Design of catalytic micro trickle bed reactors

2.1 Flow regimes

Downward flow packed bed reactors, namely trickle bed reactors (TBR), allow for a variety of flow regimes. In general, four distinct flow regimes exist in TBRs:

1. Trickle flow regime: this regime exists when both gas and liquid flow rates are relatively low. In this regime, gas phase is continuous, and the liquid phase is dispersed. It is also known as a low interaction regime, since the flow in one phase does not significantly affect the flow in the other phase;

2. Pulsed flow: a higher gas flow rate results in pulsed flow, where the interaction between the phases is also higher;

3. Spray flow: for a given liquid flow rate, if the gas flow rate is relatively increased too much, spray flow will be obtained;

4. Bubble flow: in contrast to the spray flow, if the liquid throughput is relatively high in comparison to the gas flow, the liquid phase is continuous and the gas phase is dispersed; this is the bubble flow.

Schematic representation of bubble flow, pulsing flow, trickling regime and spray flow are given in Fig. 1 [6]. The knowledge of transition between flow regimes is critical to design TBRs. A flow map for trickle bed reactors for foaming and non-foaming liquids was developed by Charpentier and Favier [10] and reported in various text books [11, 12].

Various hydrodynamic correlations are summarized in extensive reviews of two-phase flow systems in packed beds [1, 13], and the role of hydrodynamics on conventional trickle bed reactors was covered in extensive reviews [14–18]. The transition from trickle to pulse flow is generally characterized by a sharp increase in the root mean square pressure fluctuation for a small increase in gas or liquid flow rate. At low gas velocities (Re_G < 400), the flow regimes change consecutively from trickle flow to pulse flow and from pulse flow to bubble flow by increasing the liquid flow rate. For high gas flow rates, the transition is in the order of wavy, spray, pulse and bubble flow by increasing the liquid flow rate. Many correlations and models have been proposed in recent years to predict the regime boundaries. However, none of them has been entirely successful. Larachi et al. [19] systematically analyzed the prediction performances of all the transition models and correlations against all the transition data published in the literature since 1964. It was seen that the use of available phenomenological and semi-theoretical models for predicting flow transition leads to unacceptable errors. Based on all this information, Larachi et al. [19] recommended the use of a neural network correlation that turned out to be the most general correlation for predicting the trickle-to-pulse flow transition.

Fig. 1

Schematic representations of flow regimes in trickle-bed reactors. Reprinted with permission from “Ind Eng Chem Res 2006;45(15):5185–5198” [6].

© 2006, American Chemical Society.

2.1.1 Micro trickle bed reactors

The trickle flow regime prevails at relatively low gas and liquid flow rates. The liquid flows as a laminar film and/or in rivulets over the catalyst particles, while gas passes through the remaining void space. Trickle flow regime exists at low liquid and moderate gas flow rates where liquid flows as film over catalyst particles. In this regime, inertial forces are weaker compared to the interfacial forces, and liquid hold-up is controlled by capillary pressure. The trickle flow regime region widens with a decrease in liquid viscosity and/or surface tension. Heat and mass transfer rates are slower when compared to other possible flow regimes.

Transition from a trickle to pulse flow regime occurs with an increase in either the gas or liquid flow rate. In the pulse flow regime, local flow paths for gas are blocked by liquid pockets, which results in the formation of alternate gas and liquid-enriched zones. A pulse flow regime has advantages in terms of the utilization of a catalyst bed and higher heat and mass transfer rates. However, the operating window of this regime is relatively small and decreases as the particle diameter increases. The concept of flow passage blockage at large liquid holdups is reported by several authors [20, 21]. This blockage generates disturbances that propagate and grow with the length of the reactor. In other studies, the appearance of a pulse flow regime is thought to be related to the instability that occurs in the liquid film due to the shear exerted by the gas phase [22, 23].

Several experimental methods were used to detect the transition line from a trickle to pulse flow regime. A change in slope of measured pressure drop versus gas or liquid flow rate indicates the transition to the pulse flow regime [24]. Another group of methods is based on the application of different on-line transducers and imaging techniques.

Gas and liquid properties have significant effects on the transition boundary. Cyclohexane has a three times lower surface tension than water. As a result, a transition to a pulse flow regime occurs at lower liquid flow rates (Fig. 2) [25]. The influence of gas viscosity and density on the transition boundary is relatively small when compared to that of liquid.

Fig. 2

Effect of gas and liquid throughputs on trickle to pulse flow regime transition boundary. Reprinted from “Chem. Eng. Sci., volume 55, Attou A, Ferschneider G., a two-fluid hydrodynamic model for the transition between trickle and pulse flow in a cocurrent gas-liquid packed-bed reactor, 491–511” [25].

© 2000, with permission from Elsevier.

Higher liquid viscosity leads to an increase in liquid holdup and therefore earlier transition to the pulse flow regime.

Faridkhou et al. [26] employed digital image analysis to characterize the two flow regimes, hysteresis, and transition thereof. The effect of gas density on the onset of flow regime transition was studied by comparing the curves for air-water and argon-water systems. In MTBRs, the low-to-high interaction flow regime transition at a given liquid superficial velocity shifts toward higher gas superficial velocities the higher the pressures (or the gas densities) are. This makes the trickle flow operating region wider at elevated pressures or with gases with higher molecular weights. As stated by Attou et al. [25], an increase in gas density leads to a decrease in inertial forces and therefore transition occurs at higher gas velocities.

Faridkhou et al. [27] developed a method to study RTD in a MTBR. They used pellets of two different sizes: 53–63 μm and 106–125 μm and co-current operation. Due to relatively high pressure gradients of 500–3000 kPa/m, they carefully densified the bed prior to the experiments. The gas and liquid phases were brought into contact by a T-junction located upstream of the packed bed consisting of a coaxial arrangement. The liquid flows through the inner tube while the gas flows within the annular area between liquid line and the outer tube. To avoid flow instabilities within the bed, it is of importance that the contact between the two phases takes place at the start of the packed bed. They used two point electrodes inserted into the reactor wall via two holes. The authors experimentally confirmed theoretically predicted maximum liquid velocity in the high porosity zone close to the wall. The solid holdup was 0.592 and 0.576 for the 53–63 μm and 106–125 μm pellet beds, respectively. The liquid holdup was in the range 0.22–0.25 for the 53–63 μm pellet bed and this value was by 10–15% higher than that for the 106–125 μm pellet bed at the same gas and liquid velocity.

2.1.2 Semi-structured micro trickle bed reactors

Since more than a decade, various micro and milli reactor concepts have been developed for gas-liquid-solid reactions. Many of these reactors are becoming a key technology for the industry to face market pressures and match increased environmental considerations. Due to their high mass and heat transfer performances and the intrinsically safe design inherent to microstructures, they allow isothermal operation under otherwise explosive conditions. Furthermore, the small material inventory reduces the costs optimization of catalyst and/or operating conditions that often decrease time to the market. These reactors have been applied in kinetic studies of highly exothermic fast reactions, catalyst screening and process optimization.

Structured packings have the advantage that they are made to fit the dimensions of the reactor in which they are placed and thus avoid the flow maldistribution due to a channelling or bypassing of the solids. Solid foam packings represent a generation of materials combining relatively high specific surface area with low pressure drop per unit height. This is largely due to the open-celled structure with very high voidages (up to 97 vol %). The geometric surface areas of the solid foam packings increase as the voidage decreases (solids holdup increasing) because the struts making up the unit cells increase in diameter. The ppi number (pore per linear inch) of the solid foam packings is an independent parameter to describe the average cell size.

These packing materials commonly have a two-dimensional structure that redirects the liquid and gas flow in planar directions. Several examples of structured packing materials are shown in Fig. 3. In monoliths (including internally finned monoliths), the liquid and gas flows in separated channels, while in Mellapak, the fluids flow down corrugated sheets of gauze stacked to form open channels between these sheets, and in Katapak some of the open channels are filled with spherical particles. In the development of these packings, the aim was to increase their relatively low surface area, while maintaining a low pressure drop (high voidage) and adequate contact between the flowing phases. Solid foams have been produced in different materials (metal, ceramics, carbon, SiC, polymers, etc.). Banhart et al. [28] outlined the methods and procedures for producing these and many other solid foams. In the particular field of chemical engineering, ceramic foams found application as heat exchangers, solar receivers, gas filters, packing columns or catalyst support. A review of this field is proposed by Twigg et al. [29].

Fig. 3

Schematic representation of commonly available structured reactor packings. (a) Monolith, (b) internally finned monolith, (c) Mellapak (Sulzer), (d) Katapak-S (Sulzer).

The surface area of random or structured packings can be enlarged by depositing catalytic coatings. However, due to most of the area being internal, it is not hydrodynamically accessible, and diffusion limitations within the pores may still affect the transfer of components to the active catalyst. This mass transfer of components to the catalyst located on the solid support is essential for the operation of multiphase catalytic reactions. A clear understanding of the corresponding mass transfer resistances is vital.

2.1.3 Structured micro trickle bed reactors

A novel MTBR was developed where arrays of posts were built in each channel imitating the packing in conventional reactors [30] (Fig. 4). These posts in the reactor channels provide a bed porosity of around 0.4 and enhance the overall heat and mass transfer analogous to packing in conventional reactors.

Fig. 4

Micro-structured catalyst supports prior to porous silicon formation. (a) Channel view. (b) Micron scale striations produced by the etch process. Reprinted with permission from “J. Microelectromechanical Syst. 2002;11(6): 709–17” [30].

© 2002, IEEE.

In a follow up study, Wada et al. reported a multichannel microreactor in which the channels comprise uniformly distributed pillars [31]. The multi-channel microreactor was built as an integrated system with a pressure drop section at the inlet (Fig. 5). Flow regimes were studied by a reaction between oxygen and ethyl acetate (Fig. 6). At low liquid to gas flow rates and all gas flow rates, an annular flow was observed. At high liquid to gas flow ratios, slug flow existed. At high gas and liquid flow rates, churn flow was observed with rapidly undulating gas-liquid interface shapes.

Fig. 5

Multichannel microreactor and design of the pressure drop zone. (a) Relationship between structure and normalized pressure drop. The depth was measured from the surface of the Si wafer. The depths of manifold and reaction channels were 300 μm. The pressure drop across the shallow channels (25 μm deep) dominates the total pressure drop. (b) Fabricated structure. (c) Picture of microfabricated multichannel microreactor. Reprinted with permission from “Ind. Eng. Chem. Res. 2006;45(24):8036–42” [31].

© 2006, American Chemical Society.

Fig. 6

Gas-liquid flow regimes observed in the multichannel microreactor with posts: (a) slug flow (gas as dark area); (b) churn flow (the interface fluctuates rapidly and liquid periodically spans the entire channel); (c) annular flow (gas flows at the center of reactor). The conditions used in the oxidation experiments fall within the rectangle in dashed lines. Reprinted with permission from “Ind. Eng. Chem. Res. 2006;45(24):8036–42” [31].

© 2006, American Chemical Society.

In a follow up study, an integrated MTBR with cooling was developed [32]. In order to achieve an overall device capacity of 1 ml min⁻¹, the reactor consisted of 32 parallel channels with a width of 650 μm, a depth of 300 μm that were packed with 5000 posts with a diameter of 70 μm. The 8 : 1 width-to-packing diameter ratio provided uniform flow distribution. The authors observed significant differences in transition lines between different flow regimes as compared with the flow regime maps for open channels, such as the Baker coordinate map [33], Charpentier-Favier coordinate map [10] and Talmor coordinate map [34]. They also compared the data obtained from similar studies by Losey et al. [30] and Wada et al. [31]. In all cases, substantial differences in the position of regime boundaries were observed as compared to those observed in conventional TBRs. In the case of the Baker coordinate map, the annular regime was observed in microchannel systems at conditions corresponding to a dispersed regime in conventional systems. This is the direct result of substantial differences in the relative strength of capillary and gravity forces which dictate the flow regime [35]. More accurate hydrodynamic models are required for predicting flow regimes in a structured MTBR. In consecutive hydrogenation reactions, the order of selectivity to the intermediate product follows the pattern of an increased mass transfer rate [36]. Therefore, a MTBR packed with small catalyst particles could over perform monolith and other structured reactors.

Krishnamurthy et al. [37] investigated a two-phase flow across a bank of staggered circular micropillars, 100 μm long with a diameter of 100 μm and a pitch-to-diameter ratio of 1.5. The device was sealed from the top with a Pyrex cover which allows flow visualization. Pressure measurements were performed with pressure transducers placed at the inlet, exit, and in three locations along the channel length. A two phase gas-liquid flow was obtained by passing the two phases through a micromixer, which was located upstream of the main pillar array and was fabricated as a part of the device. The mixer has two inlets, one for the water and one for the nitrogen, and a series of closely spaced 50 μm diameter circular pillars with a pitch-to-diameter ratio of 1.3 (Fig. 7).

Fig. 7

Schematic view of pillared microreactor. Reprinted with permission from “Phys. Fluids. 2007;19(4):043302” [37].

© 2007, AIP Publishing LLC.

Four different flow patterns were observed, namely bubbly slug, gas-slug, bridge and annular flows. In a follow up study, the authors concluded that no significant deviations were observed between water and ethanol with respect to flow patterns [38]. However, the reduction of surface tension affected the flow pattern transition lines. The authors modified constant B in Eq. (7) to account for the effect of surface tension on pressure drop. The resulting correlation was able to predict the combined experimental data for ethanol to within ± 10 %.

2.2 Pressure drop

The calculation of pressure drop in trickle bed reactors is an important design parameter. It depends on the particle packing characteristics and the fluid properties. Many earlier correlations are based on the data obtained with non-spherical particles such as Raschig rings and cylinders. These correlations under-predict the pressure drop for spherical particles. Therefore, it is important to use a proper correlation based on flow regime, particle size and operating pressure and temperature.

2.2.2 Semi-structured micro trickle bed reactors

Mohammed et al. [41] studied pressure drop using three different liquid distributors, namely a single point distributor, spray nozzle and multipoint distributor in a cylindrical column (polyvinyl chloride) with an internal diameter of 100 mm and a height of 163 cm. The highest pressure drop was observed for the single point liquid distributor due to a high degree of liquid maldistribution. They observed an accumulation of the liquid phase in the center region in the upper part of the packing. A more uniform initial distribution of the liquid phase was observed with the multipoint and spray nozzle distributors.

The authors observed higher static liquid holdups at higher foam pore density. Moreover, especially for solid foams with high pore density (e.g. 25 and 30 ppi), the static liquid holdup increased with increasing superficial flooding velocity. The authors observed different values of the static holdup at high superficial flooding velocity in experiments with foams of high pore density. They concluded that the rapid filling of the small pores may cause forming of immobile gas pockets that cannot be replaced by the liquid phase. The number of gas pockets and their location, in turn, is of a random nature and possibly affected by local geometry-flow interactions.

They also used two different pre-wetting modes, which are defined as follows:

1. “Kan-liquid” mode: adjusting the flow rates to the desired gas and liquid superficial velocity after operating the packed bed for 10 min in the high-interaction regime. This mode represents the upper hydrodynamic boundary [42];

2. “Levec” mode: adjusting the flow rates to the desired gas and liquid superficial velocity after initial liquid flooding of the packed bed followed by draining the bed under gravity for 15 min according to Levec et al. [43] until only the residual liquid holdup remained. This mode represents the lower hydrodynamic boundary.

A distinctive pressure drop multiplicity was found for the single point distributor. A higher pressure drop branch was found for the “Kan-liquid” pre-wetting mode, which can be described as rapid flooding. This mode of operation leads to an increased static liquid holdup especially in the wall region in the upper part of the column. They concluded that uniform initial liquid distribution could dampen the occurrence of hydrodynamic multiplicity.

The authors developed correlations to predict the pressure drop and liquid holdup model in tubular reactors with solid foam packings. To account for the influence of different velocities and physical properties of the fluids, the liquid Reynolds number, the liquid Galileo number and gas Weber number were applied. Furthermore, the correlations include geometric properties of the foam, such as window diameter, specific surface area and porosity.

(5)dpdLρLg = a1ReLb1ReLc1(asdw1 − εε)d1

(6)εL = a2(ReGWeLReL)b2GaL(dpdLρLg)c2(1 − εε)d2

The window diameter (d_w) was used as a characteristic linear dimension of the foams for the dimensionless numbers. The coefficients and exponents of the correlations are listed in Tables 1 and 2.

Table 1

Coefficients and exponents for pressure drop correlation (Eq. (5)).

Pre-wetting	Pore density, ppi	Flow regime	a1	b1	c1	d1
Levec	10, 20	Trickle	21.56	0.15	3.45	9.80
Levec	25	Pulse	0.21	−0.15	0.53	0.37
Kan-liquid	10, 20	Trickle	0.03	0.20	3.02	6.86
Kan-liquid	25	Pulse	0.06	0.04	0.46	0.26

Table 2

Coefficients and exponents for total liquid holdup correlation (Eq. (6)).

Pre-wetting	Pore density, ppi	Flow regime	a2	b2	c2	d2
Levec	10, 20	Trickle	2426	0.78	−0.18	0.82
Kan-liquid	10, 20	Trickle	170.3	0.60	−0.16	0.35

In the co-current upflow configuration, Stemmet et al. [44–46] observed bubble and pulsing regimes in a 1 cm width reactor filled with 10 ppi solid foam packing with a solid holdup of 7 %. The liquid holdup increases with increasing liquid velocity and decreasing gas velocity, up to the maximum voidage of the solid foam packing. (Fig. 8 (a)). The liquid holdup increases as the ppi number of the solid foam increases due to higher capillary pressures and higher static liquid holdup.

Fig. 8

(a) Liquid holdup in a 10 ppi solid foam packing in the co-current upflow configuration; (b) Liquid holdup in 10 and 40 ppi solid foam packings in the co-current downflow configuration. Reprinted from “Chem. Eng. Sci., volume 62, Stemmet CP, Meeuwse M, van der Schaaf J, Kuster BFM, Schouten JC., gas–liquid mass transfer and axial dispersion in solid foam packings, 5444–5450” [46].

© 2007, with permission from Elsevier.

Similar behavior was observed in the co-current downflow configuration; however, the absolute value of the liquid holdup at low gas velocities was much lower and it did not exceed 0.4 (Fig. 8 (b)).

As the liquid viscosity increases from 0.8 to 2.0 mPa/s, the liquid holdup slightly increases by 10–15 %. This increase in liquid holdup is more pronounced in the 10 ppi solid foam packing due to the higher wetting of the packing material, resulting in more effective drainage of the solid foam packing.

2.3 Liquid holdup

The liquid volume fraction in trickle bed reactors is characterized by dynamic liquid holdup and static liquid holdup. The latter is proportional to the number of stagnant liquid pockets. The difference between these two parameters often determines the effective liquid residence time distribution in trickle bed reactors. For kinetically controlled reactions, the reaction rate are directly proportional to the extent of internal wetting of particles.

Liquid holdup can be expressed in two ways:

1. total liquid holdup (ε_L) defined as volume of liquid per reactor volume;

2. liquid saturation (β_L) defined as a volume of liquid per void volume.

The liquid hold-up consists of two parts: dynamic liquid holdup (ε_Ld) and static liquid holdup (ε_Ls). The latter is the volume of liquid which remains in the bed after draining the bed.

The liquid hold-up controls the liquid residence time and therefore reactant conversion. Marquez et al. [47] studied the dispersion and holdup in MTBRs. They used two reactors with lengths of 7 and 97 cm and with a diameter of 2 mm packed with particles of 106–125 μm. They observed a rather high liquid holdup between 0.65 and 0.85. Under these conditions, the particles were fully wetted, which is a beneficial effect for the catalytic applications. Besides, the gas flow rate affected neither the holdup nor the dispersion. No liquid flow was observed from the bed when both gas and liquid flow was stopped simultaneously, suggesting zero dynamic holdup. Thus the flow in MTBR is comparable with the packed beds and liquid flow only, and the operation mode does not affect the hydrodynamics in MTBRs. High liquid holdup values translate into good wetting characteristics of catalysts and are highly desired for catalytic reactor applications. These values are considerably higher as compared to those in industrial TBR [48].

A similar effect was observed by Yang et al. [49]. The use of diluent particles with a diameter of 300 μm increased liquid holdup from 0.05 to 0.20. In beds diluted with small particles, the liquid is more easily retained between the particles due to the increased liquid-solid contact and high capillary forces that lead to higher total liquid hold up. A similar observation was reported by Kulkarni et al. [50]: the dynamic hold up almost doubled and the wetting efficiency reached 90 %. Bej et al. [51] studied the effect of the diluting the catalyst bed with non-porous inert particles on the performance of MTBR. By testing various sizes of non-porous silicon carbide particles, it was determined that the 160–190 μm diameter particles packed in a relatively small MTBR with a volume of 5 ml considerably increased the liquid hold up and provided a comparable reactor performance with a larger (100 ml) TBR.

MTBRs are employed in fine chemicals and pharmaceutical synthesis. Al-Herz et al. [52] studied the hydrogenation of 1-heptyne in a trickle bed reactor over a 1 wt % Pd/Al₂O₃ catalyst. Various solvents were screened and 95 % selectivity was achieved in isopropanol. They observed that the liquid flow rate significantly affects the reactor performance. The gas flow rate was 10 ml min⁻¹ and the liquid flow rate was increased gradually from 5 ml min⁻¹ to 20 ml min⁻¹. The liquid hold up is very low and thus a partial wetting of the catalyst is expected. The wetting efficiencies were calculated by the correlation of Lebigue et al. [53] which is given below:

where Fr_L (liquid Froude number), Mo_L (liquid Morton number) and ε_B (bed porosity) are representative of the flow behaviour at the pellet scale, the physical properties of the fluid and the bed topology. In all liquid flow rates the catalyst wetting efficiency was above 91 % and increasing up to 97 % with the highest liquid flow rate. The total liquid holdup was calculated by correlation from Lange et al. [54];

where d_P and d_R are the particle diameter and reactor diameter respectively. Total liquid holdup changes from 0.28 to 0.34 with the increasing liquid flow rate. Partial catalyst wetting is expected in such low liquid holdups. Higher liquid flow rate increased the hydrogenation rate and it was leveled off after 15 ml min⁻¹. A similar trend was observed for the 1-heptene selectivity. Improper catalyst wetting could be the culprit for lower observed reaction rates at lower flow rates, and the reaction may fall within the kinetic regime above the liquid flow rate of 15 ml min⁻¹.

2.4 Flow maldistribution and start-up effects

Liquid phase maldistribution is an important factor in the design, scale-up and operation of trickle-bed reactors [55]. Large catalyst particles, uneven catalyst loading and a non-uniform liquid inlet distribution enhance channeling. For a longer reactor, a small variation in vertical orientation during installation of reactor could also lead to liquid maldistribution. Liquid maldistribution is often related to non-complete wetting of the reactor zone near the inlet. However, no quantitative information is available in the literature.

For smaller particles, larger surface to volume ratio leads to better liquid spreading [56]. Therefore inert fine particle are used to improve flow distribution. Wu et al. [57] suggested the use of an inert fine particle along with catalyst particles to improve wetting in the reactor bed. Use of spherical particles with a uniform particle size can provide better control on local packing characteristics and therefore on local mixing and transport rates.

Van Herk et al. [58] performed a flow image analysis; it revealed that segregation of inert and catalyst pellets leads to preferential pathways in the bed, where the zones with the smallest particles were filled with stagnant liquid that was refreshed much less often. The initial conversion level and the deactivation rate were only reproducible when the segregation was prevented by matching free-fall velocities of particles during the filling procedure of the reactor.

Fishwick et al. [59] compared several different reactor types in selective hydrogenation of 2-butyne-1,4-diol into the corresponding alkene over a 0.5 wt % Pd/Al₂O₃ catalyst: a stirred tank reactor; wall coated monoliths with a diameter of 5 and 10 cm; a wall coated capillary microreactor with a 2 mm diameter and a 50 mm trickle bed reactor packed with 6 mm pellets diluted with 200 μm silicon carbide. The highest initial reaction rate was observed in the TBR. However, the selectivity to alkene measured at a 90 % conversion was 100 % in the monolith, capillary and stirred tank reactors; however, it was only 93 % in the TBR. The poor selectivity was due to the presence of stagnant zones in TBR broadening the residence time distribution. All reactors operated closed to a plug flow behavior under the used flow conditions [36].

Marquez et al. [60] studied different start up procedures and step changes in flow in MTBRs. As the Bond number was in the order of 10⁻² to 10⁻³, the up flow and down flow patterns were identical and hysteresis was not observed. However, a little hysteresis was observed in the pressure drop during step changes in liquid (and gas) flow rates. The characteristic time to reach a new steady state after start up or after a step change in gas or liquid flow rates was the same. This time depends on the pressure drop and can exceed the liquid residence time by a factor of 3.2. The volume of the gas feed section, which is the volume of the tubing and the connections after the mass flow controller to the top of the bed, should be minimized to reduce the characteristic time to reach the new steady state after a step change.

2.5 Axial dispersion

Measurement of residence time distribution is the main method to determine axial dispersion. The axial dispersion model is used to describe all deviations from an ideal plug flow mode via a single parameter called dispersion coefficient (D_L), which is often expressed in terms of the liquid phase Peclet number (Pe = uL/D_L). The main factor contributing to the deviation from plug flow behavior is non-uniform porosity distribution which could lead to channeling and short-circuiting. Non-uniform bed porosity can be induced by a wide particle size distribution or bed assembly methods resulting in non-uniform bed densification.

Mears [61] proposed a criterion to estimate the minimum reactor length required to avoid dispersion effects:

where n is the reaction order, and C is the concentration. It can be seen that dispersion effects can be more pronounces at higher conversions or higher reaction orders, other parameters being the same.

At low liquid flow rates (Re_L < 4), Fu et al. [62] showed that the dispersion depends on particle diameter

where Bo is the Bodenstein number (Bo = uL/D_L).

MTBRs usually have short lengths that require accounting for axial dispersion. In conventional TBRs, the particle size has the same order of magnitude with capillary length. Thus, the gravity is an important factor; it should be taken into consideration during the reactor design. However, particle size in MTBRs is much smaller than the capillary length that makes the effect of gravity negligible. The analysis of the multiphase flow becomes the analysis of the perturbation from the single-phase flow. The single-phase dispersion in catalytic reactors is well understood; the reactor models are developed based on the dispersion coefficient in which the Peclet number appears in the dimensionless axial dispersion reactor model equation.

The Peclet number has two limits: at very low liquid velocities, the dispersion is dominated by molecular diffusion where the Peclet number becomes uL/D_M, where D_M is the diffusion coefficient. The other limit is the fully developed turbulent flow, where the particle Peclet number is around unity and independent of diffusion. Marquez et al. [47, 60] studied the intermediate regime in a 2 mm inner diameter, 97 cm long MTBR packed with 106–125 μm particles. The liquid was fed into the packed bed with a needle and the dispersion due to the packing was analyzed by subtracting the spread observed in the empty reactor (Fig. 9).

Fig. 9

Micropacked bed for stability and transient times studies. (A) 10 μl tracer injection loop. (B) Zoom showing the way that gas and liquid are introduced in the bed. (C) Differential pressure transmitter. (D) Gas-liquid separator. (E) Refractive-index cell. Reprinted with permission from “Ind. Eng. Chem. Res. 2010;49(3):1033–1040” [60].

© 2010, American Chemical Society.

Van Herk et al. [63] studied the scaling down of TBR for hydrodesulphurisation reactions. Their reactor setup involved six parallel MTBRs with a diameter of 2 mm packed with particles of 100 μm. They studied the residence time distributions with a pulse of 10 μl of colored dye. The inlet and outlet effects were eliminated by performing the RTD experiments with and without the reactor. In the single–phase experiments, they found an accurate agreement with the expected residence time within 5 % error. They claimed that the origin of the error was due to small bubbles present in the space between the catalyst particles. The liquid residence time was independent of the gas and the liquid flow rates. Thus it could be concluded that gravity plays little role in the hydrodynamics of MTBRs.

In the MTBR at the Reynolds numbers around 1, the flow patterns were either bubble flow or segregated flow. In the latter case, the gas flow could bypass the liquid resulting in low conversions. The Peclet numbers were above 100 [63]. The high value of particle Peclet number implies the validity of a plug flow estimation with short reactor lengths and an order of magnitudes of 10–20 particle diameters.

The dispersion ina MTBR and in a liquid-only flow fixed bed reactor turned out to be similar. Maiti et al. [8] reported the effect of porosity on the hysteresis observed in TBRs. They found out that the hysteric behavior of porous particles is much different from that of nonporous particles. While a hysteric behavior can expected with porous microparticles, however the effect is too small to be observed for microparticles packed in MTBRs.

Kulkarni et al. found out that the use of 6 mm porous particles rather than the nonporous glass beads have an impact on the residence time distribution in a laboratory-scale TBR with diameter of a 5 cm and 35 cm length [50]. Higher dispersion was observed with the porous particles due to the diffusion of the tracer into the catalyst pores. Moreover, the addition of 200 μm inert fine particles increased the Peclet number. The Peclet number increased as the superficial liquid velocity increased.

Bodenstein numbers were reported in the range of 10–100 in a semistructured MTBR, which indicates the importance of dispersion effects [32]. The Peclet number was studied in a MTBR with a diameter of 10 mm and a 5 cm length packed with 150–250 μm catalyst particles at temperatures of 25–65 °C and 6.1 bar [64]. The experiments were performed in step-response measurements with an inert tracer, which showed that the axial dispersion should be taken into account.

Several authors reported that effect of gas flow rate on dispersion is almost negligible [50, 65, 66]. However supercritical fluids behave like a gas rather than a fluid. Jin et al. [67] studied the hydrodynamics in a MTBR with a diameter of 12.5 mm, a length of 110 cm under supercritical conditions. The reactor was filled with particles in the range of 0.67–1.32 mm in diameter. N-hexane was utilized as the supercritical fluid at pressures of 35–45 bar and temperatures between 140 and 280 °C. Nitrogen was used as the gas. It was observed that the residence time distribution changes drastically with the gas flow rate. The mean residence time and the dispersion decreased with the increasing gas flow rate under supercritical conditions.

3.1 Mass transfer

Micro trickle bed reactors operate with a relatively small energy input as compared to other reactors as there is no rigorous mixing mechanism present. Due to complete wetting, two types of mass transfer rate are relevant for MTBR, i.e gas-liquid and liquid-solid.

3.2 Heat transfer

Many reactions performed in MTBRs are often highly exothermic, and heat released in these reactions is transported by conduction and convection. This may lead to non-uniform temperature distribution in the reactor bed. Local hot spots may form in the micro packed bed due to poor fluid distribution. It could also deactivate the catalyst and reduce conversion and selectivity [85]. Besides, solid-liquid heat transfer limitations affect the kinetics of reactions. Heat management inside the catalyst bed without causing catalyst deactivation and/or by-product formation is a critical task in the design of a MTBR. A near isothermal temperature distribution can be achieved by structuring catalyst and inert material zones, and removing heat via cooling. In some cases, the gas or liquid flows can be recycled to reduce conversion per pass and control temperature. Correct calculation of heat transfer rates is crucial for design and scale-up of MTBR.

The particle size in MTBR is usually very small, therefore no intraparticle temperature gradient inside the catalyst particle is observed. Few studies in literature deal with particle-liquid heat transfer rates in trickle-bed reactors. The main reason is probably the difficulty to find accurate experimental methods.

Heat transfer rates increase with both increasing gas and liquid flow rate. In the trickle flow regime, the effect of the gas flow rate is rather weak, while the transition to pulsing flow results in an increase in the local average heat transfer rates [86].

Boelhouwer et al. [87] proposed a correlation for the Nu number

where the Re number is based on the linear liquid velocity and the particle diameter. Ruether et al. [88] presented a correlation for particle-liquid mass transfer in which a power of 0.77 for the Re number was found.

Heidari et al. [89] have developed CFD models to describe micro and meso scale heat transfer in TBRs at the gas-liquid interphase. Their micro scale model was a modified version of a double-slit [90] model implemented to determine the interfacial heat transfer. The meso-scale model was solved numerically to determine the effect of particle shape on an interfacial heat transfer. They calculated the interfacial Nusselt number as a function of pressure drop, liquid holdup and wetting efficiency. The Nusselt number increased with an increase in gas phase Reynolds and Prandtl numbers and decreased as the liquid phase Prandtl, Reynolds and Eötvös numbers increased.

Fedorov et al. [91] investigated heat transfer in a heat sink with rectangular micro channels by developing a numerical model for fluid flow and the heat conduction in a silicon substrate. They observed rather complex heat flux patterns because of a strong coupling between convection in the fluid and conduction in the substrate. They concluded that axial heat conduction in micro packed beds should be taken into account. Lee et al. [92] experimentally studied the heat transfer in rectangular micro channels with a width from 194 to 534 μm and the depth of five times the width in each case. In case of a 1D heat transfer model with constant temperature or constant heat flux boundary conditions, the deviations from the 3D full conjugate analysis were 12.4 % and 7.1 %, respectively, demonstrating the importance of the use of numerical simulations, instead of 1D correlation.

Norton et al. [93] reported that between 60 to 80 % of the heat generated in microstructured reactors is lost to the surroundings regardless of the insulation. These losses could only be avoided by application of special insulation techniques, such as vacuum insulation. Therefore microreactors, even with thick insulating packaging, cannot be operated adiabatically; heat losses to the environment should be taken into account. A pseudo-homogeneous two parameter model with an effective thermal conductivity (λ_eff) and heat transfer coefficient at the wall (h_w) could satisfactory describe heat transfer in a MTBR [94].

Dudas et al. [95] studied a highly exothermic reaction of hydrogenation of thymol on a Ni/Al₂O₃ catalyst (1.2 mm) in a 40 mm internal diameter MTBR. The temperature profile was controlled with a pre-heater. A hot zone was observed in the reactor. The position of the hot zone moved along the bed during start up due to different catalyst wetting at low liquid flow rates. However, at a high liquid flow rate of 2.85 kg h⁻¹, a considerable temperature gradient was observed along the reactor lengths. Due to the hot zone in the beginning of the reactor bed, thermal decomposition of reaction products occurred with formation of several by-products.

To reduce temperature gradients, we proposed to use radio frequency (RF) heating instead of a preheater [5]. The reactor configuration was composed of alternating catalyst and heating zones. The heat was generated inside the reactor by magnetic particles placed in an external RF field. The heating zones contained nickel ferrite microparticles with a diameter of 110 μm. The catalytic zones were packed with catalyst particles of the same size. The temperature profile was studied along the reactor length after fast heating of a part containing magnetic particles (Fig. 10). The effective thermal conductivity of the bed (λ_eff) and convective heat transfer coefficient for the heat losses to the environment (h_ext) were calculated from a 1D transient heat transfer model (Eq. (23)).

where a is the external wall area per unit of the reactor volume, C_p is the heat capacity and q is the volumetric heat production rate. The conduction and convective losses contributed 30 and 70 % of the total energy losses, respectively. The number and relative positions of heating zones within the reactor were optimized to obtain a near-isothermal temperature profile (Fig. 11). The position of heating zones depends on the heat losses to the environment and heat generation rate. Three heating zones, positioned at specific locations inside the bed, provided a temperature nonuniformity of 2 °C over a length of 50 mm.

Fig. 10

Temperature of reactor bed along the axial direction as a function of time (symbols). Lines are the guide for an eye. The interface between the heated and non-heated zones is located at x = 40 mm. Reprinted from “Chem. Eng. J., volume 243, Chatterjee S, Degirmenci V, Aiouache F, Rebrov EV. Design of a radio frequency heated isothermal micro-trickle bed reactor, 225–33” [5].

© 2014, with permission from Elsevier.

Fig. 11

Steady state fluid temperature along the axial direction of the micro trickle bed reactor for (a) single zone, (b) two zone, and (c) three zone configuration. Reprinted from “Chem. Eng. J., volume 243, Chatterjee S, Degirmenci V, Aiouache F, Rebrov E V. Design of a radio frequency heated isothermal micro-trickle bed reactor, 225–233” [5].

© 2014, with permission from Elsevier.

3.3 Scale up

For large-scale processing, multiple MTBR may be connected in series or in parallel. For a number of fine chemical syntheses that require large liquid residence time, TBR are operated in a semi-batch mode when the liquid flow has complete recycle with a recycle to feed ratio of above 50.

Hickman et al. [96] scaled up a 14 inch TBR packed with 200–400 μm particles from the laboratory scale. They observed a substantial loss in the efficiency due to inefficient catalyst wetting. By adjusting the superficial liquid velocities for efficient catalyst wetting and after a thorough investigation on the catalyst deactivation, the reactor was scaled up by a factor of 3 × 10⁶.

The hydrodynamics of the TBRs are very sensitive to the scale of the reactor, which brings about the challenge of transposing laboratory data into a commercial-sized reactor. The scale up methods relying on 1D models are not reliable because they use pressure drop and liquid holdup; this changes within the reactor in a complex fashion with the change in a reactor’s scale. Large bed diameters lead to liquid maldistribution. Ranade et al. [97] proposed a general scale up approach (Fig. 12) based on CFD models that account for the porosity and thus consider liquid maldistribution and local hot spot formation in hydrodesulfurization and hydrodearomatization reactions.

Fig. 12

Schematic of Steps in Scale-up and Scale-down. Reprinted from “Chem. Eng. Sci., volume 62, Gunjal PR, Ranade V V. Modeling of laboratory and commercial scale hydro-processing reactors using CFD, 5512–5526” [97].

© 2007, with permission from Elsevier.

Simulations were then carried out at temperatures between 320 and 380 °C, pressures of 200–800 bar, and initial H₂S concentration between 1.0 and 2.5 vol %. In MTBRs, superficial gas-liquid velocities were much lower; this reduced the overall performance and mass transfer rates due to increased dispersion. The predicted results were compared with the experimental data. While the results were not conclusive, it was shown that slight alteration in hydrodynamic parameters led to very different predictions in the reactor performance. Simulation showed that the sensitivity of conversion at high temperature and pressure is rather low; however, it becomes very sensitive to any alteration in hydrodynamic parameters at low temperatures and high liquid flow rates. Therefore, it is crucial to develop correlations under realistic operating conditions and reactor sizes that adequately represent hydrodynamic and multi-scale transport processes in order to achieve predictive models for scale up.

1 Micro trickle bed reactors

Conversion of biomass to fuels and chemicals has attracted great attention as one of the future technologies for a sustainable the low carbon society [110]. Hydrogenation of biomass-derived molecules into sugar alcohols is an important step for the production of value added chemicals and intermediates for the chemical industry [111]. In the future, trickle bed reactors might be a choice for various reactions. Jeon et al. [112] described the hydrogenation of C₉-aldehyde into C₉-alcohol over a supported Ni-MgO catalyst. Their reactor has a diameter of 0.5 inch and a length of 80 cm. A selectivity of 90 % to C₉-alcohol was observed at full conversion at 130 °C and 400 psi.

Kilpio et al. [113] investigated the hydrogenation of glucose into sorbitol in a MTBR (Scheme 1).

Scheme 1

Hydrogenation of glucose to sorbitol.

Sorbitol is an alternative sweetener and a platform chemical for a wide variety of compounds. Their reactor has a diameter of 10 mm and a length of 70 mm packed with a commercial Ru/C catalyst pellets of 100–250 μm or 330–500 μm in diameter. 90 % selectivity to sorbitol was obtained in the 90–130 °C range. The catalyst deactivation was clearly pronounced, especially at the entrance region where the highest reactant concentration was observed. They combined reaction kinetics and deactivation kinetics in a single model that agreed well with the observed reaction behavior. The model is based on an axial dispersion model using temperature-dependent kinetics and deactivation. Their simulations predicted no temperature gradient rise inside the particles. The effectiveness factor was around 70 % due to internal diffusion limitations.

Xi et al. [114] used MTBRs with a diameter of 13 mm a length of 45 cm for a ruthenium-catalyzed hydrogenolysis of lactic acid (2-hydroxypropanoic acid) to propylene glycol (1,2-propanediol), which is an important reaction step for the biomass utilization (Scheme 2). Lactic acid can be produced from renewable sources such as carbohydrates derived from agricultural crops and waste biomass. Lactic acid is very reactive due to its hydroxyl and carboxyl functional groups and it can be converted into various chemicals through polymerization, esterification, dehydration and oxidation [115].

Scheme 2

Hydrogenolysis of lactic acid into propylene glycol.

This reaction is usually performed in a batch reactor, because mass transfer limitations could be eliminated via vigorous agitation. The MTBR was packed with catalyst particles and inert glass beads of the size 200 μm and operated at a liquid flow rate of 100 ml h⁻¹. The translation of a batch protocol to a MTBR demonstrated similar values for the reaction rate of 4.9 × 10⁻⁴ kmol m⁻³ cat s⁻¹ when the both reactors operated in the intrinsic kinetic regime. The MTBR approaches fully wetted behavior and essentially acts as a differential reactor. A detailed model was developed for MTBR that accounts for interphase mass transfer, temperature gradients and partial wetting of catalyst particles. However, the model predictions must be handled cautiously, since predicting fractional wetting and mass transport coefficients did not always match with the experimental observations.

Son et al. [116] employed a MTBR with a diameter of 16 mm and 200 mm length for the transesterification of sunflower oil with methanol to produce biodiesel over a CaO catalyst with the size of 1–2 mm. The reactor allowed for a separation of the products and reactant (methanol) when it was operated in a semi batch mode where the condensed products were collected in a reservoir attached to the bottom of the reactor and the methanol was collected in a condenser at the exit. A 98 % yield of fatty acid methyl esters was achieved at 373 K with oil and methanol flow rates of 3.8 and 4.1 ml h⁻¹.

Glycerol is the side product of biodiesel production. The growing production of biodiesel as a renewable source-based fuel leads to an increased amount of glycerol. The utilization of glycerol strongly affects the process economy. Brander et al. [117] employed a commercial MTBR (Vinci Technologies) to investigate various possible glycerol reaction pathways such as oxidation over a Pt-Bi catalyst, aqueous phase reforming and hydrogenolysis over Ru/C and Ru/TiO₂ catalysts. Glycerol is a highly functionalized molecule and thus a large number of value added chemicals can be obtained from glycerol by a variety of chemical reactions (Scheme 3). Xi et al. [118] reported the hydrogenolysis of glycerol where a MTBR was used with a diameter of 12.5 mm and 61 cm length in a similar study. A one dimensional, non-isothermal reactor model was developed based on the kinetic model, intra particle mass transport; liquid-solid and gas-liquid mass transfer coefficients were calculated using correlations for conventional reactors. They applied the reactor model to a large set of steady state trickle bed reactor experiments at a variety of operating conditions to predict glycerol conversion. The three kinetic constants were adjusted and all other constants and coefficients governing hydrodynamics and mass/heat transfer in the model are taken straight from literature or textbook correlations. As a result, good agreement was observed between the predicted and experimental results.

Scheme 3

Transformations of glycerol into high value chemicals.

The conventional batch synthesis of 6-hydroxybuspirone, an important pharmaceutical product, suffers from low process safety, difficulties in upscaling of cryogenic equipment and long reaction times. Therefore a continuous process is highly desired. LaPorte et al. [119] successfully scaled up the process of formation of 6-hydroxybuspirone from buspirone, which includes enolization, oxidation and quench steps in a TBR (Scheme 4). At first, the authors used a two stage CPC CYTOS microreactor that provided a 85–92 % conversion with a residence time of 5 min corresponding to a 300 g of product per day. Then a pilot-plant TBR with four columns was constructed. A yield of 83 % (41.4 kg) was obtained in steady state operation for 72 h. This time was limited by the amount of the available feed solution and potentially can be extended.

Scheme 4

Preparation of 6-hydroxybuspirone. Adapted from “Org. Process. Res. Dev. 2008;12(5): 956–966” [119].

© 2008, American Chemical Society.

Adipic acid is a commercially important intermediate for the production of nylon-6,6. The commercial processes include a two-step oxidation of cyclohexane that results in large amounts of N₂O, a major air pollutant and greenhouse gas. The synthesis based on oxidation with hydrogen peroxide is considered a greener alternative. An adipic acid yield of 50 % in the direct oxidation of cyclohexene by hydrogen peroxide was reported by Shang et al. [120] at 100 °C and a residence time of 20 min in a MTBR with a diameter of 10 mm and a length of 20 cm packed with 212–300 μm inert glass beads.

Scheme 5

One-step oxidative cleavage of cyclohexene to synthesize adipic acid.

Selective hydrogenation of α,β-unsaturated aldehydes, such as citral, leads to a wide range of fine chemical intermediates. A simplified reaction scheme of successive citral hydrogenation is shown in Scheme 6. Wörz et al. [121] replaced a batch operation into a continuous process. They employed 1 wt % Pd/SiO₂ and 5 wt % Pd/Al₂O₃ catalysts covered with ionic liquid layer ([BMIM][N(CN)₂]), 1-butyl-3-methylimidazolium dicyanamide in a MTBR with a diameter of 16 mm packed with catalyst pellets of 300–350 μm [121]. The use of ionic liquids enhanced the reaction selectivity towards citronellal to 100 % at 70 °C at 52 % conversion with a liquid flow rate of 1 ml min⁻¹.

Scheme 6

Consecutive reactions in citral hydrogenation.

Maki-Arvela et al. [73] employed MTBRs for the parallel screening of catalysts. They used six parallel reactors, each of which had the internal diameter of 10 mm packed with two different catalyst particle sizes of 63–90 μm and 150–200 μm. The model reaction used to observe the performance of the reactors was citral hydrogenation. They obtained excellent reproducibility.

2 Semi-structured and structured trickle bed reactors

A semi-structured MTBR based on monolithic structures [122, 123], filled with catalyst particles of 0.8 mm in diameter, was used in hydrogenation of alpha-methylstyrene over a Pd catalyst [124]. The monolith segments had a square cross section of 10 × 10 mm², a length of 5.0 cm and contained 64 parallel flow channels with a hydraulic diameter of 1 mm. The authors compared the semi-structured MTBR, a wall coated monolith and a larger scale TBR. The monolithic reactor and semi-structured MTBR demonstrated reaction rates that were at least two times higher than those achieved in the conventional TBR. Slug flow was the dominant flow regime observed at the liquid and gas superficial velocities in the range of 0.02 to 0.2 m s⁻¹ and 0.04 to 0.2 m s⁻¹ respectively.

Van Herk et al. [58] studied the hydrogenation of biphenyl over a Pt-Pd/Al₂O₃ catalyst in a MTBR with a diameter of 2.2 mm and a length of 475 mm. They have used microfabricated beds where pillars in 100–150 μm diameter were uniformly distributed. Phosphite, amine, and olefin oxidation with ozone were studied in a structured MTB and up to 100 % increase in selectivity was observed. [30, 31]

A structured MTBR was used for the production of singlet oxygen from chlorine gas and hydrogen peroxide at different pressures, flow rates and chlorine mole fraction [32]. A singlet-delta oxygen yield of 78 % was obtained at the optimal reactor performance, occurring at a gas flow rate of 75 ml min⁻¹ and an outlet pressure of 0.13 bar. A complete bottom-to-top modeling, design, construction and flow testing of a fully MTBR system were developed.