30.10.2014 Views

Comparison of configurations of a four-column simulated moving bed

Comparison of configurations of a four-column simulated moving bed

Comparison of configurations of a four-column simulated moving bed

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

1<br />

2<br />

3<br />

<strong>Comparison</strong> <strong>of</strong> <strong>configurations</strong> <strong>of</strong> a <strong>four</strong>-<strong>column</strong><br />

<strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> process by multi-objective<br />

optimization<br />

4<br />

5<br />

6<br />

7<br />

8<br />

9<br />

Yoshiaki Kawajiri* and Lorenz T. Biegler<br />

Dept. <strong>of</strong> Chemical Engineering, Carnegie Mellon University, 5000 Forbes Ave.,<br />

Pittsburgh, PA, 15213<br />

*Corresponding author, Phone:(412) 268-2238, Fax: (412) 268-7139, E-mail:<br />

kawajiri@cmu.edu<br />

June 21, 2007<br />

10<br />

Abstract<br />

11<br />

12<br />

13<br />

14<br />

15<br />

16<br />

17<br />

18<br />

19<br />

20<br />

Configurations <strong>of</strong> a <strong>four</strong>-<strong>column</strong> <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> chromatographic<br />

process are investigated by multi-objective optimization. Various existing <strong>column</strong><br />

<strong>configurations</strong> are compared through a multi-objective optimization problem.<br />

Furthermore, an approach based on an SMB superstructure is applied to<br />

find novel <strong>configurations</strong> which have been found to outperform the standard<br />

SMB configuration. An efficient numerical optimization technique is applied to<br />

the mathematical model <strong>of</strong> the SMB process. It has been confirmed that although<br />

the optimal configuration highly depends on the purity requirement, the<br />

superstructure approach is able to find the most efficient configuration without<br />

exploring various existing <strong>configurations</strong>.<br />

21<br />

1 Introduction<br />

22<br />

23<br />

Chromatographic separation processes have been applied to many products for large-<br />

scale production. The most widely used operation mode is batch elution using a single


24<br />

25<br />

26<br />

27<br />

28<br />

29<br />

30<br />

31<br />

32<br />

33<br />

34<br />

35<br />

36<br />

37<br />

38<br />

39<br />

40<br />

41<br />

42<br />

43<br />

44<br />

45<br />

46<br />

47<br />

48<br />

49<br />

<strong>column</strong>, where the feed mixture is loaded periodically and eluted with desorbent. In<br />

this mode, the purified products are collected at the end <strong>of</strong> the <strong>column</strong>. On the other<br />

hand, in the continuous mode, the feed can be continuously supplied and products<br />

can be continuously withdrawn. The representative system is the <strong>simulated</strong> <strong>moving</strong><br />

<strong>bed</strong> (SMB) process, where multiple <strong>column</strong>s are connected in a cycle, as shown in<br />

Figure 1. SMB mimics the countercurrent operation by intermittently switching the<br />

<strong>four</strong> streams, desorbent, extract, feed, and raffinate, in the direction <strong>of</strong> the liquid<br />

flow. Because <strong>of</strong> this continuous and pseudo-countercurrent operation mode, SMB<br />

has the potential to increase the productivity and reduce the desorbent consumption.<br />

However, numerical treatment <strong>of</strong> SMB processes comes with great difficulty because<br />

<strong>of</strong> the complex operation.<br />

In order to find efficient designs and operating parameters, various mathematical<br />

programming techniques have been applied to SMB. Dünnebier et al. (2000) and<br />

Toumi et al. (2002) have reported applications <strong>of</strong> sequential quadratic programming<br />

(SQP), a Newton-based nonlinear programming technique. In our previous study, we<br />

reported successful results <strong>of</strong> the full-discretization approach using an interior-point<br />

method for SMB processes (Kawajiri and Biegler, 2006d).<br />

In recent years, many modifications to the standard SMB have been attempted<br />

to further improve the performance. Fig.2(b) shows the Three-zone configuration<br />

(Ruthven and Ching, 1989; Ching et al., 1992), where the circulation from the right<br />

end to the left is cut and all liquid is withdrawn as the raffinate stream. Fig.2(c)<br />

shows the Three-zone configuration with purging, where the <strong>column</strong> at the left end<br />

is isolated and the components in the <strong>column</strong> are purged into the extract stream.<br />

As a result, the number <strong>of</strong> design alternatives for multi-<strong>column</strong> processes increases<br />

rapidly. However, only a few studies consider the <strong>column</strong> configuration <strong>of</strong> SMB.<br />

Zhang et al. (2002) used a genetic algorithm in their multi-objective optimization


50<br />

51<br />

52<br />

53<br />

54<br />

55<br />

56<br />

57<br />

58<br />

59<br />

60<br />

61<br />

62<br />

63<br />

64<br />

65<br />

66<br />

formulation. They investigated <strong>configurations</strong> <strong>of</strong> SMB with five <strong>column</strong>s. Ziomek<br />

et al. (2006) explored many different <strong>configurations</strong> <strong>of</strong> five identical <strong>column</strong>s. For<br />

this problem, we previously reported a superstructure based approach to find the<br />

optimal zone configuration and operation (Kawajiri and Biegler, 2006b). Although<br />

many novel <strong>configurations</strong> and operations have been found in previous studies, a<br />

systematic comparison <strong>of</strong> <strong>configurations</strong> has never been considered.<br />

In this work, we investigate extensively <strong>configurations</strong> <strong>of</strong> a <strong>four</strong>-<strong>column</strong> SMB.<br />

Various existing <strong>column</strong> <strong>configurations</strong> are compared through a multi-objective optimization<br />

study to analyze the trade-<strong>of</strong>f <strong>of</strong> throughput and desorbent consumption.<br />

The mathematical model is fully discretized and incorporated within a large-scale nonlinear<br />

programming (NLP) problem, which is solved using an interior-point solver,<br />

IPOPT (Wächter and Biegler, 2005). The superstructure approach, which em<strong>bed</strong>s<br />

many possible <strong>configurations</strong> such as Three-zone and standard SMB, is employed. A<br />

discussion <strong>of</strong> the standard and nonstandard <strong>column</strong> <strong>configurations</strong> is given in Section<br />

2. In Section 3, the mathematical model and superstructure approach are presented.<br />

Finally, standard and nonstandard <strong>configurations</strong> are compared with the superstructure<br />

approach in Section 4.<br />

67<br />

2 SMB <strong>column</strong> <strong>configurations</strong><br />

68<br />

69<br />

70<br />

71<br />

72<br />

73<br />

In the standard SMB configuration, the <strong>column</strong> are connected in a cycle as shown<br />

in Figure 1. It is important to note that in a <strong>four</strong>-<strong>column</strong> process, there is only one<br />

standard configuration, where one <strong>column</strong> comes between every inlet/outlet streams.<br />

On the other hand, there are numerous non-standard <strong>configurations</strong>. One representative<br />

example is Three-zone (Figure 2(a)), where the circulation is cut <strong>of</strong>f and all<br />

liquid is withdrawn in the raffinate stream. This configuration consumes more des-


74<br />

75<br />

76<br />

77<br />

78<br />

79<br />

80<br />

81<br />

82<br />

83<br />

84<br />

85<br />

86<br />

87<br />

88<br />

89<br />

90<br />

91<br />

92<br />

93<br />

94<br />

95<br />

96<br />

97<br />

98<br />

99<br />

orbent than the standard SMB configuration, since there is no recycle stream from<br />

the rightmost <strong>column</strong> to the leftmost <strong>column</strong>. Moreover, in Three-zone with purging<br />

(Figure 2(b)), the leftmost <strong>column</strong> is isolated and all components in the <strong>column</strong> are<br />

purged into the extract stream. This configuration is efficient if there is a component<br />

which has a high affinity to the adsorbent (slow <strong>moving</strong> component).<br />

Furthermore, in our previous study (Kawajiri and Biegler, 2006a), the F-shaped<br />

configuration was extracted from the SMB superstructure as the optimal <strong>column</strong><br />

configuration (Figure 2 (d)). This configuration has been found to have the highest<br />

throughput over other SMB <strong>configurations</strong> in our case studies <strong>of</strong> linear and nonlinear<br />

isotherms, but at the expense <strong>of</strong> higher desorbent consumption. This novel structure<br />

was an interesting finding; the desorbent stream is located downstream <strong>of</strong> the raffinate<br />

stream, and the recycle stream is cut <strong>of</strong>f. An analysis <strong>of</strong> this configuration is given in<br />

Section 4.<br />

It is also important to note that operations which involve multiple <strong>configurations</strong><br />

are possible. For instance, the operation <strong>of</strong> the Varicol system has multiple standard<br />

SMB <strong>configurations</strong> in one step. This feature can be generalized to operations<br />

which have multiple non-standard <strong>configurations</strong>. For this problem, we proposed a<br />

superstructure-based approach where the dynamic optimal configuration is obtained<br />

as a solution <strong>of</strong> an optimal control problem Kawajiri and Biegler (2006c). However,<br />

such dynamic <strong>configurations</strong> can be hard to implement in large-scale processes, since<br />

frequent changes <strong>of</strong> flow rates may lead to pressure instability. Therefore, investigations<br />

<strong>of</strong> operations <strong>of</strong> “static <strong>column</strong> configuration,” where the flow rates are kept<br />

constant in a step, also have significant importance.<br />

In the following section, we formulate a multi-objective optimization problem to<br />

compare the above-mentioned <strong>configurations</strong>. In this study, we only consider <strong>configurations</strong><br />

<strong>of</strong> <strong>four</strong>-<strong>column</strong> SMB. Discussion <strong>of</strong> SMBs with higher numbers <strong>of</strong> <strong>column</strong>s


100<br />

can be found in our previous studies Kawajiri and Biegler (2006b).<br />

101<br />

102<br />

3 Mathematical Modeling and Problem Formula-<br />

tion<br />

103<br />

3.1 Modeling <strong>of</strong> Chromatographic Column<br />

For the modeling <strong>of</strong> each chromatographic <strong>column</strong>, the linear driving force (LDF)<br />

model is employed, as in our previous study (Kawajiri and Biegler, 2006d). Here<br />

both axial dispersion and diffusion into adsorbent particles, which cause band broadening,<br />

are lumped into a mass transfer coefficient. The validity <strong>of</strong> such an assumption<br />

can be found elsewhere (for example, Dünnebier and Klatt (2000); van Deemter et al.<br />

(1956); Golshan-Shirazi and Guiochon (2003)). The mass balance equations in the<br />

liquid and solid phases are given by:<br />

∂C j i (x, t)<br />

ǫ b<br />

∂t<br />

(1 − ǫ b ) ∂qj i (x, t)<br />

∂t<br />

+ (1 − ǫ b ) ∂qj i (x, t)<br />

∂t<br />

+ u j ∂Cj i (x, t)<br />

= 0 (1)<br />

∂x<br />

= K appl i (C j i (x, t) − Cj,eq i (x, t)) (2)<br />

The equilibrium between the liquid and solid phase is given by the linear isotherm:<br />

q j i (x, t) = f(Cj,eq i (x, t)) (3)<br />

Here ǫ b is the void fraction, C j i (x, t) is the concentration in the liquid phase <strong>of</strong> com-<br />

104<br />

ponent i in <strong>column</strong> j, q j i is the concentration in the solid phase, u j 105<br />

is the superficial<br />

106<br />

107<br />

liquid velocity in the jth <strong>column</strong>, C j,eq<br />

i (x, t) is the equilibrium concentration in the<br />

liquid phase, q j,eq<br />

i<br />

is the equilibrium concentration in the solid phase, andK appl i is the


108<br />

109<br />

110<br />

liquid-phase based mass transfer coefficient, respectively. The subscripts i correspond<br />

to chemical components, superscript j is the <strong>column</strong> index, N c is the total number <strong>of</strong><br />

components, and N Column is the number <strong>of</strong> <strong>column</strong>s.<br />

At the CSS, the concentration pr<strong>of</strong>iles at the end <strong>of</strong> a cycle must be identical to<br />

those at the beginning <strong>of</strong> the cycle. The dynamics <strong>of</strong> the concentration pr<strong>of</strong>iles in<br />

the whole cycle need to be considered. However, the problem size can be reduced by<br />

taking advantage <strong>of</strong> the symmetry <strong>of</strong> the SMB operation. By enforcing the following<br />

constraints, the pr<strong>of</strong>iles at the end <strong>of</strong> the next step are required to be identical to those<br />

at the beginning <strong>of</strong> the current step, thus shifting the entire pr<strong>of</strong>iles downstream:<br />

C j i (x, 0) = Cj+1 i (x, t step ), j = 1, ..., N Column − 1 (4)<br />

q j i (x, 0) = qj+1 i (x, t step ), j = 1, ..., N Column − 1 (5)<br />

C N Column<br />

i (x, 0) = C 1 i (x, t step) (6)<br />

q N Column<br />

i (x, 0) = q 1 i (x, t step) (7)<br />

111<br />

3.2 Column <strong>configurations</strong> and SMB superstructure<br />

In this section, the <strong>configurations</strong> shown in Figure 2 are discussed. Note that these<br />

<strong>configurations</strong> can be extracted from the SMB superstructure shown in Figure 3.<br />

Referring to this figure, the following stream and <strong>column</strong> velocities are defined:<br />

u j , u j R , uj E , uj D , uj F ≥ 0, j = 1..N Column (8)<br />

where u j R , uj E<br />

are velocities <strong>of</strong> raffinate and extract withdrawn from jth <strong>column</strong>, and<br />

u j D and uj F<br />

are velocities <strong>of</strong> desorbent and feed supplied to jth <strong>column</strong>. Note that all<br />

velocities remain constant within each step. Furthermore, the volume and component


alance equations between the j − 1 th and j th <strong>column</strong>s are given by:<br />

u j−1 − u j−1<br />

E<br />

u N Column<br />

− u N Column<br />

E<br />

− uj−1 R + uj D + uj F = uj , j = 2, . . .N Column (9)<br />

− u N Column<br />

R<br />

+ u 1 D + u1 F = u1 (10)<br />

C j−1<br />

i (L, t)(u j−1 − u j−1<br />

E<br />

− uj−1 R ) + C F,iu j F = Cj i (0, t)uj , j = 2, . . .N Column<br />

(11)<br />

C N Column<br />

i (L, t)(u N Column<br />

− u N Column<br />

E<br />

− u N Column<br />

R<br />

) + C F,i u 1 F = C1 i (0, t)u1 (12)<br />

where C F,i is the concentration <strong>of</strong> component i in the feed. Further, in order to prevent<br />

draining the desorbent and feed into the product streams without going through a<br />

<strong>column</strong>, the following constraint is implemented:<br />

u j − u j D − uj F ≥ 0 j = 1, . . .,N Column (13)<br />

112<br />

113<br />

114<br />

115<br />

116<br />

117<br />

118<br />

119<br />

If (13) is inactive, then the liquid from a <strong>column</strong> is transported to the next <strong>column</strong>.<br />

Otherwise, the circulation loop is cut open and all liquid is withdrawn in either or<br />

both <strong>of</strong> the product streams. We previously reported that this constraint is necessary<br />

not only to prevent undesirable operating parameters, but to ensure the robustness <strong>of</strong><br />

optimization (Kawajiri and Biegler, 2006a,c). Note that this superstructure em<strong>bed</strong>s a<br />

number <strong>of</strong> different non-standard <strong>configurations</strong> such as Three-zone, Three-zone with<br />

purging, and the standard SMB. Those <strong>configurations</strong> can be extracted by enforcing<br />

the flow constraints in Table 1.


120<br />

3.3 Optimization problem formulation<br />

In this study, we investigate the trade-<strong>of</strong>f <strong>of</strong> throughput maximization and desorbent<br />

minimization with fixed purity and recovery requirements by solving the following<br />

multi-objective optimization problem:<br />

(MOO) max ū F :=<br />

min ū D :=<br />

N Column<br />

∑<br />

j<br />

N Column<br />

∑<br />

j<br />

subject to : (1) − (13)<br />

(Extract Product Purity) =<br />

u j F<br />

(14)<br />

u j D<br />

(15)<br />

(Extract Product Recovery) =<br />

N Column ∑<br />

t step<br />

∫<br />

j=1 0<br />

N Column ∑ ∑Nc<br />

j=1<br />

N Column ∑<br />

i=1<br />

u j E (t)Cj E,Prod (t)dt<br />

t step<br />

∫<br />

0<br />

t step<br />

∫<br />

j=1 0<br />

N Column t<br />

∑ step<br />

∫<br />

j=1<br />

0<br />

u j E (t)Cj E,i (t)dt ≥ Pur min<br />

u j E (t)Cj E,Prod (t)dt<br />

(16)<br />

u j F (t)Cj F,Prod (t)dt ≥ Rec min<br />

(17)<br />

u l ≤ u j ≤ u u (18)<br />

121<br />

122<br />

123<br />

124<br />

125<br />

126<br />

where t step is the valve switching interval, or step time, Pur min and Rec min are the<br />

purity and recovery requirements <strong>of</strong> the desired product which should be recovered<br />

in the extract stream, respectively. The desired product is denoted by the index<br />

Prod. u u and u l are the upper and lower bounds on the zone velocities, respectively.<br />

In addition, to examine the performance <strong>of</strong> the individual <strong>configurations</strong> shown in<br />

Figure 2, the constraints shown in Table 1 are enforced.


127<br />

3.4 Solution strategy<br />

The problem formulated in the previous section is a nonlinear multi-objective optimization<br />

problem. There have been many solution methods proposed for this type <strong>of</strong><br />

problem. For SMB optimization, Hakanen et al. (in press,s) used an interactive multiobjective<br />

optimization approach for SMB optimization with <strong>four</strong> objective functions,<br />

throughput, desorbent consumption, purity, and recovery. Their approach is useful<br />

when the number <strong>of</strong> objective function is large. Zhang et al. (2002) used a genetic<br />

algorithm for the zone configuration problem for a five-<strong>column</strong> SMB. In this study,<br />

however, the ǫ-constrained method based on a Newton-based numerical optimization<br />

approach is employed since there are only two objective functions. In this method,<br />

parameter ǫ p and the two objective functions <strong>of</strong> MOO (14)-(15) are reformulated into<br />

the following:<br />

maxū F (19)<br />

subject to : ū D ≤ ǫ p (20)<br />

128<br />

129<br />

130<br />

131<br />

132<br />

133<br />

134<br />

135<br />

136<br />

137<br />

By changing the value <strong>of</strong> ǫ p and solving the above single-objective optimization problem<br />

repeatedly, the whole Pareto curve can be approximated. Although more computationally<br />

expensive than the interactive approach in Hakanen et al. (in press),<br />

this method allows us to compare the performance <strong>of</strong> different SMB <strong>configurations</strong><br />

by examining the entire Pareto curves.<br />

The resulting set <strong>of</strong> single-objective optimization problems is handled by the approach<br />

where the partial differential algebraic equations (PDAEs) are fully discretized<br />

both in the spatial and temporal domains using centered finite difference and Radau<br />

collocation on finite elements, respectively. Our previous study has found that this<br />

full-discretization approach significantly reduces the computational effort (Kawajiri


138<br />

139<br />

140<br />

141<br />

142<br />

143<br />

and Biegler, 2006d). It is important to note that although the resulting NLP problem<br />

is large, the linear system to be solved in each iteration, i.e., the linearized Karush-<br />

Kuhn-Tucker (KKT) conditions, has a sparse structure and can be solved very quickly.<br />

To satisfy this requirement, we choose IPOPT, an interior-point solver with a filter<br />

line search method which exploits the sparse structure <strong>of</strong> linearized KKT conditions.<br />

Details <strong>of</strong> IPOPT can be found in Wächter and Biegler (2005).<br />

144<br />

4 Case study<br />

145<br />

4.1 <strong>Comparison</strong> <strong>of</strong> different <strong>configurations</strong><br />

For the numerical case study we consider the separation <strong>of</strong> fructose and glucose. The<br />

parameters in Hashimoto et al. (1983), where the equilibrium is linear, i.e.,<br />

q j i (x, t) = K iC j,eq<br />

i (x, t) (21)<br />

146<br />

147<br />

148<br />

149<br />

150<br />

151<br />

152<br />

153<br />

154<br />

155<br />

156<br />

are used in this study and are given in Table 2. Throughout this study, the number<br />

<strong>of</strong> finite elements in the temporal domain N FET and collocation points N COL are set<br />

to 5 and 3 respectively, and the number <strong>of</strong> finite elements N Column is set to 40 per<br />

<strong>column</strong>. Details <strong>of</strong> the accuracy <strong>of</strong> the discretization method and choice <strong>of</strong> numbers <strong>of</strong><br />

finite elements were determined in Kawajiri and Biegler (2006d). This optimization<br />

problem is implemented within the AMPL modeling environment (Fourer et al., 1992).<br />

It should be also noted that because the optimization problem is non-convex, our<br />

approach does not guarantee finding the global optimum. Therefore, we use several<br />

different starting points in order to avoid local optimal solutions. The starting points<br />

are obtained in the following manner: an initialization problem is solved where all operating<br />

parameters (u j ,u j R ,uj E ,uj F ,uj D ,t step) are fixed in a standard SMB configuration.


157<br />

158<br />

159<br />

160<br />

161<br />

162<br />

163<br />

164<br />

165<br />

166<br />

167<br />

168<br />

169<br />

170<br />

171<br />

172<br />

173<br />

174<br />

175<br />

176<br />

177<br />

178<br />

179<br />

180<br />

181<br />

182<br />

Note that this initialization yields a set <strong>of</strong> linear equations which has no degrees <strong>of</strong><br />

freedom. It should also be noted that such a starting point satisfies all constraints <strong>of</strong><br />

the optimization problem except the purity and recovery requirements (16)-(17). Furthermore,<br />

multiple starting points can be easily obtained by changing the operating<br />

parameters arbitrarily and resolving the initialization problem.<br />

First, the throughput maximization problem (ǫ p = ∞ in (20)) is solved for each<br />

configuration. The result is summarized in Table 3. In all cases, the computational<br />

time takes only a few CPU minutes, which demonstrates the efficiency <strong>of</strong> our numerical<br />

optimization technique.<br />

The propagation <strong>of</strong> concentration pr<strong>of</strong>iles <strong>of</strong> the throughput maximization problem<br />

are compared in Figure 4. In the standard SMB configuration, concentrations <strong>of</strong> both<br />

<strong>of</strong> the faster and slower components are nearly zero at the right and left ends <strong>of</strong> the<br />

<strong>column</strong> train (Figure 4a). In other words, the faster component should not catch<br />

up with the slower component in order to achieve the purity and recovery demands.<br />

The optimizer finds such an operation because <strong>of</strong> the following reasons: Before the<br />

faster component reaches x = 8.0 in the rightmost <strong>column</strong>, switching must occur<br />

(the inlet and outlet streams are switched in the direction <strong>of</strong> the liquid flow) to<br />

avoid contaminating Zone I (leftmost <strong>column</strong>) by the faster component. At the same<br />

moment <strong>of</strong> switching, the concentration <strong>of</strong> the slower component at x = 0 must<br />

be nearly zero to avoid contaminating Zone IV (rightmost <strong>column</strong>) by the slower<br />

component. This “zone contamination” in Zones I/IV would lead to contamination<br />

<strong>of</strong> extract/raffinate by the faster/slower component.<br />

On the other hand, in the optimal concentration pr<strong>of</strong>iles <strong>of</strong> the Three-zone configuration<br />

shown in Figure 4b, the concentration <strong>of</strong> the faster component at the right<br />

end (x = 8.0) is significantly higher. However, this would not lead to contamination<br />

<strong>of</strong> the extract, since the recycle stream from x = 8.0 to x = 0 is cut <strong>of</strong>f. Therefore the


183<br />

184<br />

185<br />

186<br />

187<br />

188<br />

189<br />

190<br />

191<br />

192<br />

193<br />

194<br />

195<br />

196<br />

197<br />

198<br />

199<br />

200<br />

201<br />

202<br />

203<br />

204<br />

205<br />

206<br />

207<br />

208<br />

internal concentration pr<strong>of</strong>iles can be higher, which leads to the higher throughput.<br />

It should be also noted that the desorbent consumption is higher because the recycle<br />

stream is cut <strong>of</strong>f.<br />

The throughput <strong>of</strong> the Three-zone with purging has the lowest throughput; this<br />

is because the leftmost <strong>column</strong>, where purging takes place, does not contribute to the<br />

separation effectively, as the concentrations remain nearly zero all the time (Figure<br />

4c). We note that this conclusion may not hold in cases where the Henry constant<br />

<strong>of</strong> the slower component K 2 is large and re<strong>moving</strong> the slower component from Zone<br />

I requires a higher amount <strong>of</strong> desorbent.<br />

Finally, note that the throughput <strong>of</strong> the F-shaped configuration is the highest.<br />

The propagation <strong>of</strong> the concentration pr<strong>of</strong>iles in the F-shaped configuration is shown<br />

in Figure 4d. This configuration allows such a situation where the slower component<br />

is “left behind” by the switching, and remains in the rightmost <strong>column</strong>. The resulting<br />

slower component left in the rightmost <strong>column</strong> is collected in the extract stream. This<br />

leads to further increase <strong>of</strong> throughput, since the concentration in the leftmost <strong>column</strong><br />

can be significantly higher.<br />

We note that the above comparison <strong>of</strong> <strong>configurations</strong> cannot be generalized to<br />

other systems, and it is not easy to determine the configuration that attains the<br />

highest throughput a priori (We will see this in the following case study where we<br />

consider different values <strong>of</strong> Pur min ). Here, an alternative approach is to use the SMB<br />

superstructure, which em<strong>bed</strong>s all possible <strong>configurations</strong> and extracts the optimal<br />

configuration. For Pur min = 95%, the optimal configuration extracted from the superstructure<br />

turns out to be the F-shaped configuration (Table 3). This configuration<br />

and optimal operating parameters can be found simultaneously in 2.87 CPU minutes.<br />

These <strong>configurations</strong> can be compared by changing ǫ p . Figure 5(a) shows the<br />

Pareto sets <strong>of</strong> different <strong>configurations</strong> with Pur min = 0.95. The solid line shows the


209<br />

210<br />

211<br />

212<br />

213<br />

214<br />

215<br />

216<br />

217<br />

218<br />

219<br />

220<br />

221<br />

222<br />

223<br />

224<br />

225<br />

226<br />

227<br />

228<br />

229<br />

230<br />

231<br />

232<br />

233<br />

234<br />

Pareto set <strong>of</strong> the optimal solution <strong>of</strong> the superstructure formulation. As can be seen<br />

in the figure, it overlaps that <strong>of</strong> the standard SMB when the desorbent consumption is<br />

low. On the other hand, the part <strong>of</strong> the higher desorbent consumption overlaps that <strong>of</strong><br />

the F-shaped configuration. Also note that the Pareto curves <strong>of</strong> other <strong>configurations</strong>,<br />

Three-zone and Three-zone with purging, are below that <strong>of</strong> the superstructure. In<br />

the region <strong>of</strong> 1.5 ≤ ū D ≤ 3.0, however, none <strong>of</strong> the fixed <strong>configurations</strong> achieves the<br />

throughput <strong>of</strong> the superstructure solution; this optimal configuration is similar to the<br />

F-shaped configuration (Figure 6(a)) except that it has a recycle stream from x = 8<br />

to x = 0 in order to reduce desorbent consumption.<br />

Finally, the multi-objective optimization problem is solved for different values <strong>of</strong><br />

Pur min . In all cases, Pareto curve <strong>of</strong> the superstructure optimization becomes higher<br />

than those <strong>of</strong> other <strong>configurations</strong>. When Pur min = 0.90 (Figure 5b), the Pareto<br />

curves show a similar trend to those <strong>of</strong> Pur min = 0.95. The optimal configuration at<br />

ū D = 3.0 is shown in Figure 6 (b). However, when Pur min is decreased to 0.85 (Figure<br />

5c), the highest throughput is achieved by the Three-zone configuration at the expense<br />

<strong>of</strong> higher desorbent consumption. The optimal configuration and concentration pr<strong>of</strong>iles<br />

at ū D = 7.0 are shown in Figure 6(c). Finally, in the case <strong>of</strong> Pur min = 0.80, the<br />

F-shaped configuration never reaches the Pareto curve <strong>of</strong> the superstructure (Figure<br />

5d). Instead, the Three-zone with purging has the highest throughput at the highest<br />

desorbent consumption; this is closely followed by the Three-zone configuration with<br />

lower desorbent consumption. The optimal configuration at ū D = 5.0, shown in Figure<br />

6(d), is again similar to the F-shaped configuration, but the feed is supplied at<br />

x = 2.0 and the liquid is recycled from x = 8 to x = 0.<br />

From this multi-objective optimization study, we conclude that the optimal configuration<br />

is highly dependent on the purity specification, although the standard SMB<br />

configuration generally has good trade-<strong>of</strong>fs when desorbent consumption should be


235<br />

236<br />

kept low. Nevertheless, our optimization approach using the SMB superstructure is<br />

able to find the most efficient configuration without exploring various <strong>configurations</strong>.<br />

237<br />

5 Conclusions and Future Work<br />

238<br />

239<br />

240<br />

241<br />

242<br />

243<br />

244<br />

245<br />

246<br />

247<br />

248<br />

249<br />

250<br />

251<br />

Standard and nonstandard <strong>configurations</strong> <strong>of</strong> <strong>four</strong>-<strong>column</strong> SMB, under optimal operating<br />

conditions, have been compared through a case study. The optimal configuration<br />

is different depending on the purity requirement, throughput, and desorbent consumption.<br />

Here, we determined that although the standard SMB configuration has<br />

good trade-<strong>of</strong>fs when desorbent consumption is low, optimal configuration is highly<br />

dependent on the purity requirement. Nevertheless, our multi-objective optimization<br />

scheme using an SMB superstructure is able to find the optimal configuration<br />

efficiently without exploring various existing <strong>configurations</strong>. Also, our numerical optimization<br />

technique has been found to be efficient and reliable.<br />

Our future work will further extend our optimization approach to find optimal<br />

operating schemes for multi-component separations, where more than two components<br />

are fractionated into multiple streams. As our superstructure formulations em<strong>bed</strong> a<br />

number <strong>of</strong> both standard and nonstandard operating schemes, we aim to develop<br />

further efficient designs and operating schemes.<br />

252<br />

Acknowledgment<br />

253<br />

254<br />

Funding from the National Science Foundation under Grant CTS-0314647 is gratefully<br />

acknowledged.


255<br />

References<br />

256<br />

257<br />

258<br />

Ching, C.B., Chu, K.H., Hidajat, K., and Uddin, M.S. Comparative study <strong>of</strong> flow<br />

schemes for a <strong>simulated</strong> countercurrent adsorption separation process. AIChE<br />

Jounal, 38, 1744–1750 (1992).<br />

259<br />

260<br />

261<br />

Dünnebier, G. and Klatt, K.U. Modelling and simulation <strong>of</strong> nonlinear chromato-<br />

graphic separation processes: a comparison <strong>of</strong> different modelling approaches.<br />

Chemical Engineering Science, 55, 373–380 (2000).<br />

262<br />

263<br />

264<br />

Dünnebier, G., Fricke, J., and Klatt, K.U. Optimal design and operation <strong>of</strong> simu-<br />

lated <strong>moving</strong> <strong>bed</strong> chromatographic reactor. Industrial and Engineering Chemistry<br />

Research, 39, 2290–2304 (2000).<br />

265<br />

266<br />

Fourer, R., Gay, D.M., and Kernighan, B.W. AMPL: A Modeling Language for<br />

Mathematical Programming. Duxbury Press, Belmont, CA (1992).<br />

267<br />

268<br />

Golshan-Shirazi, S. and Guiochon, G. <strong>Comparison</strong> <strong>of</strong> the various kinetic models <strong>of</strong><br />

non-linear chromatography. Journal <strong>of</strong> Chromatography, 603, 1–11 (2003).<br />

269<br />

270<br />

271<br />

Hakanen, J., Kawajiri, Y., Miettinen, K., and Biegler, L.T. Interactive multi-objective<br />

optimization for <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> processes. Control and Cybernetics (in<br />

press).<br />

272<br />

273<br />

Hakanen, J., Kawajiri, Y., Miettinen, K., and Biegler, L.T. Interactive multi-objective<br />

optimization <strong>of</strong> superstructure SMB processes (submitted for publication).<br />

274<br />

275<br />

276<br />

Hashimoto, K., Adachi, S., Noujima, H., and Maruyama, H. Models for the separation<br />

<strong>of</strong> glucose/fructose mixture using a <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> adsorber. Journal <strong>of</strong><br />

Chemical Engineering <strong>of</strong> Japan, 16(5), 400–406 (1983).


277<br />

278<br />

279<br />

Kawajiri, Y. and Biegler, L.T. Large scale nonlinear programming for asymmetic<br />

design and operation <strong>of</strong> <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong>s. Journal <strong>of</strong> Chromatography A,<br />

1133, 226–240 (2006a).<br />

280<br />

281<br />

282<br />

283<br />

Kawajiri, Y. and Biegler, L.T. Large-scale optimization strategies for zone configuration<br />

<strong>of</strong> <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong>s. In 16th european symposium on computer aided<br />

process engineering and 9th international symposium on process systems engineering,<br />

(Eds.) W. Marquardt and C. Pantelides, 131–136. Elsevier (2006b).<br />

284<br />

285<br />

286<br />

Kawajiri, Y. and Biegler, L.T. A nonlinear programming superstructure for optimal<br />

dynamic operations <strong>of</strong> <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> processes. Industrial & Engineering<br />

Chemistry Research, 45(25), 8503–8513 (2006c).<br />

287<br />

288<br />

Kawajiri, Y. and Biegler, L.T. Optimization strategies for <strong>simulated</strong> <strong>moving</strong> <strong>bed</strong> and<br />

PowerFeed processes. AIChE Journal, 52(4), 1343–1350 (2006d).<br />

289<br />

290<br />

Ruthven, D.M. and Ching, C.B. Counter-current and <strong>simulated</strong> counter-current ad-<br />

sorption separation processes. Chemical Engineering Science, 44, 1011–1038 (1989).<br />

291<br />

292<br />

293<br />

Toumi, A., Hanisch, F., and Engell, S. Optimal operation <strong>of</strong> continuous chromato-<br />

graphic processes: Mathematical optimization <strong>of</strong> the VARICOL process. Industrial<br />

and Engineering Chemistry Research, 41, 4328–4337 (2002).<br />

294<br />

295<br />

296<br />

van Deemter, J.J., Zuiderweg, F.J., and Klinkenberg, A. Longitudinal diffusion and<br />

resistance to mass transfer as causes <strong>of</strong> nonideality in chromatography. Chemical<br />

Engineering Science, 5, 271–289 (1956).<br />

297<br />

298<br />

299<br />

Wächter, A. and Biegler, L.T. On the implementation <strong>of</strong> an interior point filter line<br />

search algorithm for large-scale nonlinear programming. Mathematical Program-<br />

ming, 106(1), 25–57 (2005).


300<br />

301<br />

302<br />

Zhang, Z., Hidajat, K., Ray, A.K., and Morbidelli, M. Multiobjective optimization<br />

<strong>of</strong> SMB and Varicol processes for chiral separation. AIChE Journal, 48(12), 2800–<br />

2816 (2002).<br />

303<br />

304<br />

305<br />

Ziomek, G., Antos, D., Tobiska, L., and Seidel-Morgenstern, A. <strong>Comparison</strong> <strong>of</strong> possi-<br />

ble arrangements <strong>of</strong> five identical <strong>column</strong>s in preparative chromatography. Journal<br />

<strong>of</strong> Chromatography A, 1116, 179–188 (2006).


Figure 1: Schematic diagram <strong>of</strong> 4 <strong>column</strong> SMB<br />

Figure 2: Examples <strong>of</strong> <strong>configurations</strong>: (a) standard SMB, (b) Three-zone (c) Threezone<br />

with purging (d) F-shaped.


Figure 3: SMB superstructure


Figure 4: <strong>Comparison</strong> <strong>of</strong> concentration pr<strong>of</strong>iles. Vertical dotted lines show the ports


Figure 5: Pareto curves


Figure 6: Configurations in Figure 5


Table 1: Configuration constraints<br />

Standard SMB 3-zone 3-zone purging F-Shaped<br />

u j D = 0, j ≠ 1 uj D = 0, j ≠ 1 uj D = 0, j ≠ 1, 2 uj D = 0, j ≠ 1<br />

u j F = 0, j ≠ 3 uj F = 0, j ≠ 3 uj F = 0, j ≠ 3 uj F = 0, j ≠ 3<br />

u j E = 0, j ≠ 1 uj E = 0, j ≠ 1 uj E = 0, j ≠ 1 uj E = 0, j ≠ 4<br />

u j R = 0, j ≠ 3 uj R = 0, j ≠ 4 uj R = 0, j ≠ 4 uj R = 0, j ≠ 3<br />

u 4 = u 4 R u 4 = u 4 R , u1 = u 1 E u 4 = u 4 E<br />

Table 2: Parameters <strong>of</strong> Fructose / glucose separation<br />

parameter value parameter value<br />

ǫ b 0.389 L [m] 2.0<br />

K 1 0.518 K appl 1 [1/s] 6.84 × 10 −3<br />

K 2 0.743 K appl 2 [1/s] 6.84 × 10 −3<br />

C F,1 [%] 50.0 C F,2 [%] 50.0<br />

u l [m/h] 0.0 u u [m/h] 8.0<br />

N Column 4 Rec min [%] 80.0<br />

N FET 5 N COL 3<br />

N FEX 40<br />

Table 3: Statistics <strong>of</strong> throughput maximization problem for Pur min = 0.95<br />

Objective function Φ Standard SMB 3-zone 3-zone purging F-Shaped Superstructure<br />

ū F [m/h] 0.410 0.521 0.344 0.737 0.737<br />

ū D [m/h] 1.199 4.836 2.083 4.669 4.669<br />

Extract purity [%] 95.0 95.0 95.0 95.0 95.0<br />

Extract recovery [%] 80.0 80.0 80.0 80.0 80.0<br />

Number <strong>of</strong> iterations 39 44 48 57 54<br />

CPU time [min] 1.76 1.65 1.87 3.12 2.87

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!