the calculated temperature on this tray does not agree with the measured
temperature, a new guess of X, is made and the procedure is repeated.
Shah verified the effectiveness of the nonlinear estimator by both simulation
studies and experimental tests on a pilot-scale column. The work of Weber and
M o ~ l e r ’should
~ also be noted. They proposed several control schemes using
the sum of two or more temperatures, the difference between two temperatures,
and the distillate-to-feed ratio to adjust manipulated variables.

1. Griffin, D. E., J. R. Parsons, and
D. E. Smith, “The Use of Process
Analyzers for Composition Control
of Fractionators,” ISA Trans., 13(1)
(1979).
2. Rademaker, O., J. E. Rijnsdorp, and
A. Maarkveld, DynamM and contrd
OfContinuOw Di&llathn Units” El-
sevier, New York, 1975.
3. Boyd, D. M., CEP 71(6):55-60 (June
1975).
4. Webber, W. O., Pet. Rt$38(5): 187-
191 (May 1959).
5. V e d o n , W. L., Oil and GasJ., 98-
100 (Aug. 21, 1961).
6. Bondla, J. A., Hyahfux Pnx. 24-0 (Nov.
1976).
7. Tivy, V. “Automatic Control of Frac-
tionating ~ h m n s , P
”et. (Nov.
1948).
REFERENCES
8. Luyben, W. L., IEC Fund., 8: 739-
744 (Nov. 1969).
9. Grote, H. W., U.S. Patent 2,725,351
(Nov. 29, 1955).
10. Luyben, W. L., AIChE J., 18: 238
(1972).
11. Webber, R., and C . B. Brosilow,
AIChE 1. 18: 614 (19721.
12. Joseph, B.: et al. Hyd. h.127
, (Mar.
1976).
13. Joseph, B., et al., AIChE J., 24:
485(1978).
14. Shah, M. K., and W. L. Luyben, I m .
Chem. En.. Symp. Series, no. 56,
2.6/1(1979).
15. Weber, R., and H. A. Mosler, U.S.
Patents 3,733,627 (1973) and
3,855,074 (1974).