Vous êtes sur la page 1sur 12

Article

pubs.acs.org/IECR

Control of Rotating Wave Trains in a Loop Reactor


Pietro Altimari and Erasmo Mancusi*,,

Dipartimento di Chimica, Universita Sapienza di Roma, Piazzale Aldo Moro 5, 00185 Roma, Italia

Facolta dIngegneria, Universita del Sannio, Piazza Roma, 82100, Benevento, Italia

Departamento de Engenharia Qumica e de Alimentos, Universidade Federal de Santa Catarina, 88040-970, Florianopolis, SC, Brazil

ABSTRACT: A control strategy is proposed to stabilize thermal wave trains in networks of catalytic reactors with periodically
rotated inlet and outlet ports. The implemented approach entails updating the time between two successive port rotations based
on the current values of reaction- and thermal-front velocities. These latter values are estimated online by an algorithm derived
from geometric analysis of the spatiotemporal temperature pattern. Guidelines for the placement of temperature sensors and
controller design that enforce robust stability at regime and during startup are illustrated. The illustrated strategy is validated on a
simulated problem of controlling a loop reactor for the combustion of volatile organic compounds.

INTRODUCTION
Spatiotemporally varying patterns, such as traveling fronts and
temperature fronts with the periodic rotation of the inlet and
outlet ports. This rotation enables, for example, a reaction front
periodic waves, can autonomously form in catalytic reactors as a traveling in the ow direction to be trapped within the bed and
result of the interplay between reaction and transport.16 The can thus ensure autothermal operation at low adiabatic
emergence of these solutions can lead to thermal runway and temperature rise. This has motivated signicant interest in
reaction extinction and has therefore been considered a critical the application of such network technology to the combustion
concern for reactor operation.7,8 Numerous theoretical and of gaseous streams with low contents of volatile organic
experimental studies over the past decades have thus focused compounds.18,2128 The eect of radial heat losses in the
on the development of design and control methodologies system for a lean mixture of volatile compounds was analyzed
preventing the formation of spatiotemporally varying pat- by Barresi et al.29 This latter application covers the largest
terns.9,10 Typically, stationary operation is targeted, and reactor portion of the studies on such networks presented during the
parameter values far from the stability boundaries of such past two decades.
solutions are selected. Some studies have also provided indications about the
In contrast with such an approach, theoretical and network potential to enhance the yield and/or selectivity in
experimental evidence has been reported demonstrating the reversible reactions or more complex reaction schemes. This
possibility of purposefully exploiting spatiotemporally varying has been conrmed for the synthesis of methanol,3033 the
patterns to optimize the performance of catalytic processes.11,12 selective reduction of nitrogen oxides,3436 and the production
The fundamental idea is to drive the motion of traveling fronts of synthesis gas through the coupling of partial oxidation and
and waves to construct a prescribed spatiotemporal pattern.13 steam reforming.37
This methodology has been demonstrated to be successful in Despite the satisfactory results obtained in these applications,
enhancing the yield and selectivity in reactiondiusion the fundamental mechanisms governing the emergence of high-
systems.11,12,14 Scarce attention has been paid, however, to its conversion regimes have only recently been identied, shedding
application in convectionreactiondiusion systems. This light on the actual network potential. Sheintuch and
class of systems congures the most appropriate modeling Nekhamkina38 elucidated the fundamental role played by the
framework to describe the dynamics of catalytic reactors and interaction between traveling fronts and periodic forcing. They
thus represents an obligatory benchmark for testing the demonstrated the existence of single thermal waves traveling in
reliability of the introduced control ideas. the ow direction for ns/Vth < < ns/Vfr, where Vfr and Vth are
Particular interest has been attracted in this context to the the velocities of a reaction front and of a purely thermal front
application of a network of tubular catalytic reactors with (that is, a temperature front moving in the absence of reaction),
periodically rotated inlet and outlet ports.15,16 A schematic respectively. Experimental demonstration of this latter solution
description of the network, also referred to as a loop reactor or regime was also reported in ref 39.
a simulated moving-bed reactor, is shown in Figure 1. The Subsequent studies were focused on the formation of trains
number of reactors rotated in the ow direction is dened as ns, of traveling thermal waves with switching times of 0 < < ns/
and the time between two successive modications of the Vth. These regimes allow the conversiontemperature pattern
feeding sequence is referred to as the switching time . minimizing the amount of catalyst needed to attain a prescribed
The network is usually operated in T*-periodic regimes,
where T* is the time needed to recover the initial feeding Received: April 4, 2013
conguration, although multiperiodic, quasiperiodic, and Revised: July 10, 2013
chaotic solutions can arise.1720 The emergence of high- Accepted: August 1, 2013
conversion solutions is governed by the interaction of traveling Published: August 1, 2013

2013 American Chemical Society 12134 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 1. Schematic representation of the network. Inlet and outlet ports are periodically rotated in a stepwise manner by ns reactors.

conversion in equilibrium-limited processes to be reproduced40 A proportional feedback control law was proposed by
and can signicantly reduce the maximum temperature Smagina and Sheintuch,47 allowing the switching time to be
experienced by the catalyst in combustion processes.41 Mancusi computed in response to the dierence between the catalyst
et al.42 demonstrated that increasing ns can enlarge the values temperature at a prescribed axial position and a set-point value.
and the range sustaining the emergence of thermal wave-train Stability analysis of the controlled network was also performed
solutions. This is of practical importance, as the achievement of on a simplied model reported earlier by Sheintuch and
low switching times would require frequent modication of the Nekhamkina,38 providing indications for the selection of
feeding sequence and, thus, cause damage to the external valve controller parameters.
system. An analogous control structure was employed by Sheinman
Altimari et al.43 later demonstrated the possibility of and Sheintuch48 for the robust stabilization of the single-wave
controlling the structure of thermal wave trains. Innitely solution of a network carrying out methanol synthesis. They
many domains of thermal wave trains were demonstrated to implemented a proportional controller and a proportional
exist for any value of ns and any number of reactors composing derivative controller to update the switching time in response
the network, and analytical approximations were derived for the to the dierence between the axial position reached by the
stability boundaries of the predicted solutions based on reaction front at the end of each cycle and a set-point value.
geometric analysis of the spatiotemporal temperature pattern. The set point was updated during each cycle based on the
The analysis was extended to reversible exothermic reactions.40 online estimation of the reaction-front velocity. For this
The coexistence of wave-train solutions with low-conversion purpose, an algorithm, whose detailed description was not
multiperiodic and extinguished regimes was analyzed in ref 44. reported, was employed by the authors for the reconstruction
The main obstacles to network operation around T*-periodic of the temperature pattern. Other advanced approaches, such as
wave-train regimes are the narrow switching time stability range the use of observers or soft sensors for forced unsteady-state
of such solutions and their coexistence with low-conversion reactors, have been proposed to control a single traveling
multiperiodic and extinguished regimes. Under these con- temperature wave.4951
ditions, external disturbances and model uncertainties can likely Even though they ensure the stability of the single-wave
cause network shutdown or transition to low-conversion solution, each of the aforementioned strategies is inadequate for
regimes. For this reason, the switching time cannot be xed operating the network in wave-train regimes.52 A study
a priori but must be tightly controlled. addressing the problem of stabilizing wave-train solutions was
Several approaches have been developed for network control, recently presented.52 In that study, the proportional control
all of them concerned with single-wave solutions. The proposed structure proposed by Sheinman and Sheintuch48 was
control strategies employ the switching time as a manipulated employed, and a controller design procedure enforcing the
variable and can be grouped into two dierent classes: (1) stability of wave-train solutions was described. The imple-
strategies triggering switching when temperature values at one mented control was adaptive as the set point was updated
or two prescribed axial positions cross certain thresholds and during each cycle based on the estimated value of the reaction-
(2) strategies updating the implemented switching time based front velocity. Analytical relationships previously determined
on an explicit feedback control law. Among the former group, for the stability limits of wave-train solutions43 were used for
the strategy proposed by Barresi et al.45 entails switching when this purpose.
the catalyst temperature at the beginning of the fed reactor falls The main limit of the approach presented in ref 52 is related
below a rst set point and the temperature at the inlet of the to the application of an approximate form for the expression of
subsequent reactor is above a second set point (signicantly the stability limits of wave-train solutions. These limits depend
larger than the rst). Zahn et al.46 investigated the performance on both the reaction- and thermal-front velocities, Vfr and Vth,
of three dierent controllers, again falling in group 1, respectively. However, only Vfr was estimated online with the
concluding that the most eective approach, requiring the approach described in ref 52, and a simplied form for the
minimum number of temperature sensors, is to install a expression of the stability limits was employed to update the set
thermocouple at the end of each reactor and to switch when the point based on the assumption Vfr Vth. However, this
temperature at the end of the fed reactor becomes lower than a assumption is valid at zero adiabatic temperature rise, whereas
prescribed set point. The feasibility and robustness of this Vfr becomes signicantly lower than Vth as the adiabatic
control strategy was also conrmed experimentally. temperature rise is increased. Because the minimum adiabatic
12135 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

temperature rise sustaining autothermal operation increases z tu T T0 C


with the number of waves,41 diculties can be found as = , t* = 0 , = ,x=1 ,
L L T C0
solutions with increasing number of waves are considered.
A further weak point of the control method described in ref E u ( H )C0
= ,v= ,B= ,D
52 is related to the implemented startup strategy, which RT0 u0 (c p)f T0
imposes that network operation be started with the catalyst
AL (c p)eff (c p)f Lu0
uniformly heated at temperatures as high as possible. This a= exp( ), Le = , Peh = ,P
approach is based on the observation that higher initial u0 (c p)f ke
temperatures reduce the risk of transition of the controlled Lu0 L
network to multiperiodic and extinguished solutions. However, em = ,L= 0
there is no guarantee that increasing the initial temperature is Df N (2)
sucient to denitively rule out this risk. Danckwerts boundary conditions are applied at inlet and outlet
In this article, a control strategy overcoming the previous sections of each reactor and are modied after each switching
diculties is presented. In accordance with the proposed to take into account the permutation of the network feeding
approach, a switching time value ensuring stability of the sequence. The boundary conditions can be written in
prescribed solution is determined during each cycle based on dimensionless form as
estimated values of reaction- and thermal-front velocities.
Again, analytical relationships previously determined for the 1 xi
stability limits of wave-train solutions are used for this purpose, = [1 fi (t *)]x in fi (t *) xi 1(i 1, t *)
Pem 0
but unlike in ref 52, no proportional control is employed.
Because both the reaction- and thermal-front velocities are + xi(i 1, t *)
estimated during each cycle, the diculties related to the
1 i
application of an approximate form for the expression of the = [1 fi (t *)]in fi (t *) i 1(i 1, t *)
stability limits are overcome. Also, it is demonstrated that the Peh 0
proposed control strategy signicantly enhances robustness + i(i 1, t *)
during startup.
The rest of this article is structured as follows: The xi i
mathematical model of the network is described next. Then, = = 0, i = 1, ..., N
1
1
the stability characteristics and spatiotemporal temperature
pattern of wave-train solutions of the uncontrolled network are (3)
reviewed. Subsequently, the proposed control strategy and its where f i(t) = h{t* mod[(i 1)ns, N]}, mod(*, ) denotes
implementation are analyzed. Concluding remarks end the the standard modulo function,53 and h(t*) is the following
article. piecewise-constant periodic function

MATHEMATICAL MODEL
We consider networks of catalytic reactors organized in a loop


h(t ) =
t*
0 if 0 mod , N < 1

(see Figure 1 for a representative scheme). In the following t*


1 if mod , N > 1
discussion, the number of reactors composing the network is (4)
denoted as N. The inlet and outlet sections are periodically
rotated in the ow direction so as to jump ns reactors each time. It is worth noting that the implemented operating strategy
The number of shifted reactors, ns, is kept constant during (that is, ns) enters the mathematical model through the forcing
operation. The switching velocity is dened as Vsw = ns/. N function h(t*) in the boundary conditions 3.
1 switching strategies can be identied as ns varies between 1 The numerical analysis illustrated in the article was derived,
and N 1.43 unless otherwise specied, using the parameter values reported
A rst-order irreversible exothermic reaction is selected as in Table 1. We investigated parameter values capable of
representative of a large class of chemical processes. For the sustaining the formation of a reaction front moving at a
sake of simplicity, constant physical properties and negligible constant velocity in the ow direction, as these conditions are
interphase gradients are assumed. Mass and energy balances for of practical interest. Indeed, under stationary operation, the
the ith reactor unit can be then written as
Table 1. Dimensionless Parameter Valuesa
xi x 2
1 xi parameter value
+v i = 2
+ Da(1 x) exp i
t * Pem i + Damkohler number, Da 0.0129/N
dimensionless activation energy, 14.13
2
i 1 i dimensionless adiabatic temperature rise, B 4
Le +v i = + BDa(1 x) dimensionless feed temperature,b in 4.82
t * Peh 2
dimensionless velocity, v 1
exp i , i = 1, ..., N Lewis number, Le 800
i + Peclet number for heat conduction, Peh 312.5/N
(1) Peclet number for mass diusion, Pem 500/N

with the following denitions for dimensionless variables and a


Reference dimensional parameters: u0 = 0.5 m/s, L0 = 0.5 m, C0 = 1
parameters (see also the Nomenclature section) 103 kmol/m3, T0 = 293 C. bCorresponding to Tin = 100 C.

12136 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145


Industrial & Engineering Chemistry Research Article

emergence of a reaction front traveling in the ow direction section. This determines the warming of the cold region ab
causes extinction, and thus, periodic rotation of inlet and outlet generated during the previous cycle. The network energy
ports is required to attain high conversions. balance is closed, as a new cold region ab is generated at the
The total length L0 of the network is maintained constant as same time by displacement of the upstream reaction front.
N is varied. Therefore, the Peclet and Damkohler number The warming of cold region ab generated by displacement of
values reported in Table 1 are scaled by N. the reaction front is possible only if thermal fronts forming in a

SPATIOTEMPORAL STRUCTURE OF WAVE-TRAIN


SOLUTIONS
and b reach the network outlet section during the same cycle.
For the pattern displayed in Figure 2, these two thermal fronts
must reach the network outlet section within a time interval
Thermal waves are formed in the network by the interactions of ranging between one and two cycles. This imposes the
traveling temperature fronts with the rotating inlet and outlet condition that the distance covered in two cycles by the front
sections. To elucidate this mechanism, we analyze in Figure 2 forming in a is larger than the axial position reached by the
network outlet section within the same time interval, that is,
2Vth 2. Analogously, it can be derived that the distance
covered by the front forming in b in a single cycle must be
lower than the axial position reached by the network outlet
section within the same time interval, that is, Vfr + Vth 2.
The derived inequalities can be manipulated to determine the
switching-time stability limits of the considered spatiotemporal
pattern leading to 1/Vth 2/(Vfr + Vth).
Innitely many T*-periodic wave-train regimes can be
generated as varies and the two fronts forming in a and b
reach the network outlet section after the same number of
cycles. In particular, a train of p thermal waves is found when
the fronts forming in a and b spend between p 1 and p cycles
traveling through the network. Again, the switching-time values
sustaining the emergence of such a solution can be determined
by the geometric analysis illustrated earlier. This gives
mod(nsp , N ) mod(nsp , N )

Vthp Vfr + Vth(p 1) (5)

The term mod(ns p, N) appearing in inequalities 5 denes


the axial position reached by the network outlet section within
p cycles. It is important to remark that values dened by
inequalities 5 can also sustain the emergence of multiperiodic
and extinguished solutions.44 There is therefore no guarantee
that a T*-periodic wave-train solution will be reached even
when parameter values fullling inequalities 5 are selected.

CONTROL OF ROTATING THERMAL WAVE TRAINS


In the following subsection, a strategy for network control is
illustrated. The objective is to bring the network to operate
around a prescribed T*-periodic wave-train regime and to
Figure 2. T*-periodic two-wave solution (ns = 3 and N = 4 at = ensure stability in the presence of disturbance and model
830): (a) temperature prole predicted at one-half cycle and (b) uncertainties.
spatiotemporal temperature pattern, where blue and red identify cold Control Strategy. In accordance with the analysis
and hot regions, respectively. Each line separating such regions presented in ref 43 and reviewed in the preceding section,
describes the evolution of the axial position of a temperature front.
ensuring the stability of T*-periodic wave-train regimes requires
that inequalities 5 be satised. Once ns, N, and p are xed, this
the spatiotemporal temperature pattern of a T*-periodic regime objective can be achieved by varying the velocities of the
corresponding to a train of two thermal waves. Vertical bold temperature fronts or the switching time . The velocities of the
segments are used to denote the positions of the inlet and temperature fronts can be modied by varying the feeding
outlet sections during each cycle. conditions (e.g., feed temperature and composition). This
It is apparent from Figure 2 that the fresh reactants are approach is less ecient than varying the switching time,
always fed after switching to a hot reactor section. This ensures however. This latter variable can be instantaneously changed to
the formation of a new reaction front moving in the ow a prescribed value without any additional cost for network
direction with velocity Vfr and prevents transition to operation, whereas modications in the feed characteristics
extinguished or multiperiodic regimes. Owing to the displace- increase operation cost and take a nite time interval to be
ment of the reaction front, a cold region, ab, with a width of introduced. These considerations suggest that switching time
Vfr is generated during each cycle (Figure 2). Two thermal should be employed as the manipulated variable. Obviously,
fronts then form after switching in a and b and move at improved control performance could be attained by simulta-
constant velocity Vth = 1/Le until reaching the network outlet neously manipulating the switching time and the feed
12137 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

characteristics. However, the design of such a multivariable


control system goes beyond the scope of the present study. In
the following development, therefore, the switching time is
employed as the only manipulated variable.
When reaction- and thermal-front velocities are known,
inequalities 5 can be used to determine a switching-time value
ensuring stability. Owing to model uncertainties and dis-
turbances, variations in Vth and Vfr can occur, however, causing
the selected switching time to fall outside the stability range of
the prescribed solution. The approach we followed to prevent
this possibility was to update the switching time during each
cycle based on the current values of Vth and Vfr. Estimates for
these velocities were derived online and substituted into
inequalities 5 to determine the switching time. To enhance
robustness, the middle point of the stability range dened by
inequalities 5 was selected as the switching time i to be
implemented during the ith cycle. This gives the following
control law
mod(nsp , N ) 1 1
i = +
2 Vth, ip Vfr, i + Vth, i(p 1) (6)
where Vth,i and Vfr,i are the thermal- and reaction-front velocity Figure 3. Geometry-based online estimation of reaction- and thermal-
values, respectively, estimated during the ith cycle. The front velocities. The spatiotemporal temperature pattern displayed in
problem to be solved is to determine Vth,i and Vfr,i. The Figure 2 is considered. The axial positions of the thermocouples used
algorithm that we implemented for this purpose was derived are indicated by vertical dashed gray lines, while vertical bold gray
segments are used to denote the time intervals t*th and t*fr . The
from the geometric analysis of the network spatiotemporal instant at which a temperature front reaches the thermocouple placed
pattern illustrated in the previous section. in the ith reactor is determined by thhe intersection between the front
Consider the wave-train solution displayed in Figure 2. Each spatiotemporal path and the vertical dashed gray line at = i + .
temperature front separates a region characterized by temper- This allows t*th and t*fr to be computed during each cycle.
atures close to the feed temperature from a region where
signicantly larger temperature values are attained. Owing to
the attainment of large Peclet values (typically found in whereas the thermocouples placed in the fourth, third, and
industrial practice), temperature fronts are very steep, and second reactors allow t*fr to be determined during the second,
transitions between adjacent hot and cold regions take place third, and fourth cycles, respectively.
within a very narrow spatial interval. Under these conditions, This approach to estimate Vfr is identical to the one
the instant at which any temperature front crosses a prescribed employed in ref 52. We next demonstrate that Vth can be
axial position is marked by a sudden transition of in the estimated without introducing any additional thermocouple.
temperature at that position, T(), between a value close to the From geometric analysis of the spatiotemporal pattern, it is
feed temperature Tin and a value considerably larger than Tin. A known that a thermal front always forms at a switching in a
decreasing or rising temperature front is found when the (Figure 3). If the kth cycle is considered, the point a is the inlet
transition is accompanied by an increase or decrease in section of the reactor fed during the (k 1)th cycle. To
temperature, respectively. In this framework, the reaction-front estimate Vth, it is sucient to record the time interval tth *
velocity Vfr can be estimated by placing a thermocouple at a needed for the thermal front forming in a to cover the distance
distance from the inlet section of the fed reactor and separating a from the subsequent thermocouple. This can
recording the time interval tfr* needed for the measured be computed as the time needed for the temperature measured
temperature to reach Tin. Because the reaction front forms at by the thermocouple at a distance from a to become larger
switching near the inlet of the fed reactor, it must be the case than Tin. It can thus be found that
that

Vth =
Vfr = th* (8)
tfr* (7)
In Figure 3, t*th is determined during the rst cycle by
In network operation, the feed section is periodically modied, monitoring the temperature measured by the thermocouple
and an estimate for Vfr can therefore be derived during each placed in the second reactor. The thermocouples placed in the
cycle provided that a thermocouple is installed at a distance rst, fourth, and third reactors are then exploited to estimate
from the inlet of each reactor. Vth during the second, third, and fourth cycles, respectively. It
A schematic representation illustrating the approach should be noted that, because Vth > Vfr, a thermal front takes
implemented to estimate Vfr is reported in Figure 3. The less time than the reaction front to cover the distance ,
spatiotemporal temperature pattern displayed in Figure 2 is implying that tfr* > tth
*. This ensures that an estimate for Vth
considered as a representative example, and the axial position of is available during each cycle before Vfr is computed.
each thermocouple is indicated. The time interval t*fr is Inuence of Controller Parameters. Three parameters
determined during the rst cycle by monitoring the temper- need to be xed to implement the formulated control
ature measured by the thermocouple placed in the rst reactor, algorithm: and two temperature limit values. Temperature
12138 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 4. Impact of front deformation on the estimation of reaction- and thermal-front velocities: (a) snapshots of the reaction front crossing the
axial position * = 0.5, (b) snapshots of a decreasing thermal front crossing the axial position * = 0.5. Owing to the deformation induced by thermal
dispersion, the temperature T(*) is dierent from Tin at the time instant when the front crosses *.

Figure 5. Inuence of the limit temperature values Ts,1 and Ts,2 on the estimated reaction- and thermal-front velocities (a) Vfr and (b) Vth,
respectively. The displayed data were computed by numerical simulation of the controlled network. The reported Vfr and Vth values correspond to
the achievement of the two-wave T*-periodic regime illustrated in Figure 2.

limit values are required to determine the instant at which a the front deformation caused by thermal dispersion, the use of a
temperature front crosses the thermocouple axial position. The temperature T s,2 larger than Ts,1 is recommended, as
passage of a rising temperature front, for example, must be demonstrated in Figure 4b. The evolution of the thermal
identied by recording the time instant at which the measured front displayed in this latter gure indicates that Ts,2 values
temperature becomes lower than a limit value Ts,1 > Tin and not exceedingly close to Tin would lead to an underestimation the
when it reaches Tin. Because of the front deformation caused by time needed to cross a prescribed axial position.
thermal dispersion and the nitely large reaction rate, the tail of Increasing the enthalpy Peclet number, on the other hand,
the front extends upstream in a wide spatial interval with makes each temperature front reducing the required dierences
temperatures higher than Tin.41 The time needed for the between Ts,1, Ts,2, and Tin steeper. Because of the large Peclet
measured temperature to reach Tin is consequently longer than values found in industrial practice, the stability of the controlled
the time needed for the front to cross the thermocouple axial network is ensured within a wide range of variation of Ts,1 and
position. This is clearly shown in Figure 4a, where snapshots of Ts,2. To conrm this statement, we analyze in Figure 5 the
the reaction front crossing a prescribed axial position * are eects of Ts,1 and Ts,2 on the Vfr and Vth values estimated with
displayed. It is apparent that the front has entirely crossed and the proposed algorithm. The results displayed in Figure 5 were
is far from * when T(*) = Tin, whereas it becomes derived by numerical simulation of the controlled network
progressively closer to * as T(*) increases above Tin. Ts,1 operated around the two-wave T*-periodic regime arising with
values slightly larger than Tin must therefore be selected to N = 4 and ns = 3. The curves displayed in Figure 5 can
determine the instant at which the front crosses the therefore be considered stable branches of this latter solution.
thermocouple axial position. The estimated Vfr value increases slightly with Ts,1 (Figure 5a).
A second temperature limit value Ts,2 must be dened to This can be explained by noting that, because of the front
identify the passage of a decreasing thermal front. Because of deformation caused by thermal dispersion (Figure 4a),
12139 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 6. Startup of the controlled network; (a,c) temperature proles computed at one-half of each cycle under regime conditions and (b,d)
spatiotemporal temperature patterns. (a,b) Implementation of a 1 value fullling inequality 10 allows the T*-periodic ve waves solution arising with
N = 4 and ns = 3 to be reached. (c,d) Application of a 1 value violating inequality 10 leads to the emergence of a four-wave 5T*-periodic solution
coexisting with the prescribed T*-periodic one.

increasing Ts,1 determines a reduction in the time interval tfr* each cycle. When implementing the illustrated control
needed for the temperature at to reach Ts,1. Figure 5a also algorithm, the distance covered by the reaction front during
shows that the estimated Vfr value is barely aected by Ts,2. The the ith cycle is Vfr,ii, where i is dened by eq 6. The upper
reason for this latter behavior is that the reaction front forms bound for is therefore given by
near the network inlet section and travels in the ow direction
regardless of the implemented Ts,2 value. mod(nsp , N ) 1 1
Increasing Ts,2 causes an increase in the estimated Vth value max = +
2 p 1 + (p 1) (9)
(Figure 5b) This eect can, again, be explained by analyzing the
front deformation induced by thermal dispersion. Vth is
estimated by evaluating the time interval tth * needed for the where = maxi(Vth,i/Vfr,i). max is therefore determined by the
thermal front forming in a to reach the thermocouple axial type of regime (the p value in eq 9) to be stabilized and by .
position. This time interval increases with Ts,2 because of front Because Vfr,i and Vth,i depend on the reactor parameters, the
inection (Figure 4b). Larger Ts,1 values, on the other hand, value of to be used in inequality 9 should be selected based
cause a reduction in the estimated Vfr, thereby decreasing the on the explored region of parameter space. At parameter values
implemented switching time, in accordance with eq 6. This, in of practical relevance, the inequality Vth,i 1.5Vfr,i is typically
turn, causes a reduction in the maximum temperature attained fullled. Therefore, = 1.5 is generally sucient to determine
by the front forming in a,41 leading to a reduction in tth *. max.
It is important to remark that, in accordance with the results Startup. A fundamental obstacle to network operation
displayed in Figure 5, the implemented control law can ensure around T*-periodic wave-train solutions is multistability.
stabilization of wave-train solutions within a wide region of Because of the coexistence of the target T*-periodic solution
temperature limit values Ts,1 and Ts,2. with extinguished and low-conversion multiperiodic regimes,
The choice of must also be based on an analysis of the temperature hot spots compromising catalyst activity and/or
evolution of the temperature fronts. must be lower than the reductions in the average conversion are likely to occur owing
distance covered by the reaction front during a single cycle. The to disturbances and/or model uncertainty. The most complex
fulllment of this requirement is essential to ensure that the step is in this framework startup. Because large variations in
reaction front can cross the thermocouple axial position during temperature and concentration occur during this phase of
12140 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 7. Evolution of the outlet temperature and switching time during startup, corresponding to the numerical simulations illustrated in (a,b)
Figure 6a,b and (c,d) Figure 6c,d. (a,c) Temperature and (b,d) switching-time values computed at the end of each cycle are reported. Consequently,
(a,b) a xed point is attained when the prescribed ve-wave T*-periodic solution is reached, whereas (c,d) a periodic evolution (with period = 5) is
found as a result of the emergence of the four-wave 5T*-periodic solution.

operation, the risk of transitions to undesired solutions is mod(nsp , N ) p 1 1 1


signicantly greater than for operation at regime. 1 1 +
The approach typically followed for network startup is to Vfr,1 2 p p 1 + 1/
preheat the catalytic bed with a hot inert gas ow and (10)
subsequently start to feed fresh reactants and to rotate inlet and
outlet sections. It was observed in ref 52 that increasing the Derivation of inequality 10 is reported in the Appendix. As
initial bed temperature reduced the risk of transitions to found for inequality 9, the value = 1.5 can be used, with
extinguished and multiperiodic solutions. However, neither any reactor parameter values of industrial relevance, to determine
indication about the limit initial temperature to be imposed nor an upper bound for 1. An estimate for such a bound, valid at
any guarantee to reach the target solution was provided. We any set of reactor parameter values, can be derived in the limit
next demonstrate that the control algorithm illustrated in the 1/ 0. This limit corresponds to a thermal-front velocity that
previous section can successfully cause the network to operate is innitely larger than the reaction-front velocity and cannot
around any T*-periodic wave-train solution. The only therefore be realized. It can nonetheless be imposed to
prerequisite to be fullled for this purpose is that the initial determine the theoretical minimum of the right-hand side of
bed temperature be large enough to ensure the formation of a inequality 10.
reaction front close to the network inlet. The pattern displayed in Figure 6a,b was generated by
To demonstrate the eectiveness of the formulated control implementing an underestimated 1 value computed based on
strategy for startup, the problem of bringing the network to the eq 10. In particular, the employed 1 value was selected to be
ve-wave solution arising with N = 4 and ns = 3 is considered. large enough to generate an upstream cold region ab and
Adiabatic temperature rise values sustaining this latter solution signicantly lower than the minimum switching time ensuring
generate Vfr values signicantly larger than the corresponding stability, that is, 1/Vth, computed with the actual Vth value.
Vth values, making the application of the control strategy Once the region ab had been created by displacement of the
proposed in ref 52 inadequate. The evolution of the reaction front, switching was performed, leading to the
temperature prole resulting from implementation of control formation of a thermal front in a. An estimate for Vth,2 was
law 6 during startup is displayed in Figure 6a,b. A reaction front derived during the second cycle by monitoring the temperature
is formed at t* = 0 at the network inlet section and moves in at a distance from the inlet of the rst reactor. A new
the ow direction during the rst cycle. By monitoring the reaction front also formed at the beginning of the second cycle,
temperature at a distance < max from the inlet, an and an estimate for Vfr,2 was determined by means of the
estimate for Vfr,1 can be obtained. In contrast, no estimate for thermocouple placed in the fourth reactor. The switching time
Vth can be determined during the rst cycle as no thermal front 2 was thus computed. Iterating this procedure enabled the
is present at this stage. This rules out the possibility of prescribed solution to be reached. The evolutions of the
computing 1 based on eq 6. It can be demonstrated that a switching time and of the network outlet temperature during
transition to the prescribed solution is ensured provided that startup are reported in Figure 7a,b. It is apparent that about 50
the following condition is fullled cycles were needed to reach the prescribed regime and
12141 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 8. Response of the controlled network to disturbances in the (ac) feed reactant concentration C0 and (df) gas velocity u. The illustrated
results describe the startup to the three-wave T*-periodic solution arising with N = 4 and ns = 3.

moderate deviations of the outlet temperature and switching This can, in turn, cause transitions to multiperiodic or
time from the corresponding regime values are found. extinguished solutions.
It is important to remark that, in accordance with the The eects of step variations in C0 and u are examined in
illustrated analysis, no lower limit was found for 1. The Figure 8. The problem of bringing the network to the T*-
implementation of a 1 value even much lower than the periodic three-wave solution arising with N = 4 and ns = 3 is
minimum value actually required to ensure stability is sucient considered. To analyze the ability of the controller to reject
to reach the prescribed solution. In contrast, extinction or disturbances both at regime and during startup, numerical
transition to a coexisting multiperiodic solution was found simulations were performed starting from a uniform initial
when 1 values not fullling inequality 10 were selected. This temperature prole with disturbances introduced before the
latter circumstance is illustrated in Figure 6c,d. In this latter target T*-periodic regime was reached. Moreover, the
case, the controller could not recover the error made during the minimum C0 and u values considered in the numerical
rst cycle, and a transition to a four-wave multiperiodic solution simulation were selected close to the stability boundaries of
was found even though i values enforcing stability were the target solution. It must be emphasized that the achievement
of such parameter values would make the target solution
implemented for all i > 1. The evolutions of the switching time
unstable if an open-loop control strategy (i.e., an a priori xed
and the outlet temperature during startup are illustrated in
switching time) were implemented.
Figure 7c,d. It is worth noting that a multiperiodic solution with
The evolutions of the network outlet temperature and
a period of 5T* was attained, with the switching time oscillating switching time reported in Figure 8 demonstrate the ability of
at regime within the stability range of the T*-periodic solution the controlled network to reject disturbances. The controller
displayed in Figure 6a,b. This was possible because of the modies the switching time to prevent that the stability
coexistence of the two displayed solutions. boundaries of the target solution are crossed. Large variations in
The implementation of an underestimated 1 value is the outlet temperature were initially found corresponding to a
recommended. It is nonetheless important that 1 be selected transition from the initial at temperature prole to a three-
large enough to generate a region ab at a temperature around wave temperature pattern. Also, a sudden increase in the
Tin. This is essential to form a thermal front in a and thus switching time was obtained following the rst cycle. Because
estimate Vth during the second cycle. an underestimated 1 value was implemented, a sudden increase
Controller Performance. The ability of the formulated in the switching time was observed as an estimate for Vth was
control strategy to reject disturbances in the feed concentration derived during the second cycle. Following the illustrated initial
C0 and in the gas velocity u was analyzed. Variations in these transient, moderate variations in the outlet temperature and
two parameters typically occur in industrial practice, causing switching time were ensured by the controller in response to
signicant changes in reaction- and thermal-front velocities. the introduced disturbances.
12142 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

Figure 9. Response of the controlled network to disturbances in the (ac) feed reactant concentration C0 and (df) gas velocity u. The illustrated
results describe the startup to the ve-wave T*-periodic solution arising with N = 4 and ns = 3 and analyzed in Figure 6a,b.

The analysis of controller robustness was completed by analysis of the network spatiotemporal temperature pattern.
extending this numerical experiment to a solution characterized The implementation of the illustrated control strategy for
by a larger number of temperature waves. This test was network startup was analyzed. A procedure was presented in
essential to demonstrate the eectiveness of the proposed this framework enabling the controlled network to be brought
control approach, as control diculties increase signicantly to any T*-periodic wave-train solution. Numerical simulations
with increasing number of waves. Indeed, increasing the were used to demonstrate the eectiveness of the proposed
number of waves causes an increase in the number of coexisting controller in rejecting disturbances both at regime and during
multiperiodic solutions and reduces the network stability range. startup.
These two latter eects contribute to an increase in the risk of We conclude by noting the need for further investigation of
transitions to undesired solutions. the impact of the implemented control strategy on the
The problem of bringing the network to the ve-wave spatiotemporal structure and stability of the loop reactor. In
solution displayed in Figure 6a,b is considered. The evolutions particular, the eects of the values of controller parameters on
of the outlet temperature and switching time in response to the stability limits of multiperiodic solution regimes coexisting
successive step variations in C0 and u are displayed in Figure 9. with the target T*-periodic regimes should be performed
Again, disturbances were introduced before the regime was through a detailed bifurcation analysis. This study can allow for
reached, and the C0 and u values imposed in the numerical indications to be derived on how to design the proposed
simulation were selected very close to the stability boundaries control structure to reduce the risk of transitions to undesired
of the target solution. Again, the controller eectively rejected regimes.
the disturbances. Following the initial transient, corresponding
to the transition from a at prole to a ve-wave temperature
pattern, oscillations of moderate amplitude were observed for
APPENDIX: DERIVATION OF INEQUALITY 10
Stability limits derived for the uncontrolled network in ref 41
outlet temperature and switching time, and the target solution and reviewed in the section Spatiotemporal Structure of Wave-
was eventually attained.

Train Solutions of the present article apply almost unchanged


to the controlled network. The only dierence to be taken into
CONCLUSIONS account in the analysis of the controlled network is the
A control strategy has been proposed to stabilize thermal wave possibility of variations in the reaction- and thermal-front
trains in a network of catalytic reactors with periodically rotated velocities. Any T*-periodic wave-train solution can be sustained
inlet and outlet ports. The strategy relies on updating the in the controlled network provided that the thermal fronts
switching time during each cycle based on the current values of forming in a and b (Figure 2) reach the network outlet section
the reaction- and thermal-front velocities. To estimate these during the same cycle. For a train of p thermal waves, this
two velocities online, an algorithm was derived from geometric imposes the following two conditions:41
12143 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145
Industrial & Engineering Chemistry Research Article

First, the distance covered in p cycles by the thermal front q = number of reactors covered by a purely thermal front =
forming in a is greater than the distance (modulo the network mod(nsp,N)
length) covered during the same time interval by the network r = reaction rate
outlet section, that is R = gas constant
p+1 t = time
Vth,ii mod(nsp , N ) T = temperature
i=2 (11) t* = dimensionless time
T * = time needed to recover the initial feeding
Second, the distance covered in p 1 cycles by the thermal conguration (T*=N)
front forming in b is less than the distance (modulo the Tl = lowest maximum temperature of a temperature wave
network length) covered during the same time interval by the Tm = highest maximum temperature of a temperature wave
network outlet section, that is u = gas rate
p v = dimensionless gas ow rate
Vfr,11 + Vth,ii mod(nsp , N ) Vfr = dimensionless reaction-front velocity
i=2 (12) Vsw = dimensionless switching velocity
Vth = dimensionless purely thermal-front velocity
Inequalities 11 and 12 coincide with inequalities 5 in the case x = conversion
of constant reaction- and thermal-front velocities. Inequality 11 z = axial coordinate
involves only i values with i > 1. These latter values are
computed based on eq 6, as estimates for Vth,i and Vfr,i can be Greek Letters
derived during startup for all i > 1. This ensures the fulllment = maximum value assumed by the ratio of the thermal-
of inequality 11. Inequality 12, in contrast, involves switching front velocity to the reaction-front velocity = maxi (Vth,i/Vfr,i)
time 1, which cannot be computed based on eq 6 because of H = heat of reaction
the impossibility of estimating Vth,1 during startup. Inequality tfr* = time interval needed for the reaction front to cover
12 can be exploited to obtain an upper bound for 1. By the distance
replacing the implemented control law (eq 6) with i for i > 1, t*th = time interval needed for the thermal front to cover the
inequality 12 can be recast in the form distance
= distance of each thermocouple from the upstream
p
reactor inlet
Vfr,11 11 1 = reactor void fraction
1 + Vfr, i
mod(nsp , N ) 2 p p1+ = dimensionless activation energy
i=2
Vth, i
= dimensionless temperature
p 1 1 1 = density
1 + = dimensionless switch time
2 p p 1 + 1/ (13) = dimensionless axial coordinate
where = maxi(Vth,i/Vfr,i). Subscripts and Superscripts

AUTHOR INFORMATION
Corresponding Author
0 = reference conditions
f = uid
h = enthalpy
*E-mail: mancusi@unisannio.it (E.M.), pietro.altimari@ in = inlet
uniroma1.it (P.A.). m = mass
Notes max = maximum
The authors declare no competing nancial interest.

NOMENCLATURE
REFERENCES
(1) Rhee, H. K.; Lewis, R. P.; Amundson, N. R. Creeping profiles in
A = Arrhenius constant catalytic fixed bed reactors. Continuous models. Ind. Eng. Chem.
B = dimensionless adiabatic temperature rise Fundam. 1974, 13, 317.
(2) Gatica, J. E.; Puszynski, J.; Hlavacek, V. Reaction front
C = concentration propagation in nonadiabatic exothermic reaction flow systems.
cp = specic heat capacity AIChE J. 1987, 33, 819.
Df = mass axial dispersion coecient (3) Sheintuch, M. Excitable waves in a fixed bed reactor:
Da = Damkohler number Observations and analysis. Chem. Eng. Sci. 1990, 45, 2125.
E = activation energy (4) Yakhnin, V. Z.; Rovinsky, A. B.; Menzinger, M. Convective
f(t) = forcing function instability induced by differential transport in a tubular packed bed
gl(T) = function describing the dependence of Tl on Tm reactor. Chem. Eng. Sci. 1995, 50, 2853.
gm(T) = function describing the dependence of Tm on Tl (5) Sheintuch, M.; Nekhamkina, O. Pattern formation in models of
h(t) = piecewise-constant function dened by eq 4 fixed-bed reactors. Catal. Today 2001, 70, 369.
ke = solid-phase axial heat conductivity (6) Yochelis, A.; Sheintuch, M. Principal bifurcations and symmetries
in the emergence of reactiondiffusionadvection patterns on finite
L = length of a single reactor unit domains. Phys. Rev. E 2009, 80, 056201.
L0 = network length (7) Pinjala, V.; Chen, Y. C.; Luss, D. Wrong-way behavior of packed-
N = number of reactors composing the network bed reactors: II. Impact of thermal dispersion. AIChE J. 1988, 34,
ns = number of reactors jumped by inlet and outlet sections 16631672.
p = number of temperature waves (8) Luss, D. Temperature Fronts and Patterns in Catalytic Systems.
Pe = Peclet number Ind. Eng. Chem. Res. 1997, 36, 2931.

12144 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145


Industrial & Engineering Chemistry Research Article

(9) Yakhnin, V. Z.; Menzinger, M. Convective instability and its (32) Velardi, S.; Barresi, A. A.; Manca, D.; Fissore, D. Complex
suppression in packed-bed- and monolith reactors. Chem. Eng. Sci. dynamic behaviour of methanol synthesis in the ring reactor network.
1999, 54, 4547. Chem. Eng. J. 2004, 99, 117.
(10) Gudekar, K. G.; Riggs, J. B. Bifurcation and Stability Analysis of (33) Sheinman, R.; Sheintuch, M. Loop Reactor Design and Control
an Ethylene Oxide Reactor System. Ind. Eng. Chem. Res. 2003, 42, for Reversible Exothermic Reactions. Ind. Eng. Chem. Res. 2009, 48,
3285. 5185.
(11) Wolff, J.; Papathanasiou, A. G.; Kevrekidis, I. G.; Rotermund, H. (34) Fissore, D.; Garran Tejedor, D.; Barresi, A. A. Experimental
H.; Ertl, G. Spatiotemporal addressing of surface activity. Science 2001, investigation of the SCR of NOx in a Simulated Moving Bed Reactor.
294, 134. AIChE J. 2006, 52 (9), 3146.
(12) Wolff, J.; Papathanasiou, A. G.; Rotermund, H. H.; Ertl, G.; Li, (35) Fissore, D.; Penciu, O. M.; Barresi, A. A. SCR of NOx in loop
X.; Kevrekidis, I. G. Gentle dragging of reaction waves. Phys. Rev. Lett. reactors: Asymptotic model and bifurcational analysis. Chem. Eng. J.
2003, 90, 018302. 2006, 122, 175.
(13) Mikhailov, A. S.; Showalter, K. Control of waves, patterns and (36) Fissore, D.; Barresi, A. A.; Botar-Jid, C. C. NOx removal in
turbulence in chemical systems. Phys. Rep. 2006, 425, 79. forced unsteady-state chromatographic reactors. Chem. Eng. Sci. 2006,
(14) Papathanasiou, A. G.; Wolff, J.; Kevrekidis, I. G.; Rotermund, H. 61, 3409.
H.; Ertl, G. Some twists and turns in the path of improving surface (37) Fissore, D.; Barresi, A. A.; Baldi, G. Synthesis Gas Production in
activity. Chem. Phys. Lett. 2002, 358, 407. a Forced Unsteady-State Reactor Network. Ind. Eng. Chem. Res. 2003,
(15) Matros, Y. S., Ed. Catalytic Processes under Unsteady State 42, 2489.
Conditions; Elsevier: Amsterdam, 1989. (38) Sheintuch, M.; Nekhamkina, O. The asymptotes of loop
(16) Haynes, T. N.; Caram, H. S. The simulated moving bed reactors. AIChE J. 2005, 51, 224.
(39) Madai, A. Y.; Sheintuch, M. Demonstration of loop reactor
chemical reactor. Chem. Eng. Sci. 1994, 49, 5465.
operation. AIChE J. 2008, 54, 2413.
(17) Mancusi, E.; Russo, L.; Continillo, G.; Crescitelli, S.
(40) Altimari, P.; Mancusi, E.; Crescitelli, S. Formation of Thermal
Computation of frequency locking regions for a discontinuous
Wave Trains in Loop Reactors: Stability Limits and Spatiotemporal
periodically forced reactor. Comput. Chem. Eng. 2004, 28, 187.
Structure for Reversible Reactions. Ind. Eng. Chem. Res. 2012, 51, 9609.
(18) Russo, L.; Mancusi, E.; Maffettone, P. L.; Crescitelli, S.
(41) Altimari, P.; Mancusi, E.; Brasiello, A.; Maffettone, P. L.;
Nonlinear analysis of a network of three continuous stirred tank Crescitelli, S. Temperature wave trains of the loop reactor: The effect
reactors with periodic feed switching: Symmetry and symmetry- of thermal dispersion. Chem. Eng. Sci. 2012, 76, 108.
breaking. Int. J. Bifurcation Chaos Appl. Sci. Eng. 2004, 14, 1325. (42) Mancusi, E.; Altimari, P.; Russo, L.; Crescitelli, S. Temperature
(19) Altimari, P.; Maffettone, P. L.; Crescitelli, S.; Russo, L.; Mancusi, and conversion patterns in a network of catalytic reactors for methanol
E. Nonlinear dynamics of VOC combustion loop reactor. AIChE J. synthesis for different switch strategies. Chem. Eng. Sci. 2010, 65, 4579.
2006, 52, 2812. (43) Altimari, P.; Mancusi, E.; Russo, L.; Crescitelli, S. Temperature
(20) Russo, L.; Altimari, P.; Mancusi, E.; Maffettone, P. L.; wave-trains of periodically forced networks of catalytic reactors. AIChE
Crescitelli, S. Complex dynamics and spatiotemporal patterns in a J. 2012, 58, 899.
network of three distributed chemical reactors with periodical feed (44) Mancusi, E.; Russo, L.; Altimari, P.; Crescitelli, S. Multiplicities
switching. Chaos, Solitons Fractals 2006, 28, 682. of temperature wave trains in periodically forced networks of catalytic
(21) Brinkmann, M.; Barresi, A. A.; Vanni, M.; Baldi, G. Unsteady reactors for reversible exothermic reactions. Chem. Eng. J. 2011, 171,
state treatment of a very lean waste gases in a network of catalytic 655.
burners. Catal. Today 1999, 47, 263. (45) Barresi, A. A.; Vanni, M.; Brinkmann, M.; Baldi, G. Control of
(22) Fissore, D.; Barresi, A. A. Comparison between the reverse-flow an autothermal network of nonstationary catalytic reactors. AIChE J.
reactor and a network of reactors for the oxidation of lean VOC 1999, 45, 1597.
mixtures. Chem. Eng. Technol. 2002, 25 (4), 421426. (46) Zahn, V. M.; Yi, C. U.; Morgenstern, A. S. Analysis and
(23) Fissore, D.; Barresi, A. A.; Velardi, S.; Vanni, M. On the demonstration of a control concept for a heat integrated simulated
properties of the forced unsteady-state ring reactor network. Chin. J. moving bed reactor. Chem. Eng. Sci. 2011, 66, 4901.
Chem. Eng. 2004, 12 (3), 408. (47) Smagina, Y.; Sheintuch, M. Control of rotating pulses in a loop
(24) Sheintuch, M.; Nekhamkina, O. Comparison of flow-reversal, reactor. J. Process Control 2009, 19, 954.
internal-recirculation and loop reactors. Chem. Eng. Sci. 2004, 59 (19), (48) Sheinman, R.; Sheintuch, M. Loop Reactor Design and Control
4065. for Reversible Exothermic Reactions. Ind. Eng. Chem. Res. 2009, 48,
(25) Barresi, A. A.; Baldi, G.; Fissore, D. Forced Unsteady-State 5185.
Reactors as Efficient Devices for Integrated Processes: Case Histories (49) Fissore, D.; Pisano, R.; Barresi, A. A. Observer design for the
and New Perspectives. Ind. Eng. Chem. Res. 2007, 46 (25), 8693. selective catalytic reduction of NOx in a loop reactor. Chem. Eng. J.
(26) Mancusi, E.; Russo, L.; Altimari, P.; Maffetone, P. L.; Crescitelli, 2007, 128, 181.
S. Effect of the Switch Strategy on the Stability of Reactor Networks. (50) Fissore, D. Robust control in presence of parametric
Ind. Eng. Chem. Res. 2007, 46, 6510. uncertainties: Observer-based feedback controller design. Chem. Eng.
(27) Zahn, V. M.; Mangold, M.; Krasnyk, M.; Seidel-Morgenstern, A. Sci. 2008, 63, 1890.
Theoretical Analysis of Heat Integration in a Periodically Operated (51) Fissore, D.; David, E.; Hammouri, H.; Barresi, A. A. Nonlinear
Cascade of Catalytic Fixed-Bed Reactors. Chem. Eng. Technol. 2009, soft-sensor design for unsteady-state VOC afterburners. AIChE J.
32, 1326. 2006, 52 (1), 282.
(28) Zahn, V. M.; Mangold, M.; Seidel-Morgenstern, A. Autothermal (52) Altimari, P.; Mancusi, E. Control of temperature wave trains in
operation of an adiabatic simulated counter current reactor. Chem. Eng. periodically forced networks of catalytic reactors for methanol
Sci. 2010, 65, 458. synthesis. Chem. Eng. Process.: Process Intensif. 2013, 63, 25.
(29) Barresi, A. A.; Vanni, M.; Brinkmann, M.; Baldi, G. Influence of (53) Koshy, T. Elementary Number Theory with Applications; Elsevier:
Amsterdam, 2007.
the wall heat losses on the performance of a network of nonstationary
catalytic burners. Chem. Eng. Technol. 1999, 22, 237.
(30) Vanden Bussche, K. M.; Froment, G. F. The STAR
configuration for methanol synthesis in reversed flow reactors. Can.
J. Chem. Eng. 1996, 74, 729.
(31) Velardi, S. A.; Barresi, A. A. Methanol synthesis in a forced
unsteady state reactor network. Chem. Eng. Sci. 2002, 57, 2995.

12145 dx.doi.org/10.1021/ie401067x | Ind. Eng. Chem. Res. 2013, 52, 1213412145

Vous aimerez peut-être aussi