www.icheme.org/cherd doi: 10.

1205/cherd05027

02638762/06/$30.00+0.00 # 2006 Institution of Chemical Engineers Trans IChemE, Part A, May 2006 Chemical Engineering Research and Design, 84(A5): 363 369

THE MICROREACTOR A Systematic and Efcient Tool for the Transition from Batch to Continuous Process?
S. LOMEL1 , L. FALK1, J. M. COMMENGE1, J. L. HOUZELOT1 and K. RAMDANI2
1

nie ChimiqueCNRS Groupe ENSIC, Nancy, France Laboratoire des Sciences du Ge 2 Rhodia, Centre de Recherches de Lyon, Saint Fons, France

n the basis of the experimental works of Krummradt et al. (1999), this paper shows how it is possible, for certain systems of reactions, to reduce considerably the operations times from a semi-batch stirred tanks to a continuous microreactor, in which mixing and heat transfer are intensied. In the case of the Grignard reaction, we demonstrate that it is so possible to reduce the operation time from 5 h to less than 10 s. However, to reach the productivity of the industrial 6 m3 stirred tank, it is necessary to use ve minireactors in parallel. For such exothermic reactional systems, the eld of interest for an intensied reactor use as an industrial production tool is specied. Keywords: process intensication; feed batch reactor; continuous process transposition; exothermic reaction; microreactor.

INTRODUCTION There has been for several years an increasing interest for microreactors and more generally for processes intensication, bringing forth over 1500 publications in the last few years and several books (Stankiewicz and Moulijn, 2004; Hessel et al., 2004). Numerous articles praise the advantages of microreactors for the pharmaceutical or for the ne chemical industries, thanks to their excellent capability for mixing and for thermal exchanges which allows an increase in yields and selectivities of reactions. The experimental conditions can be tested and optimized directly at laboratory scale, and it is possible, thanks to the numbering-up technique consisting in extrapolating a process by putting devices in parallel, to propose fast development of production processes. The problems of scale extrapolation are then eliminated, so reducing considerably difculties, time and costs of studies allowing short time to market of new molecules. However, realized works consist very often in experimentally testing the relevance of microreactors and the conclusions stay in the empirical state. There are few detailed analytical works which allow us to understand the conditions in which the adoption of microreactors present a real interest or not. For that purpose, we have taken as base of study the experimental works of Krummradt et al. (1999), showing that it is possible to
nie Correspondence to: Professor L. Falk, Laboratoire des sciences du Ge Chimique, CNRS Groupe, ENSIC, Nancy, France. E-mail: laurent.falk@ensic.inpl-nancy.fr

reduce, for the same organic synthesis, the operation time from 5 h in an industrial 6 m3 stirred tank to less than 10 s in a microreactor. We propose, by a simple analysis based on characteristic times of the system, an explanation of these a priori surprising experimental results.

PRESENTATION OF THE EXPERIMENTAL RESULTS Krummradt et al. (1999), studied experimentally in various scales and in various structures of processes the exothermal synthesis (DR H 300 KJ mol1 ) of an organic compound of ne chemistry. The objective of this study aims at comparing the performances of a microreactor associated to a micro heat exchanger with regard to a classic stirred jacketed tank reactor. This organometallic synthesis includes side reactions (parallel and consecutive) which are thermally activated. The following main reaction is fast but the reaction rate is not exactly known:

To control the temperature of the system, the common practice is to operate these reactions in semi-batch stirred tanks. The feeding rate of one of the reagents is then the operating key parameter for the control of the reactor internal temperature. This conguration is adopted in the 363

364
Table 1. Experimental results in the semi batch conguration. Reactor type Laboratory (0.5 l) Industrial (6000 l) Feed time (h) 0.5 .5 Yield 88% 72%

LOMEL et al. Unfortunately there are numerous articles of this kind in the literature, where the authors test in an empirical way the microreactors performances without developing a detailed analysis of the results. The objective of the present article is to explain, thanks to a characteristic times analysis in various processes types, the different results obtained and to show if microreactors really constitute an effective alternative to conventional stirred tanks. To develop and expose our approach, we focus on three process structures: a semi batch and a continuous stirred tank and a plug ow reactor, in which the ow is assumed laminar. The following analysis is relevant for cases which are heat transfer limited.

experimental study as reference point. First of all, the reaction is operated in a 0.5 l laboratory semi-batch reactor and then scaled up in a 6000 l tank (scale up factor of 23 on diameters). The surface/volume ratio A/V for each tank are approximately equal to 80 m21 at the laboratory scale and 4 m21 at the industrial scale, orders of magnitude in agreement with the relation (A/V) (4.9 + 0.6)V 21/3 proposed by Schweich (2001). To respect the isothermal condition, the authors empirically adjusted the feed time in the laboratory reactor. The temperature at laboratory scale is maintained at 2 408C, while at the industrial scale, the reaction is operated at 2 208C, the cooling system power not being sufcient enough to reach lower temperatures. The experimental results are presented in Table 1. For the feed time adopted in the 6000 l reactor, the isothermal condition is not strictly respected and the authors observed an important temperature rise favouring the side reactions which partially explains the difference of yields. The difference between the operation times during the extrapolation is important. How to explain such an increase of feed time during the scale-up? Does this value arise from experimental attempts or is it calculated? Considering the size of the reactor, one can believe that this value was not experimentally adjusted and this shows the difculty of dening operating conditions during extrapolation. Krummradt et al. (1999) then suggested to operate this reaction in microreactors proposed by the Institute of Microtechnique of Mainz (IMM). At rst, they realized experiments in a system composed of 10 micromixers in parallel cooled at 2108C, each one including 32 oblong channels characterized by a length of 220 mm and a width of 40 mm for a ratio A/V 104 m21. Because of fouling problems leading to frequent stops of the installation, they adopted minireactors characterized by a ratio A/V 4000 m21 with a ow rate of 3 1025 m3 s21. The results are presented in Table 2. The use of microreactors appears very efcient for this reaction, especially since ve minireactors in parallel are able to deliver the same productivity as the semi-batch industrial tank. The thermal transfer intensication, associated to a better mixing, leads to very short residence times with a high yield. However, the authors do only present the obtained raw values and bring no explanation to these a priori surprising results. Were such results foreseeable? Is it absolutely necessary to use a micromixer presenting important risks of plugging or is it possible to work on an intermediate scale presenting practical advantages combined with high thermal transfer performances?
Table 2. Experimental results with microreactors. Reactor type IMM microreactor Minireactor Residence time , 10 s , 10 s Yield 95% 92%

CHARACTERISTIC TIME ANALYSIS The methodology consists of establishing mass and heat balances in isothermal condition for which parallel and consecutive reactions are assumed not to occur. So, the expression of the operation characteristic time will be determined by solving the mass and heat balance equations. The case of the semi-batch reactor will be treated in details and only the nal results in the others process structures will be presented.

Semi Batch Reactor The adopted notations are specied in Figure 1. The expressions of heat balance and mass are given in equations (1) and (2) respectively: UA(Tp T ) (V rCp GR ) dT dt QrCp (T TE ) DR H rp V dC A Q(CA CA0 ) rp V V dt

(1) (2)

The reactor isothermicity condition imposes dT/dt 0 and setting rp . V from the equation (2) in equation (1), the differential equation (3) is established: atf t UA(Tp T ) dCA CA CA0 QDR H dt

r C ( TE T ) DR H

(3)

where a V0/Vf is the ratio of the volume V0 present in the tank at initial conditions to the fed volume Vf. The reactant concentration variation in the reactor, described by equation (3), is a function of the adiabatic temperature rise, J, the characteristic time of heat exchange ttr and the feed time tf, dened as:   tf ( TE T ) CA CA0 1 (Tp T ) (a 1) J J ttr     0 t ( DR H )CA0 (4) , J a tf t rC p

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369

THE MICROREACTOR

365

Figure 1. Adopted notations in the fed batch stirred tank case.

where
0 CA0

CA0 1a

where the characteristic time of heat exchange is the same as in expression (5). We can notice that the operation characteristic time depends on the characteristic time of heat exchange in a proportionally way as previously. Plug ow reactor The mass and heat balances resolution lead to the following expression of the residence time:

and Vf 1 V tf Q 1aQ The characteristic time of heat transfer is expressed ttr

t
(5)

r V Cp UA

J X ttr T TP

(8)

This time accounts for the speed of the heat transfer: a system characterized by a low time for heat transfer allows to quickly remove the heat produced by the reaction and conversely. This term is a constant for a constant heat transfer coefcient U, which is the case when the vessel heat transfer is shell side limited. Generally, the adoption of a classic stirred jacketed tank reactor conducts to a shell side heat transfer limitation. For this kind of reactor, the characteristic time is directly proportional to the diameter D of the tank. The expression of the operation characteristic time, i.e., the feed time, which is related to the reactant concentration by equation (4), is the following:   TE T ttr J X , tf J (a 1) ( T Tp ) C 0 CAf where X A0 0 (6) CA0 As the result, the characteristic time of heat exchange is the sensitive parameter to scale up the process and x the operation characteristic time. As can be noticed in equation (5), this operation time is independent of the reaction kinetic due to the isothermicity constraint.

In the case of microreactor, it can be reasonably assumed that the ow is laminar and fully developed. In that case, the Nusselt number is constant and the characteristic time of heat transfer writes: ttr

rCP D2 4lNu

(9)

where D is the diameter of the channel. It can be noticed, that this characteristic time, proportional to the channel dimension to the square, strongly decreases when the diameter in the mini or microreactor diminishes. This explains the strong impact of miniaturization on intensication, compared to classical stirred tanks with the characteristic times of heat transfer only linearly dependent on D (Commenge et al., 2005). From equations (6) (8), it can be seen that for all the three structures of processes, the operation time is proportional to the characteristic time of heat transfer, logical result in a heat transfer limited scenario, and to a function cprocess which is characteristic of the structure of the process as follows:

top eration cprocess (J , X ) ttr

(10)

Continuous Process Continuous stirred tank reactor The same reasoning allows to determine the characteristic operation time, i.e., the residence time, t as follows:   T S TE ttr J X (7) t ( TP T ) J

Numerical application shows that function cprocess has almost the same order of magnitude for the three types of reactors investigated here, so the key parameter to modify the operation time is the time for heat transfer and not the process structure. Generally, equation (10) illustrates how it is possible to reduce the operation characteristic time by reducing the characteristic time of heat exchange. Therefore, the reduction of the process size, which decreases the characteristic time of heat exchange, explains the very low values for the residence time in the microstructured reactor compared to the large vessel.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369

366 ANALYSIS VALIDATION: FIELD OF MICROSTRUCTURED REACTOR INTEREST? Comparison with Experimental Results

LOMEL et al. strongly the operation characteristic time with the detriment of the productivity. In order to meet the required industrial production, the numbering-up method has been applied and showed its success. However, in the case of large production, have the miniaturization and numbering-up delivered profound advantages?

We estimate the operation characteristic time in the different processes structures thanks to the previous determined relations. For a given conversion yield, we assumed the following conditions to calculate the characteristic parameters of the process: T TS, T 2 TE TS 2 TE T 2 TP 20 K, CA 1 mol L21, (A/V ) (4.3) . V 21/3 and U 200 W m22 K21 for the stirred tank (heat transfer limitation in the jacket). To simplify the case of the plug ow reactor, the theoretical value of 3.66 is assumed for the estimation of the Nusselt number for fully developed temperature proles. The results presented in Table 3 show that the best efciency in term of operation characteristic time is obtained for the mini and micro plug ow reactor, running in a laminar ow, followed by the continuous and semi batch stirred tank. We highlight the effect of the miniaturization which allows to signicantly reduce the operation characteristic time in the plug ow conguration. In the semi-batch process, we calculate a feed time of 7 h. For comparison, the adopted experimental feed time, 5 h, has not enabled to respect the isothermal criteria; it can be checked by the good agreement between the experimental data and the results predicted by our analysis. The case of the plug ow reactor, with a channel diameter equal to 100 mm, leads to a very short space time (5 1023 s) which may be not realistic. Indeed, the chemical regime will probably be reached and the reaction rate must be the limiting phenomenon which will x the space time as a consequence. The following paragraph intends to design a microreactor which is likely to represent the experimental data. A minireactor composed of 10 cm length rectangular channels with a hydraulic diameter of 1 mm is considered. A laminar ow is assumed and the value of the Reynolds number is supposed equal to 100, which corresponds to a linear velocity of 0.1 m s21. Garimella et al. (2001), reported the Nusselt number value of 5.5 for the chosen conditions which leads to a heat transfer coefcient of 3300 W m22 K21. The characteristic time of heat exchange and the operation time is estimated respectively to 0.3 s and 1 s for each channel. The acceptable ow rate of such a minireactor, composed of 380 channels, is equal to the experimental minireactor composed of 320 channels. The proposed methodology allows to nd with a good agreement the experimental conditions and the minireactor design reported by Krummradt et al. (1999). The intensication of the characteristic time of heat exchange, via the channels miniaturization, reduces

Field of Interest To approach this aspect, attention must be paid to the global process productivity. To quantify this term, we consider here, for each of the processes structures, the evolution of the productivity according to the characteristic size of the system. Fed batch reactor The productivity of the reactor can be estimated by the ratio of the moles of converted organic compound divided by the operation time required to carried out the reaction: PFBR
0 V CA0 X tf

Setting the expression of the feed time [equation (6)], the productivity is expressed as a function of the tank diameter: PFBR  V V D3  D2   tf ttr D

As a result, the productivity can be related to the reactor geometric parameter and is proportional to the diameter to the square.

Continuous stirred tank In the same way,

A similar dependence between the productivity and the reactor diameter is established.

Plug ow reactor In laminar ow, the Nusselt number is equal to 3.66 and independent of the channel geometric parameters, so that the heat transfer coefcient is inversely proportional to the channel diameter. Thus, the productivity can be

Table 3. Predicted values for the operation characteristic time. Reactor Fed batch Continuous stirred tank Plug ow reactor (minireactor) Plug ow reactor (microreactor) Diameter (m) 2 2 0.001 0.0001 U (W m22 K21) 200 200 2200 22000 A/V (m21) 2.4 2.4 4000 40000 Characteristic time of heat transfer 2.5 h 2.5 h 5 1021 s 5 1023 s Operation characteristic time 7h 5.6 h 1.5 s 1.5 1022 s

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369

THE MICROREACTOR expressed as follows:

367

V 0 V D2 L CA0 X   L ttr D2 t

of the obvious purposes is to minimize the objective function to optimize the intensied reactor design. The determination of the objective function minimum, considering the channel diameter as the variable, leads to the following expression of the optimal diameter:   wpressure drop 512 m Q (1=6) wvolume p2 N

The productivity is independent of the channel diameter and only depends on the channel length, so that, with xed length, the diameter size decreasing has no impact on the productivity. To equalize the industrial productivity of the stirred tank reactor of volume Vtank, the number of elementary channels is given by the ratio of the industrial fed batch reactor to productivity of the elementary plug ow reactor: N PFBR PPFR

Doptimal

(13)

This leads to the following relation: N

(11)

With the microreactor design proposed to represent the experimental results, the number of channels is estimated to 3200 using equation (11). As a result, nine microreactors are able to deliver the industrial productivity, or ve microreactors working two times longer in parallel which is the experimental adopted conguration by Krummradt and coworkers (1999). Thus, the whole process volume is estimated to Vinstallation 2(pD 2/4)L N, where 2 is an arbitrary weighting factor to take account of the thickness of the metal walls, channels connections and distribution chambers. The process volume scales as the channel diameter to the square, that shows that a very small volume can be attained by using very small channel diameter. However, the pressure drop, estimated by the Poiseuille relation. DP 128 m Q L p D4 where m is the viscosity, is inversely proportional to the channel diameter to the fourth, which reveals the main drawback of mini and microreactors. In fact, it is obvious that between a very small microreactor inducing high pressure drops and a larger minireactor but saving energy and pumping equipments costs, one should nd an optimal technical solution corresponding to an intermediate channel size. So that to dene an optimal characteristic size for the channel diameter, an objective function F is introduced, which takes into account the installation volume, the pressure drop per channel and the global process productivity as follows:

In the present case, Figure 2 plots the variation of the optimal diameter with respect to the ratio Wpressure drop/Wvolume. If the pressure drop does not constitute the most important parameter for the objective function optimization, the previous ratio is less than 1 and conversely. However, whatever the considered case, it is highlighted that the millimetre scale constitutes the relevant order of magnitude for the channel diameter and appears to be the best compromise between the volume intensication, the pressure drop and the global process productivity. For an identical productivity with regard to the industrial stirred tank, Figure 3 shows how it is possible, by using intensied micro and minireactor characterized by a channel diameter of some millimetres, to decrease the installation volume while controlling the pressure drop which can become particularly prohibitive for very small channel. For channel diameters below centimetre scale, the whole reactor volume is less than 50 l. However, for channel diameters with an order of magnitude of some hundred of micrometers, the pressure drop and the risk of plugging the installation due to product deposition are the limiting parameters to ensure the use of microreactors at industrial scale (318 bars/m21 for a 100 mm diameter). By adopting the minireactor we have designed (channel size equal to 1 mm), Figure 4 illustrates the strong reduction of the installation volume compared to classic process congurations. In the case of large production, the required channel number increases as a function of the tank diameter to

F 1 wvolume V wpressure drop DP 3 X 1 wproductivity where Wi 1 PPFR i1

(12)

wi is a dimensionless weighting factor, between 0 and 1, introduced to take account of the considered parameter importance (volume, pressure drop or productivity). One

Figure 2. Channel optimal diameter plotted against the ratio of the pressure drop and the installation volume weighting factors.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369

368

LOMEL et al. millimetres is the most relevant dimension to succeed the transition of a semi-batch to a continuous process via the integration of intensication tools. The intensication of heat transfer, resulting in smaller characteristic time, which implies a decreasing of the operation time, leads to a reduction of the process volume. However, each channel has a very weak productivity and the benets of using microreactors are dened as a function of the industrial vessel diameter to be replaced. Among the various technical solutions studied by Krummradt et al. (1999), the solution of a semi-batch reactor equipped with an external thermal heat exchanger delivering higher heat exchanges coefcients was not considered. Nevertheless, it is known that this structure allows to deliver coefcients from four to ve times higher than in classic jacketed stirred tank, so decreasing the operation time approximately 1 h. On the only level of the thermal control, it may thus be possible that the minireactor is not the best technico-economical solution with regard to the stirred tank equipped with an external heat exchanger. On the other hand, the Grignard reaction is a fast reaction in which selectivity is sensitive to mixing. The control of the mixing process in large stirred tank is particularly difcult and the minireactor, in which mixing is intensied, has demonstrated its higher performance. This shows that to completely answer the question posed by the title of the paper requires application of a technico-economical study to the choice of investments.

Figure 3. Process volume and pressure drop evolutions versus the adopted channel diameter.

the square and the intensied reactors integration in the production process has to be considered and claried through a technico-economical evaluation. Moreover, in order to transpose the methodology to other exothermic reactional systems, the constraints of mechanical resistance and manufacturing of the walls, which are functions of the exchanged thermal power, have to be taken into account.

CONCLUSIONS On the one hand, we have demonstrated that a simple methodology, based on the characteristic times analysis, allows to explain the a priori surprising experimental work of Krummradt et al. (1999), in the different process structures characterized by heat transfer limited conditions. On the other hand, the eld of interest by using intensication tools, in laminar ow, has been determined in order to meet industrial production. A characteristic size of some
a A Ci Ci0 0 Ci0 Cif CP D DP DRH J L N Nu PFBR PCST PPFR Q rP t ttr tf T TE TP TS U V V0 Vf wi X

NOMENCLATURE
ratio of V0 to Vf, dimensionless heat exchange surface of the reactor, m2 concentration of component I, mol m23 initial concentration of component i, mol m23 initial concentration of component i after reactants volume mixing, mol m23 concentration of component i when t tf, mol m23 mass caloric capacity, J kg21 K21 reactor diameter, m pressure drop, Pa reaction enthalpy, J mol21 adiabatic temperature rise, K channel length, m number of elementary channels Nusselt number productivity of the fed batch reactor, mol s21 productivity of the continuous stirred tank, mol s21 productivity of the plug ow reactor, mol s21 ow rate, m3 s21 reaction rate referred to the total liquid volume, mol m23 s21 time, s characteristic time for heat transfer, s feed time, s temperature, K feed uid temperature, K temperature of the tank wall, K outlet uid temperature, K overall heat transfer coefcient, W m22 K21 volume, m3 volume in the tank at the initial condition, m3 fed volume, m3 weighting factor of the parameter i, dimensionless yield

Figure 4. Installation volume and operation characteristic time (space time, residence time or feed time) evolutions for the considered process structures and an identical productivity.

Greek symbols l thermal conductivity, W m21 K21 GR mass reactor caloric capacity, J kg21 K21

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369

THE MICROREACTOR
r m t
volumique mass, kg m23 viscosity, Pa s21 residence time, s

369

REFERENCES
Commenge, J.M., Falk, L., Corriou, J.P. and Matlosz, M., 2005, Analysis of microstructured reactor characteristics for process miniaturization and intensication, Chem Eng Technol, 28(4): 446. Garimella, S., Dowling, W., Veen, M.V.D. and Killion, J., 2001, Heat Transfer Engineering, 22(6): 12.

we, H. (eds), 2004, Chemical Micro Process Hessel, V., Hardt, S. and Lo Engineering (Wiley-VCH). Krummradt, H., Kroop, U. and Stoldt, J., 1999, Experiences with the use of microreactors in organic synthesis, Proceedings of the Third International Conference on Microreaction Technology, 181. nie de la Re action Chimique (Tec&Doc, Schweich, D. (ed.), 2001, Ge Paris, France). Stankiewicz, A. and Moulijn, J.A. (eds), 2004, Re-Engineering the Chemical Processing Plant (Marcel Dekker). The manuscript was received 11 October 2005 and accepted for publication after revision 28 March 2006.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A5): 363 369