Base catalyzed ethanolysis of soybean oil in microreactors: Experiments and kinetic modeling

Chemical Engineering Science 104 (2013) 610–618

Contents lists available at ScienceDirect

Chemical Engineering Science journal homepage: www.elsevier.com/locate/ces

Base catalyzed ethanolysis of soybean oil in microreactors: Experiments and kinetic modeling Stefan Schwarz a, Ekaterina S. Borovinskaya a,b, Wladimir Reschetilowski a,n a b

Dresden University of Technology, Institute for Industrial Chemistry, Zellscher Weg 19, D-01069 Dresden, Germany Saint-Petersburg State Institute of Technology (Technical University), System Analysis Department, Moskovsky Avenue 26, 190013 St. Petersburg, Russia

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T


Continuous ethanolysis of soybean oil was carried out in seven microreactors.
Micromixing efficiency was analyzed by modified Villermaux–Dushman-method.
Better performance of microreactors was observed compared to batch process.
Kinetic modeling under consideration of mass transfer limitations was performed.
Heavy influence of reactor geometry on kinetics was observed.

art ic l e i nf o

a b s t r a c t

Article history: Received 14 June 2013 Received in revised form 5 September 2013 Accepted 23 September 2013 Dedicated to Prof. Dr.-Ing. Gerhard Emig on the occasion of his 75th birthday. Available online 1 October 2013

The base catalyzed alcoholysis of vegetable oil is a common industrial process for the production of biodiesel. The performance of the methanolysis reaction could be improved by applying continuous processes and using microreactors because of the strong mass transfer limitation of this reaction. This limitation is often neglected for the base catalyzed ethanolysis reaction. The KOH catalyzed ethanolysis of soybean oil was carried out in continuous and in batch. The experimental results and kinetic modeling of the data show strong mass transfer limitations of the reaction. The limitation is reduced by increasing the efficiency of micromixing which leads to performances exceeding the batch process under otherwise identically reaction conditions. & 2013 Elsevier Ltd. All rights reserved.

Keywords: Ethanolysis Soybean oil Mass transfer Microreactor Kinetic modeling

1. Introduction The ongoing industrial development worldwide leads to a high inquiry of diesel fuels, especially in the rural and transport sector.

n

Corresponding author. Tel.: þ 49 351 4633 7056; fax: þ 49 351 4633 2658. E-mail addresses: [email protected], [email protected] (W. Reschetilowski). 0009-2509/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ces.2013.09.041

Due the increasing shortage of mineral oil there is a growing demand on alternative resources. The heavy usage of fossil fuels will lead to an increasing environmental impact. Regarding these economic and ecologic problems the industrial production of biofuels derived from renewable sources is an important field of research (Meher et al., 2006; Stamenković et al., 2011). Biodiesel is a mixture of esters of low aliphatic alcohols and high fatty acids obtained by alcoholysis of biological feedstocks with a high amount of triacylglycerols. In industrial scale biodiesel

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

is commonly produced by homogenous base catalyzed methanolysis or ethanolysis of vegetable oils applying NaOH or KOH as catalyst (Stamenković et al., 2011). The most common feedstocks are highly refined rapeseed, sunflower, soybean or palm oils (Santacesaria et al., 2012b). To achieve high yields under industrial conditions the raw materials must not contain free fatty acids and water above 0.5 wt% (Moser, 2009). Otherwise the reaction is strongly hindered because of soap formation and catalyst deactivation. The reaction occurs in three reversible consecutive steps shown in Eqs. (1)-(3). Triacylglycerol (T) reacts with the alcohol (ROH) to diacylglycerol (D) and one equivalent of fatty acid ester (E). In two additional steps monoacylglycerol (M) and finally glycerol (G) is formed, each under formation of additional E (Richard et al., 2013). T þROH⇄D þ E

ð1Þ

D þ ROH⇄M þ E

ð2Þ

M þ ROH⇄G þ E

ð3Þ

Methanol is the most common used alcohol in biodiesel production because of its favorable physico-chemical properties (Stamenković et al., 2011). Methanol is usually produced from petrochemical materials. Ethanol on the other hand is produced in a high amount from biomass and is therefore from high interest for the production of completely bio-based biodiesel (Moser, 2009). The rate determing step for each reaction is the nucleophilic attack of an alkoxide on a carbonyl carbon of the acylglycerol leading to a tetraedric intermediate (Kulkarni et al., 2007). The rate of the homogeneous base catalyzed alcoholysis depends on the reactivity of the alkoxide. Since the reactivity of methoxide is higher than the reactivity of ethoxide the reaction rate of the methanolysis is higher. In general there are three stages of the transesterification reaction (Marjanović et al., 2010). The initial phase is highly influenced by mass transfer limitations because the reactants are not completely miscible and two phases are formed. In the following stage the reaction is controlled by the chemical kinetics and continues hereby much faster. In the final step the chemical equilibrium is setting up slowly. While at the methanolysis reaction all stages have been thoroughly investigated, mass transfer limitations of the ethanolysis reaction often can be neglected because of the much higher solubility of vegetable oil in ethanol. The solubility of soybean oil in methanol is 5.7 g l  1 at 30 1C (Boocock et al., 1996) while ethanol can solve 58 g l  1 at the same temperature (Gandhi et al., 2003). Most commercial processes use a large excess of alcohol and a high catalyst loading in order to reach high yields of biodiesel. Under industrial conditions mass transfer limitations do not influence the ethanolysis reaction (Richard et al., 2013). Common reaction conditions for ethanolysis and methanolysis are a molar alcohol–oil-ratio of 6:1, 0.5 wt% of alkali catalyst (with respect to triacylgylcerol) and a temperature close to the boiling point of the alcohol. The reaction is usually carried out in a batch reactor with above 600 rpm for 1 h to give total conversion to fatty acid esters (Moser, 2009). However it is desirable to use as less alcohol and catalyst as possible in order to reduce the costs of production and product separation. Various studies upon the continuous methanolysis reaction in microreactors showed high yields in short reaction times and under mild conditions (Qiu et al., 2010; Sun et al., 2008; Wen et al., 2009). Sun et al. (2008) carried out the KOH catalyzed methanolysis of rapeseed oil in tube reactors. Reducing the inner diameter of the tube from 2.0 mm to 0.25 mm lead to an increase of the yield of esters from 80% to 98.8% at a reaction time of 6 min at 60 1C and a MeOH–oil-ratio of 6:1. According to the authors the interfacial area increased due the decrease of the tube diameter which led to an improved mass transfer. By utilizing a split-and-

611

recombination micromixer combined with a tube reactor Guan et al. (2008) observed yields of 99% at the KOH catalyzed methanolysis of sunflower oil after a reaction time of less than 2 min at 60 1C. The MeOH–oil-ratio was quite high (22.9:1). Application of an unstructured t-shaped mixer under identical conditions lead only to yields of 50%. The increased mixing at the beginning of the reaction formed higher interfacial areas and caused an acceleration of the reaction. The ethanolysis reactions have not been thoroughly studied yet. Investigations considering mass transfer limitations of ethanolysis reactions were published by Richard et al. (2013) recently. The KOH catalyzed ethanolysis of sunflower oil was carried out in a capillary reactor combined with a t-shaped mixer at 65 1C and various EtOH–oil-ratios. Using a EtOH–oil-ratio of 6:1 a yield of 82% after 12 min was observed (Richard et al., 2013). In this work the KOH catalyzed ethanolysis of soybean oil is carried out in seven different continuous reactor systems and in batch reactor. Commercially refined soybean oil is used because of its low amount of water and free fatty acids (Bertoldi et al., 2009). By varying the total flow rate, reaction time and reactor geometry independently the effects on the yields of fatty acid ethyl esters (FAEE) were analyzed. The effectivity of micromixing in the micromixers was investigated by a modified Villermaux–Dushman-method. The kinetic data obtained was modeled by modification of a kinetic model proposed by Richard et al. (2013). It was shown that the base catalyzed ethanolysis of soybean oil is controlled by mass transfer limitations. The application of continuous reactors with a high effectivity of micromixing leads to an acceleration of the reaction. The continuous process outperforms the batch process under otherwise identically conditions.

2. Experimental 2.1. Reactor composition The continuous ethanolysis of soybean oil was carried out in microreactors which were assembled by combining three mixers each with three different tube reactors. The experimental setup is shown in Fig. 1. The reactants were provided into the mixer C1 by two syringe pumps P1 and P2. The outlet of the mixer was connected by a PTFE-tube with an inner diameter of 1.8 mm to the tube reactor C2. The outgoing stream was captured in a container B1. The microreactor was tempered by a water bath B2, W1. The specifications of the three continuous mixers used in this work are shown in Table 1. The T-mixer made from stainless steel

Fig. 1. Experimental setup for the continuous ethanolysis reaction.

612

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

Table 1 Specifications of the micromixers used in this work. Name, supplier

Scheme

Material, mixing principal

Channel diameter, mixing volume

T-mixer, Swagelok

Stainless steel, t-mixing

2.3 mm, 0.07 ml

LTF-MS, Little Things Factory

Borosilicate glass, lamellar mixing

1 mm, 0.2 ml

LTF-MX, Little Things Factory

Borosilicate glass, split and recombine

1 mm, 0.2 ml

had an inner diameter above the mm scale and was therefore not considered a micromixer. The devices LTF-MS and LTF-MX were provided by company Little Things Factory (König, 2010). The mixing structures of these devices were etched into plates made of borosilicate glass. The inner diameter given in Table 1 represents the dimension of the reactor input and output. The structures shown in the schemes had dimensions below the mm scale. Both devices were therefore considered micromixers. The mixing in device LTF-MS occurred by changes of the inner diameter along the mixing distance. In micromixer LTF-MX the incoming flow was splitted up into two streams which cross each other periodically over the mixing distance. This lead to a very efficient micromixing according to the supplier. The mixers were connected to three different tube reactors. The tubes LTF-V and LTF-VS with inner diameters of 1 mm were etched into borosilicate plates and had therefore a defined volume of 1.7 ml and 1.1 ml, respectively. While the circular tube of LTF-V was unstructured, LTF-VS showed the same geometry as the mixing range of micromixer LTF-MS. PTFE-capillaries with a channel diameter of 1.8 mm and different volumes between 1.3– 16.6 ml were used as well. The combination of the mentioned mixers and tubes led to seven different reactor setups shown in Table 2. The reactors MSV, MX-V and T-V were used at temperatures 30–60 1C while the other experiments were carried out at 30 1C and 40 1C.

2.2. Kinetic experiments The reactants, KOH dissolved in EtOH and soybean oil, were provided by two syringe pumps P1 and P2 threw 1/16 in. PTFE tubes to the micromixer C1. The flow continued into a tube reactor C2 and the outgoing product was captured in a vessel B1. Mixer and tube were connected by 1/16 in. UNF fittings threw short PTFE capillaries. The reactor system was sunken into a heated waterbath B2, W1. Samples were taken by collecting about 3 ml of reaction mixture in vials containing 8 ml 0.75 wt% aqueous HCl. The reaction was quenched immediately and two phases are formed. Batch processes were performed by adding 20 ml of a solution of 1.46 g KOH in 50 ml EtOH to 49.62 g (53.76 ml) soybean oil provided in a 100 ml heated glass vessel under magnetically stirring with 1000 rpm. Samples of about 2 ml were taken periodically and quenched in 0.75 wt% aqueous HCl.

2.3. Effectivity of micromixing The experimental setup in Fig. 1 was modified by connecting the micromixer C1 with a PTFE capillary directly to a UV detector K-2500 (Knauer). An aqueous buffer solution containing 60 mM H3BO3, 30 mM NaOH, 10 mM NaI and 2 mM KIO3 was prepared and mixed with an equal volume flow of 5 mM H2SO4. The absorption at 286 nm was measured for total flow rates between 0.25 ml min  1 and 2 ml min  1. The experiment was done twice for each mixer and the mean of the results was calculated. 2.4. Analysis A HPLC system (Knauer) equipped with a pump K-500, UV detector K-2500 with an analytic flow cell A4061, a manual injector A1365 and a 250 mm  3 mm column Eurosphere-C18100-5 with a 30 mm  3 mm pre column was used. The column was heated to 40 1C in a waterbath. Samples were injected manually threw a 20 ml injection loop. Acetonitrile/acetone (7:3 v/v) was used as eluent. Analysis run 40 min with a flow rate of 1 ml min  1. Wavelength detection was at 205 nm with a time constant of 1 s. Peak areas were determined with software Eurochrom 2000. The upper phase of collected samples containing soybean oil, FAEE, di- and monoacylglycerole was analyzed by HPLC. Therefore 100 mg of the sample were diluted with 2 ml nheptane. The content of triacyglylcerol and FAEE was determined by calibration with external standards. Standards for FAEE were prepared by mixing 8.5 g of soybean oil with a solution of 100 mg KOH in 3.4 ml EtOH for 5 h at room temperature under magnetically stirring. 40 ml 0.75 wt% aqueous HCl were added and the biphasic mixture was kept in a separation funnel for 12 h. The lower aqueous phase was removed and the product was analyzed by HPLC. Crude soybean oil was used as standard for triacylglycerol. For every analysis standards were prepared by diluting mixtures of 10–100 mg FAEE and 10–100 mg oil with 2 ml n-heptane. Concentrations of the standards were calculated according to Eqs. (4) and (5) and a calibration function was received by linear regression of the area of the standard in the chromatogram and the concentration. coil ¼ moil =ð2 ml þ ðmoil þ mFAEE Þρoil Þ

ð4Þ

cFAEE ¼ mFAEE =ð2 ml þ ðmoil þ mFAEE Þρoil Þ

ð5Þ

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

Table 2 Resulting reactor setups and conditions. Name Micromixer, tube

Reactor volume, flow rate

Annotations

MS-V MX-V T-V MS-R

2.26 ml, 0.18–1.90 ml min  1 2.26 ml, 0.14–0.29 ml min  1 2.14 ml, 0.17–1.80 ml min  1 1.30–9.04 ml, 0.216– 1.507 ml min  1 1.80–9.04 ml, 0.301– 1.818 ml min  1 3.33–13.86 ml, 0.555– 2.311 ml min  1 1.45 ml, 0.17–1.80 ml min  1

Constant Constant Constant Constant time Constant time Constant time Constant

LTF-MS, LTF-V LTF-MX, LTF-V T-mixer, LTF-V LTF-MS, tube

MX-R LTF-MS, tube T-R

T-mixer, tube

T-VS

T-mixer, tube

volume volume volume reaction reaction reaction volume

Standards for the by-products D and M were not available. The chromatograms showed two peaks which did not correspond to FAEE or T. It was assumed that these peaks derive from the side products. The total area of the additional peaks was determined to calculate the amount of side products (SB). For calibration the area of FAEE in the calibration sample with the lowest FAEE content was determined. It was always slightly higher than the total area of SB. A two point calibration for SB was performed by assuming a total area of 0 if the concentration of SB is 0. The analysis gave weight fractions for T, SB and FAEE according to Eq. (6). wanalyte ¼ manalyte =ðmT þ mNB þ mFAEE Þ

ð6Þ

2.5. Materials All materials were used as received: CH3CN (VER, 499.9%), H3BO3 (VEB Laborchemie Apolda, p.a.), EtOH (Roth, 99.8%), n-C6H14 (Sigma-Aldrich, 499%), KOH (Grüssing, 85%), KIO3 (Merck, 499.7%), NaOH (Sigma-Aldrich, 499%), NaI (Merck, 499.5%), HCl (SigmaAldrich, 36%), H2SO4 (Sigma-Aldrich, 95–98%), and soybean oil (local grocery store).

3. Results and discussion 3.1. Micromixing efficiency The effectivity of micromixing for each mixer was determined. Therefore the modified Villermaux–Dushman-method proposed by Panić et al. (2004) was used. This method allows to qualitatively rank different mixers and operation conditions (Bourne, 2008). The method was successfully devoted to characterize micromixers with channel sizes from 40–1200 mm at flow rates from 0.1 to 40 ml min  1 (Panić et al., 2004). A high efficiency of micromixing may indicate a high interfacial area during biphasic reactions and therefore a good reactor performance for the ethanolysis of soybean oil. The procedure is based on the mixing sensitive conversion of two competing reactions. Two aqueous solutions are brought into the mixing device. The first solution contains a borate buffer, KI and KIO3. The second one contains diluted sulfuric acid. When these reactants are brought together the reactions in Eqs. (7)–(9) can occur: NaH2 BO3 þ H 2 SO4 ⇄H3 BO3 þ NaHSO4

ð7Þ

5KI þ KIO3 þ 3H2 SO4 ⇄3I2 þ 3H2 O þ 3K2 SO4

ð8Þ

I2 þ KI⇄KI3

ð9Þ

In this reaction system the neutralisation in Eq. (7) is much faster than the redox reaction in Eq. (8). If the micromixing is perfect the sulfuric acid is consumed by the buffer completely. Otherwise iodine is formed which leads in addition to the

613

formation of pottasiumtriiodide (Eq. (9)). This compound shows absorption at a wavelength of 286 nm. A high absorption of triiodide indicates insufficient micromixing. Fig. 2 shows the volume flow dependent absorption of pottasiumtriiodide for the three mixers. Micromixer LTF-MX shows the lowest absorption and therefore the best micromixing performance. Due the mixing principle of this device incoming streams are splitted into two different layers which cross each other periodically. This leads to a lot of small mixing areas and therefore to an efficient micromixing. For increasing flow rates the T-mixer and LTF-MX show increasing micromixing efficiency. This effect derives from changes in the fluid dynamics (Panić et al., 2004). Due stronger turbulences in the mixing region micromixing is improved. The micromixing efficiency of micromixer LTF-MS has local maximum at 0.7 ml min  1 and keeps increasing at flow rates above 1.5 ml min  1. This effect is known for other mixing devices as well but cannot be explained yet (Panić et al., 2004). At low flow rates below 1.0 ml min  1 the micromixing in mixer LTF-MS is better than in the T-mixer and vice versa at flows above 1.0 ml min  1. The improved mixing performance probably derived from the structuring of mixer LTF-MS. The effect of the geometry seems to become minor compared to the hydrodynamic influences at higher flows. 3.2. Ethanolysis experiments The results of the Villermaux–Dushman-method characterize the effectivity of micromixing in monophasic systems. The performance of the mixers at the ethanolysis is evaluated to see if the micromixing efficiency influences the outcome of the biphasic reaction. Panić et al. (2004) suggested a dye extraction method to determine the micromixing efficiency in biphasic mixtures. Since an analogous treatment of the ethanol/oil system seemed impractical it was decided to compare the yield of FAEE for different mixers at the same flow rate. The experiments MS-V, MX-V and T-V (see Table 2) were carried out in reactors with a constant volume. By variation of the total flow the reaction time was varied. The achieved weight fractions of FAEE at a reaction time of 8 min from 30–60 1C are shown in Fig. 3. Over the whole temperature range the highest yields could be observed with the mixer LTF-MX. The T-mixer showed the worst performance. The total flow at a reaction time of 8 min was about 0.3 ml min  1. As shown in Fig. 2 the ranking of the micromixing performance is the same as the ranking of the reached yields in the ethanolysis reaction. These results show that the obtained yield of FAEE correlates with the micromixing efficiency of the used mixing device. The results are comparable to investigations of the base catalyzed methanolysis of

Fig. 2. Effectivity of micromixing according to the Villermaux–Dushman-method.

614

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

Fig. 4. Yield of FAEE after 6 min reaction time for various flow rates. Fig. 3. Yield of FAEE after 8 min reaction time.

sunflower oil in a batch reactor performed by Stamenković et al. (2007). They observed the formation of a MeOH–oil-emulsion at the beginning of the reaction. The size of the droplets formed depended on the agitation speed of the mixer. Better mixing of the biphasic mixture led to an increase of the interfacial are. The results obtained by us demonstrated that the continuous base catalyzed ethanolysis is accelerated by an increase in mixing performance and therefore is strongly influenced by mass transfer phenomena. The investigations based on the Villermaux–Dushman-method pointed out that the effectivity of micromixing strongly depends on the total flow rate. Therefore the effect of the total flow on the yield of FAEE was investigated by using the reactor setups MS-R, MX-R and T-R. The total volume of the reactors was varied by using PTFE capillaries of various lengths and the residence time was kept constant by simultaneously varying volume and flow rate. The flow rate dependent yields observed at a reaction time of 6 min and a temperature of 30 1C is shown in Fig. 4. At lower flow rates below 1.2 ml min  1 the FAEE yield increased with increasing flow for the micromixer LTF-MS and the T-mixer while the yield reached with micromixer LTF-MX stayed constant. In this micromixer the efficiency of micromixing correlates with the reached yields like discussed above. At higher flow rates the yield observed for micromixer LTF-MX was the lowest while the performance of LTF-MS and the T-mixer kept rising. At a total flow of 1.5 ml min  1 the yield using LTF-MS was doubled to 60% from 30% at 0.2 ml min  1. At even higher flow rates of 2.3 ml min  1 nearly total conversion could be reached with the T-mixer. The different volume flow dependency of the mixing performance in the ethanolysis reaction derives from the different mixing principles. High flows lead to higher turbulences in the LTF-MS and T-mixer and therefore an increasing surface area. In micromixer LTF-MX the total energy of the incoming flow was splitted between the crossing points and therefore fluid dynamic phenomena showed a lesser effect on the mixing performance of the device. To analyze the effect of the inner diameter and geometry of the tube reactor on the yield of FAEE the influences of flow rate and reaction time had to be eliminated. Therefore the results of the experiments MS-V, MX-V and T-V at a residence time of 6 min rate were compared to the results obtained by the experiments MS-R, MX-R and T-R at a flow rate of 0.36 ml min  1. These experimental points are characterized by the same reactor volume, flow rate and residence time. Differences in the FAEE yield should be caused by differences of the used tube reactors. The observed yields for both experiments at a temperature of 30 1C are shown in Fig. 5. For every mixer a higher yield could be achieved using the tube

Fig. 5. Yield of FAEE depending on the inner diameter of the tube reactor.

reactor with an inner diameter of 1.0 mm. The increase of the yield is higher if the mixer shows lower micromixing efficiency. An increase of the yield of esters at the base catalyzed methanolysis of rapeseed oil by using tube reactors with small diameters was reported by Sun et al. (2008). They observed the formation of a slug-flow during the reaction. A decrease of the tube diameter led to smaller slugs and therefore a higher interfacial area. A similar effect could influence the yield of the base catalyzed ethanolysis. To determine the influence of the tube geometry on the performance of the ethanolysis reaction experiment T-VS was carried out at 30 1C and 40 1C. The structure of the tube reactor LTF-VS was similar to the structure of the micromixer LTF-MS. It was shown, that this geometry leads to an increase of reactor performance compared to the T-mixer. It was assumed that combining LTF-VS with the T-mixer leads to higher FAEE yields than the combination of T-mixer and LTF-V. The observed yields after 8 min reaction time are shown in Fig. 6. For both temperatures a slightly lower yield was reached with the tube reactor LTF-VS, but the differences to the tube LTF-V were very small. It appears that the reaction was not accelerated by the geometry of the tube after the mixture passed the T-mixer. The mixing at the very beginning seems to influence heavily the outcome of the continuous ethanolysis reaction. Santacesaria et al. (2012a) mentioned a similar phenomena at the methanolysis of vegetable oil. The initial interfacial area strongly effected the final conversion. The observation of this effect for the ethanolysis in this work shows once more the mass transfer dependency of this reaction.

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

Fig. 6. Yield of FAEE after 8 min reaction time for T-V and T-VS.

615

Fig. 8. Yield of FAEE after a reaction time of 8 min for MX-V and the batch experiments.

published by other authors mostly neglected mass transfer limitations and evaluated the chemical kinetics with a simplified reaction scheme shown in Eq. (10) (Marjanović et al., 2010; Shahla et al., 2012). T þ 3EtOH⇄G þ 3E

Fig. 7. Yield of FAEE in the batch processes.

To compare the investigated continuous processes with the state of the art industrial methods the transesterification was performed discontinuous under magnetically stirring. The yields reached with the batch process at different temperatures are shown in Fig. 7. After short reaction times a constant amount of FAEE was formed. The yield strongly depended on the temperature. Marjanović et al. (2010) reported that the amount of FAEE was temperature independent after 20–30 min at comparable conditions (Marjanović et al., 2010). Fig. 8 shows the yield of FAEE reached after a reaction time of 8 min for the experiments MX-V and the batch process. While at 30 1C the yield at this reaction time is higher in the batch process starting at 40 1C results of both reactors are comparable. Taking into account that a very low flow rate was applied to the continuous reactor (0.28 ml min  1) the results show a very effective mixing in the micromixer LTF-MX. The results with the batch reactor were achieved at an agitation speed of 1000 rpm. The results in Fig. 4 demonstrate the importance of the total flow for the performance of the continuous process. With the T-mixer and a flow rate of 2.3 ml min  1 complete conversion was observed after 6 min reaction time and at 30 1C. However, with the batch process even at 60 1C this conversion was not reached after 12 min (see Fig. 7). 3.3. Kinetic modeling All results obtained so far show the heavy influence of mass transfer limitations of the ethanolysis of soybean oil. Kinetic studies

ð10Þ

Richard et al. (2013) published a kinetic model for the KOH catalyzed ethanolysis of sunflower oil taking into account mass transfer limitations. The chemical kinetics were described with three consecutive reversible reactions like shown in Eqs. (1)–(3) while mass transfer was considered with a two film model. The simplicity of the model and the well agreement between experimental data and optimization observed by Richard et al. (2013) made this approach seem applicable for modeling our experimental data. The model was modified considering the differences in our general setup. As explained in the experimental section the analytics did not allow to distinguish between di- and monoacylglycerol. A pseudo-product called side product (SB) was introduced as the sum of D and M. The ratio of D to M was represented by the factor x shown in Eq. (11). x ¼ ½D=ð½D þ ½MÞ

ð11Þ

The resulting reaction system for the ethanolysis is represented in Eqs. (12)-(14). By addition of Eqs. (13) and (14) a mechanism with 2 reversible consecutive reactions shown in Eqs. (15) and (16) was introduced representing a compromise between the strongly simplified expression given in Eq. (10) and the more accurate description in Eqs. (12)–(14) T þ EtOH⇄xSB þE

ð12Þ

xSB þ EtOH⇄ð1  xÞSB þ E

ð13Þ

ð1  xÞSB þEtOH⇄G þ E

ð14Þ

k1

T þ EtOH ⇄ xSB þ E

ð15Þ

k1

k2

xSB þ 2EtOH ⇄ G þ 2E

ð16Þ

k2

In analogy to the studies referenced (Marjanović et al., 2010; Shahla et al., 2012) the kinetics of reaction 16 were expressed in first order according to EtOH, SB, E and G. In the mentioned papers this assumption was made for the overall reaction given in Eq. (10) with good agreement between kinetic model and measured data. Since there was no way to distinguish between mono- and

616

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

diacylglycerole this assumption seemed reasonable to describe the second part of the reaction. The following hypotheses were formulated based on the assumptions of Richard et al. (2013):

fraction of each component was obtained by Eq. (32).

 All reactions are first order according to all components of the proposed equations.  The catalytic species is EtO  which forms by total consumption of the initial added KOH.  Forth and back reactions are catalyzed by EtO  and first order according to the concentration of catalyst.  The overall reaction is biphasic. A polar ethanol phase (index p) and an apolar oil phase (index a) are formed.  The reaction occurs in the polar phase. There are phase transitions for T, SB and E.  A chemical equilibrium is present at the end of the reaction.

For the determination of the model parameters the quasiNewton-method was used. As optimization criteria the sum of squared residuals (SSR) of calculated and observed weight fractions was used (Eq. (33a)). At low weight fractions below 1% the absolute error (Eq. (33b)) was calculated instead of the relative error for reasons of numerically stability.

Introducing the phase transition of E is an extension of the original model to make the back reaction in Eq. (16) possible. These assumptions lead to the reaction rates formulated in Eqs. (17)–(20). The mass transfer was described by a two film model shown in Eq. (21). In this case A is an apolar component T, SB or E. r 1 ¼ k1 ½Tp ½EtOH½EtO  

ð17Þ

r  1 ¼ k  1  ½SBp ½Ep ½EtO   ¼ k  1 ½SBp ½Ep ½EtO  

ð18Þ

r 2 ¼ k2  ½SBp ½EtOH½EtO  

ð19Þ

r  2 ¼ k  2 ½G½Ep ½EtO   ¼ k  2 ½G½Ep ½EtO  

ð20Þ

n

n

J PA ¼ kL;A aA ð½Aa  ½Ap Þ ¼ β A ð½Aa –½Ap Þ

ð21Þ p

a

In Eqs. (17)–(20) the reactants without indices or are only present in one phase. All reactions occur in the polar ethanolic phase. There are mass transfers defined for T, SB and E in Eq. (21). The other components EtOH, EtO  and G are only present in the polar phase. The given reaction rates are all third order. This is due the hypotheses adopted from Richard et al. (2013) mentioned above. The resulting reaction system with nine components was represented by 9 differential equations (Eqs. (22)–(30)). The initial conditions for t0 ¼0 s were stated as [T]a0 ¼1.057 mol l  1; [EtOH]0 ¼16.72 mol l  1; [EtO  ]0 ¼ 0.442 mol l  1; other concentrations were set to 0 mol l  1. The initial concentrations derived from the reaction conditions described in the experimental section. d½Tp =dt ¼ J pT  r 1 þ r  1

ð22Þ

d½Ta =dt ¼  J pT

ð23Þ

d½SBp =dt ¼ J pSB þ r 1  r  1  r 2 þ r  2

ð24Þ

d½SBa =dt ¼  J pSB

ð25Þ

d½G=dt ¼ r 2  r  2

ð26Þ

d½EtOH=dt ¼ r 1 þ r  1 2r 2 þ 2r  2

ð27Þ

d½Ep =dt ¼ J pE þ r 1  r  1 þ 2r 2  2r  2

ð28Þ

a

d½E =dt ¼

 J pE

½EtO   ¼ ½EtO  0 ½EtOH0 =½EtOH

ð29Þ ð30Þ

Kinetic modeling was performed with Scilab ver. 0.5.4.0 (Linux x84). The analysis by HPLC gave weight fractions of T, SB and E but not the distribution in the two phases. To obtain the total amount of apolar products formed according to the model the mass concentration of the apolar products was calculated with Eq. (31). The weight

ma =V ¼ ð½Ta þ ½Tp ÞMT þ ð½SBa þ ½SBp ÞMSB þ ð½Ea þ ½Ep ÞME

ð31Þ

wA ¼ ½AMA =ðma =VÞ

ð32Þ

SSR ¼ Σ i ððwacalc;i þwpcalc;i  wobs;i Þ=wobs;i Þ2 for wobs;i 4 0:01

ð33aÞ

SSR ¼ Σ i ðwacalc;i þ wpcalc;i  wobs;i Þ2 for wobs;i o 0:01

ð33bÞ

The result for three continuous mixers at 30 1C is a highly reversible second consecutive reaction leading to an unreasonable high amount of side products. Aside from this effect, the overall trends at 30 1C stay similar to the previously applied fitting method. For all continuous reactors the first consecutive reaction is highly reversible in contrast to the batch process. Furthermore mass transfer coefficient β1 has nearly the same values like at the separate optimization runs. β2 And β3 may be biased by the low values of k2 and k  2. Regarding these results we assume that in between the limitations of the chosen mathematical model, the observable changes in the kinetics caused by the use of different reactors result in changing reaction constants. This effect is not strongly compensated by the mass transfer coefficients. The obtained results show the influence of these unconsidered mass transfer effects on the observable reaction rate. This purely qualitatively description reveals fundamental differences between the kinetics in a continuous microreactor and the corresponding batch process. The obtained mass transfer coefficients β depended strongly on the interfacial surface a. At the batch process the surface depends on the agitation speed in the vessel (Stamenković et al., 2007). Since it was constant for all experiments β can be assumed constant. For continuous runs the value of β depends on the total flow rate (Santacesaria et al., 2012a). For the T-mixer and the micromixer LTF-MS a raise of the FAEE yield with higher flow rate was observed (Fig. 4) which indicates that the mass transfer was not constant for different flows. Micromixer LTF-MX on the other hand showed a nearly constant performance at lower flow rates. Here β was assumed to be constant. The flow rate dependence of the mass transfer coefficient for the mixers LTF-MS and T-mixer was determined by modeling the experimental data shown in Fig. 4. Therefore the rate constants k1, k  1, k2 and k  2 obtained by the kinetic modeling of the batch experiments were used to describe the chemical reaction. Two different models for the flow rate dependence of β were used. The linear approach (Eq. (34)) was proposed by Santacesaria et al. (2012a) in a kinetic model for the continuous base catalyzed methanolysis of soybean oil. We showed in Fig. 2 that the effectivity of micromixing increases with the flow rate. Our results for the T-mixer and the work of Panić et al. (2004) showed a correlation which seems to be more likely exponential than linear. Therefore we tested an exponential approach as well (Eq. (35)). _ ¼ β ðV_ ¼ 0Þ
V= _ V_ βðVÞ ref

ð34Þ

_ ¼ β ðV_ ¼ 0Þ
EXPfV= _ V_ g βðVÞ ref

ð35Þ

The results of the optimization of the mass transfer coefficients for the micromixer LTF-MS and the T-mixer are shown in Table 3. The exponential approach gave the best fit and was therefore used for modeling the experimental date obtained with these mixers. Tables 4 and 5 sum up the results of the kinetic modeling of the experiments MS-V, MX-V, T-V and the batch experiments at 30 1C

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

617

Table 3 Modeling of the flow rate dependent mass transfer coefficients. Mixer

β1

β2

β3

β From

0.184 1.158 0.184 0.195

0.0002 0 0.0003 0.05

Eq. Eq. Eq. Eq.

[dm min  1] LTF-MS LTF-MS T-mixer T-mixer

0.093 0.051 0.069 0.035

(34) (35) (34) (35)

Table 4 List of parameters obtained by modeling the experimental data at 30 1C. Rate constants k are given in [l3 mol  3 min], mass transfer coefficients β in [dm min  1], equilibrium constants K and SSR. Parameter

MS-V

MX-V

T-V

Batch

k1 k1 k2 k2 β1 β2 β3 K1 K2 SSR

1.544 0.146 0.191 0.436 0.049 0.146 0.069 10.57 0.44 0.222

1.1 0 0.022 0.001 0.363 0.25 0 Irreversible 22 0.095

4.016 0.144 0.09 0.159 0.048 0.156 0.084 27.89 0.57 0.601

5.626 3.231 0.272 0.677 3.572 0.222 0.154 1.74 0.4 1.183

Fig. 9. Rate constants of the ethanolysis reaction at 30 1C.

Table 5 List of parameters obtained by modeling the experimental data at 60 1C. Rate constants k are given in [l3 mol  3 min], mass transfer coefficients β in [dm min  1], equilibrium constants K and SSR. Parameter

MS-V

MX-V

T-V

Batch

k1 k1 k2 k2 β1 β2 β3 K1 K2 SSR

0.925 0.060 0.228 0.260 0.079 0.144 0.062 15.42 0.87 0.443

0.5372 0.396 0.176 0.117 1.011 0.101 0.086 1.36 1.50 0.077

2.518 0.039 0.174 0.244 0.064 0.141 0.066 64.56 0.71 0.371

7.904 5.677 6.282 3.178 6.941 3.344 0.185 1.39 1.98 0.327

and 60 1C, respectively. The optimization at 40 1C and 50 1C was carried out as well and gave results with comparable agreement to the experimental data. The values of the rate constants at 30 1C are shown in Fig. 9 and Table 4. The constant k1 was the highest one for every reactor, the conversion of triacylglycerol to side products and FAEE occurred very fast. Experiment MX-V indicated the lowest rate constant k1. This was unexpected because this microreactor showed the best performance in the ethanolysis reaction out of the three considered continuous experiments. The batch process indicated the highest rate constants for every reaction. It appeared that the second reaction characterized by k2 and k  2 was much slower than the first one. To give a better overview about the influence of the equilibrium on the results of the ethanolysis the temperature dependent equilibrium constants K1 ¼k1/k  1 and K2 ¼ k2/k  2 are shown in Tables 4 and 5. The first equilibrium characterized by K1 was for all continuous experiments shifted strongly to the side of the products while for the batch process K1 was for all temperatures close to 1. At 30 1C and 40 1C the reaction was irreversible for MX-V. With increasing temperature K1 of batch process and MX-V got closer and were nearly the same at 60 1C.

Fig. 10. Results of the kinetic modeling (full lines) for the experiments (points) with batch reactor (top) and microreactor MX-V (bottom) at 30 1C. Dashed lines represent the weight fractions of 0.8 and 0.2.

The second equilibrium constant K2 behaves much different. For MS-V and T-V the value of K2 was below 1 independent of the reaction temperature. The reaction was more irreversible at 30 1C when the micromixer LTF-MX was used. The experiments MX-V showed a shift of the equilibrium towards the side of the reactants with increasing temperature while the equilibrium of the batch process got shifted towards the product side at higher temperatures. The product favouring state of the equilibrium constant K2 seems to be the reason for the

618

S. Schwarz et al. / Chemical Engineering Science 104 (2013) 610–618

products more than the continuous process a slightly higher yield was observed for the batch reaction. 4. Conclusions The continuous base catalyzed ethanolysis of soybean oil was carried out in seven different reactors. The modified Villermaux– Dushman-method was successfully used to rank three mixing devices by their effectivity of micromixing. It was shown that the yield of fatty acid ethyl esters depends strongly on mass transfer limitations. The mass transfer increased with growing efficiency of micromixing, low inner diameters of the applied tube reactors and the application of high flow rates. It was shown that with continuous reactors higher yields are obtained than with a batch reactor at lower temperatures and shorter reaction times under otherwise identically conditions. The experimental data was modeled by modifying a kinetic model published by Richard et al. (2013). Kinetic modeling showed that a shift of the underlying chemical equilibriums appears when continuous reactors are used. This effect disappears with increasing temperature. References

Fig. 11. Results of the kinetic modeling (full lines) for the experiments (points) with batch reactor (top) and microreactor MX-V (bottom) at 60 1C. Dashed lines represent the weight fractions of 0.9 and 0.1.

better performance of the reactor system MX-V compared to MS-V and T-V. An equilibrium constant for the overall process would require neglecting the formation of the side products resulting in an equilibrium state were only T, EtOH, E and G are present. Since in nearly all experiments a significant amount of side products was formed an overall equilibrium cannot be formulated. The results of the kinetic modeling showed indeed a shift of the equilibrium state for both consecutive steps depending on the used reactor. This may not be interpreted as an influence on the chemistry involved. Moreover this result shows a heavy impact of the reactor type on the mass transfer and micromixing effects. This leads to a shift of the rate constants because the applied kinetic model does not describes these effects properly. Figs. 10 and 11 show the results of the kinetic modeling for the batch experiments and MX-V at 30 1C and 60 1C, respectively. At both temperatures the conversion of triacylglycerol is much slower in the micromixer LTF-MX. However at 30 1C the amount of side products produced with the continuous reactor is half of the amount formed in the batch process (Fig. 10). Since the reaction of the side products to FAEE is much slower than the primary reaction a higher amount of FAEE is formed after reaction times about 12 min using the micromixer. The change of the kinetics by using microreactor LTF-MX leads to higher yields compared to the batch process at otherwise identically reaction conditions. At 60 1C (Fig. 11) for both reactors the equilibrium constants are nearly equal (comp. Table 5). Since the equilibrium of the second reaction of the batch process is favouring the

Bertoldi, C., da Silva, C., Bernardon, J., Corazza, M., Filho, L., Oliveira, J., Corazza, F., 2009. Continuous production of biodiesel from soybean oil in supercritical ethanol and carbon dioxide as cosolvent. Energy Fuels 23, 5165–5172. Boocock, D.G.B., Konar, S.K., Sidi, H., 1996. Phase diagrams for oil/methanol/ether mixtures. Journal of the American Oil Chemists' Society 73, 1247–1251. Bourne, J.R., 2008. Comments on the iodide/iodate method for characterising micromixing. Chemical Engineering Journal 140, 638–641. Gandhi, A.P., Joshi, K.C., Jha, K., Parihar, V.S., Srivastav, D.C., Raghunadh, P., Kawalkar, J., Jain, S.K., Tripathi, R.N., 2003. Studies on alternative solvents for the extraction of oil-I soybean. International Journal of Food Science and Technology 38, 369–375. Guan, G., Kusakabe, K., Moriyama, K., Sakurai, N., 2008. Continuous production of biodiesel using a microtube reactor. Chemical Engineering Transactions 14, 237–244. König, B., 2010. Mikroreaktionstechnik in der organischen Synthese. In: Umwelt, D.B. (Ed.). Kulkarni, M.G., Dalai, A.K., Bakhshi, N.N., 2007. Transesterification of canola oil in mixed methanol/ethanol system and use of esters as lubricity additive. Bioresource Technology 98, 2027–2033. Marjanović, A.V., Stamenković, O.S., Todorović, Z.B., Lazić, M.L., Veljković, V.B., 2010. Kinetics of the base-catalyzed sunflower oil ethanolysis. Fuel 89, 665–671. Meher, L., Vidyasagar, D., Naik, S., 2006. Technical aspects of biodiesel production by transesterification—a review. Renewable and Sustainable Energy Reviews 10, 248–268. Moser, B.R., 2009. Biodiesel production, properties, and feedstocks. In Vitro Cellular and Developmental Biology – Plant 45, 229–266. Panić, S., Loebbecke, S., Tuercke, T., Antes, J., Bošković, D., 2004. Experimental approaches to a better understanding of mixing performance of microfluidic devices. Chemical Engineering Journal 101, 409–419. Qiu, Z., Zhao, L., Weatherley, L., 2010. Process intensification technologies in continuous biodiesel production. Chemical Engineering and Processing 49, 323–330. Richard, R., Thiebaud-Roux, S., Prat, L., 2013. Modelling the kinetics of transesterification reaction of sunflower oil with ethanol in microreactors. Chemical Engineering Science 87, 258–269. Santacesaria, E., Turco, R., Tortorelli, M., Russo, V., Di Serio, M., Tesser, R., 2012a. Biodiesel process intensification by using static mixers tubular reactors. Industrial and Engineering Chemistry Research 51, 8777–8787. Santacesaria, E., Vicente, G.M., Di Serio, M., Tesser, R., 2012b. Main technologies in biodiesel production: state of the art and future challenges. Catalysis Today 195, 2–13. Shahla, S., Ngoh, G.C., Yusoff, R., 2012. The evaluation of various kinetic models for base-catalyzed ethanolysis of palm oil. Bioresource Technology 104, 1–5. Stamenković, O.S., Lazić, M.L., Todorović, Z.B., Veljković, V.B., Skala, D.U., 2007. The effect of agitation intensity on alkali-catalyzed methanolysis of sunflower oil. Bioresource Technology 98, 2688–2699. Stamenković, O.S., Veličković, A.V., Veljković, V.B., 2011. The production of biodiesel from vegetable oils by ethanolysis: current state and perspectives. Fuel 90, 3141–3155. Sun, J., Ju, J., Ji, L., Zhang, L., Xu, N., 2008. Synthesis of biodiesel in capillary microreactors. Industrial and Engineering Chemistry Research 47, 1398–1403. Wen, Z., Yu, X., Tu, S.T., Yan, J., Dahlquist, E., 2009. Intensification of biodiesel synthesis using zigzag micro-channel reactors. Bioresource Technology 100, 3054–3060.