Professional Documents
Culture Documents
CONTROL OF A CONTINUOUS-FLOW
AGITATED-TANK REACTOR
T HO M A S W . W E B E R , State University o j .Vew Y o r k , Buffalo, iV. Y
P E T E R H A R R I 0 T T , Cornell Uniuersity, Ithaca, N . Y,
The stability criteria for a tank-flow reactor are reviewed for the case where there is no mixing delay and
the dynamics of the cooling system are not important. Stability criteria are then developed for cases where
the dynamics of a cooling coil are significant. Zero-order kinetics are assumed, so the stability criteria
are conservative. A zero-order, exothermic reaction was simulated in a 2-foot tank. Hot water was used
as a feed and live steam was sparged into the tank to simulate the heat of reaction. The reactor was made
inherently stable or unstable by controlling the change in steam rate with temperature. The reactor tem-
perature was controlled by the cooling water rate. For a stable reactor, the system was stable for con-
troller gains below a certain maximum; when the reactor was inherently unstable, the system was conditionally
stable. Good control was achieved with about the same controller settings for both cases because the
absolute value of the largest time constant was much larger than the second largest.
HEAT
EXCHANGER FLOW RECORDER I
-
UNSTA0LE REACTOR
PHEUMATIC
VALVES <: i TWO-PEN
TEMPERATURE
RECORDER
v
\ 7r
i FLOW
RECORDING DRAIN
I
n
SYSTEM
metal. T h e transfer function for this model was modified to Control of Inherently Stable Reactor
include the heat of condensation of the steam. This involved With a bare bulb, the calculated maximum gain was 182,
recalculating time constant TI4 as previously indicated. T h e far beyond that obtainable from a standard three-mode con-
calculated reactor transfer function for this case was : troller. At the highest gain setting of the controller. which
0 - ___~ -0.0131' F./lb./hr. corresponded to about 65, the system was only very slightly
-
W, (637s -t 1)(7.55s +
1)(0.981s f 1)
underdamped and the offset resulting from a 10' change in
feed temperature barely detectable.
T h e gain factor is the true value rather than the one predicted With the bulb in a well. the calculated maximum gain was
by the model. T h e measured frequency response was com- about 40 and the critical frequency. 0.049 radian per second.
pared with that predicted from this transfer function. Agree- T h e critical frequency was observed to be about 0 044, which
ment was excellent out to phase lags of about 140' and the agrees well with the calculated value. In one transient test
measured amplitude response was very closely predicted by the with a controller gain of 17, the response was underdamped
model. with a damping ratio of about 0.3. Therefore. the predicted
Prior to the beginning the control tests, calibrations were maximum gain of 40 appears reasonable. Some reset action
made for all components in the system. T h e time constants improved" the transient response and removed offset. T h e
of the cooling water flow process were assumed negligible. value predicted by the Ziegler-Nichols method was satisfactory
Valve time constants are usually small-1 second or less-
especially if the valve motor is small as it was in this case. Control of Inherently Unstable Reactor
T h e lags of flow processes are generally small. Considering
As discussed in the theory, there is generally a range of
that most of the pressure drop occurred across the control
controller gains over which the system is stable in this case.
valve, the lag was probably about a 0.1 second.
Therefore, these control studies involved finding not only the
Excluding the reaci.or and controller, transfer functions
maximum controller gain but the minimum as well.
found for the components in the control loop were:
T h e reactor was made inherently unsiable by increasing the
Valve 85.4 lb. water/hr./cm. Hg gain of the proportional controller in the steam loop. The
degree of instability was only moderate-that is, the rate of
1.22 cm. Hgj' F . change of heat generation by condensation of steam per degree
Pneumatic bare bulb
2.2s +1 Fahrenheit was about 60YGgreater than the rate of heat re-
1.22 cm. Hg/' F. moval per degree. M'ith an agitator speed of 140 r . p . m . , the
Pneumatic bulb in a dry well
(26s + 1)(6s f 1) calculated transfer function for the reactor was :
GREEK
SYMBOLS
8 = steady-state reactor temperature, F.
e = deviation of reactor temperature from steady-state
value, O F.
8., = steady-state temperature of cooling water a t coil inlet,
O F.
CONSTANTS
I
64 RPM 129 RPM
The dynamics of a homogeneous tubular flow reactor with nonlinear kinetics, in response to changes in f e e d
concentration and flow rate, a r e solved, under the assumptions of plug flow, no backmixing, constant proper-
ties, and irreversible nth-order decomposition of a single reactant. The exact solution for this distributed-
parameter, nonlinear, parametrically forced system i s compared with the solutions to the linearized equations
to assess the accuracy of representation of such systems b y their linearized dynamics. It is shown that linear
solutions a r e accurate when disturbances a r e less than 25% of design values. Extensions of the results to
more complex reactions a r e discussed.
THE
general practice for establishing the theoretical and ex- true dynamics. I n the present work, a class of nonlinear
perimental dynamics of tubular chemical reactors has been reactors is studied. This class has the property that analytical
thar of linearization. I n most cases? the choice for theoretical expressions for the responses may be obtained, even though
dynamics is bettveen linearization and numerical solution, their dynamic behavior is nonlinear, distributed, and para-
since analytical solutions are precluded by the nonlinear, non- metrically forced. T h e major objective is to compare analyti-
stationary behavior of tht: partial line expressions for reaction cal and linearized solutions, on both qualitative and quantita-
rates. or in variation of physical properties. Flow rate changes tive aspects.
cause parametric forcing and: therefore, nonstationary be-
havior. Linearization i ! j often preferred over numerical Statement of Problem
solution because of its simplicity, but may require significant
Consider a homogeneous, tubular flow reactor in which sub-
sacrifice of accuracy.
stance A is decomposing irreversibly and isothermally according
Even in experimental tests of reactor responses, linearity is
to a n nrh-order rate expression :
frcquently assumed. .4 typical approach is to introduce a step
change in a n input and to treat the output as a linear response. nA + products
'l'his response is analyzed for time constants and dead time.
RA = kCA"
Often the presence of nonlinearities is detected by changing the
amplitude or direction of the step input, and observing any Assuming plug flow, no backmixing, and constant physical
change in the time constants. Even if nonlinearities are in- properties, the equation of continuity takes the form:
dicated. they may be accounted for by simply using average
valties of the time constants over the expected range of inputs.
Becaiise linearization is so widely used, it is essential to have
as much information as possible about the accuracy of the Since the velocity is to be a time-dependent quantity, it is
linearized dynamics: and the nature of their relation to the divided into steady and fluctuaririg components