Riccardo
Uglietti‡
ab,
Daniele
Micale‡
a,
Damiano
La Zara
b,
Aristeidis
Goulas
b,
Luca
Nardi
a,
Mauro
Bracconi
a,
J. Ruud
van Ommen
*b and
Matteo
Maestri
*a
aLaboratory of Catalysis and Catalytic Processes, Dipartimento di Energia, Politecnico di Milano, via La Masa 34, 20156 Milano, Italy. E-mail: matteo.maestri@polimi.it
bDepartment of Chemical Engineering, Delft University of Technology, Van der Maasweg 9, 2629 HZ Delft, The Netherlands. E-mail: J.R.vanOmmen@tudelft.nl
First published on 5th May 2023
We show the potential of coupling numerical and experimental approaches in the fundamental understanding of catalytic reactors, and in particular fluidized beds. The applicability of the method was demonstrated in a lab-scale fluidized bed reactor for the platinum-based catalytic oxidation of hydrogen. An experimental campaign has been carried out for synthesizing the catalyst powders by means of atomic layer deposition in a fluidized bed reactor and characterizing them. Catalytic testing has been also run to collect data both in fixed and fluidized bed configurations. Then, after the validation of the in-house first-principles multiscale Computational Fluid Dynamic – Discrete Element Method (CFD–DEM) model, the fundamental understanding which can be achieved by means of detailed numerical approaches is reported. Thus, the developed framework, coupled with experimental information, results in an optimal design and scale-up procedure for reactor configurations promising for the energy transition.
These properties are provided by the complex fluid dynamics behaviour, related to the movement of the solid phase, which has a strong influence on gas–solid interaction and hence, on the conversion and selectivity of catalytic reactions.1,2 On the one hand, the multiphase flow induces a mixing which affects the species contact time. On the other hand, the formation of fluid dynamic structures (e.g. bubbles, particle clusters) can introduce additional transport resistances which decrease the catalyst utilization.
In this context, the numerical models are not always capable to catch the aforementioned complexities, hindering the accurate description of fluidized systems specially in non-conventional reactor geometries which can be adopted to develop novel sustainable processes. Hence, the fundamental understanding of these reactor units requires advanced computational approaches, such as the detailed multiscale modeling.3–6 According to this approach, the first-principles descriptions of the phenomena occurring in the reaction environment are coupled in a unique framework, which allows for the understanding of their interplay and the fundamental analysis of the desired system.
This modeling approach has provided interesting insights in the context of fixed bed and structured reactors.7–12 However, literature investigations of fluidized bed reactors for several catalytic processes (e.g. ozone decomposition,13 biomass gasification,14,15 methanation,16 methanol to olefins17) do not usually use first-principles multiscale modeling. These computational approaches neglect the description of phenomena that could occur inside the reactive environment. Indeed, in these works the surface chemistry is solved without accounting for the mass transport resistances, which can strongly affect the outcomes of the process.18 Consequently, Maestri and co-workers19 extended the multiscale modeling to fluidized systems by coupling the Computational Fluid Dynamic – Discrete Element Method (CFD–DEM) model of the gas–solid flow20–22 with the description of the physico-chemical phenomena involved in the fluidized systems (i.e. gas–solid species/heat transport and catalytic reactions via detailed microkinetic models).23 This approach has been applied to accurately reproduce the behaviour of systems operated in chemical regime,19 and it has been additionally combined with speed-up methodologies to significantly reduce the computational cost associated with the solution of the detailed catalytic kinetics,24 enabling the analysis of million particle fluidized reactors.24,25 In this work, we show the potential of combining computational multiscale modeling with information obtained through experimental campaigns to investigate complex systems, such as catalytic fluidized bed reactors. In particular, we have analysed a lab-scale fluidized bed in which hydrogen oxidation takes place over a platinum on alumina supported catalyst.
During the experimental campaign, we have synthesized the Pt catalyst by means of atomic layer deposition (ALD) in a fluidized bed reactor,26,27 starting from alumina Geldart B powder.28 Catalytic particles with two different Pt loadings (1 and 2.3% w/w) have been synthetized and subsequently mechanically and chemically characterized to quantify all the properties needed to properly simulate the system by means of the multiscale CFD–DEM framework. We then operated the reactor unit in fixed bed configuration in order to collect the data needed to derive a kinetic expression at the selected operating conditions (i.e. ambient temperature and with oxygen as the limiting reactant). Finally, we performed the experimental testing in the fluidized bed reactor configuration for both the two different Pt loadings and analysed the outlet species composition at different inlet gas velocities.
The corresponding operating conditions have been also simulated with the multiscale CFD–DEM, properly configured by means of all the experimentally derived properties, to show the reliability of the approach not only in chemical regime but also in case of an important contribution of the mass transport resistances. To do so, the comparison of the performances, in terms of oxygen conversions at the outlet of the reactor, between the experimental data and the multiscale one has been carried out by achieving an excellent agreement (maximum deviation 5%). Hence, the fundamental understanding that can be provided by the computational multiscale approach is finally presented focusing both in the fluid dynamic and the chemical aspects.
All in all, this work shows the potentiality of coupling first-principles numerical approaches with experiments which can allow for the fundamental investigations of both existing and novel catalytic fluidized systems29–32 and also for the hierarchical refinement of less detailed numerical approaches (i.e., Euler–Euler models33) for the simulation and testing of industrial-scale devices.
The angle of repose tests has been used to measure the particle friction factor (μ). The particles have been deposited over a parallelepipedal box using a funnel, forming a pyramid from which the angle of repose (θ) can be graphically evaluated and then used to compute the friction factor as follows (eqn (1)):
μ = tan(θ) | (1) |
(2) |
(3) |
The particle size distribution has been measured by means of a Beckman Coulter laser diffraction particle analyser.
After the characterization of the Pt loading, the size distribution of the Pt nanocluster, deposited with the ALD experiments onto the alumina surface, has been assessed with transmission electron microscopy (TEM) images of the catalytic alumina surface. In particular, a sample of approximately 0.1 g has been crushed, successively diluted in ethanol and, then, dispersed onto the copper TEM grid of 3.05 mm diameter. The TEM images of several different catalyst portions have been performed by means of a JEOL JEM-1400 electron microscope operating at 120 kV, with a field of view of few tens of nanometers.
(4) |
Fig. 1b reports the computational grid with a cell-to-particle diameter ratio of 3 with respect to the average particle diameter and 2 with respect to the maximum particle diameter observed in the solid granulometry. Fig. 1c reports the results of the initial packing procedure by means of the DEM particle tracking algorithm. At the beginning of the procedure, the computational domain is empty, and the particles are injected from the top of the reactor. The packing simulation has been stopped once a steady-state packing has been obtained on the basis of the recorded particle velocities, i.e. maximum velocity below 10−4 m s−1.
Table 1 lists the bed and the particle properties in the simulation. In particular, it reports the minimum, maximum and average particle diameters of the experimentally measured size distribution reported in the ESI† (section S1). Additionally, further details on the measurement of the friction factor and the catalytic particle density are given in the following Results and discussion section, while the particle restitution coefficient has been selected equal to 0.8 from the alumina data available in work of Gorham and Kharaz.38
Catalytic bed properties | |
---|---|
Particle diameter Dp [μm], min/max/average | 150/500/300 |
Particle density ρp [g cm−3] | 1.36 |
Particle number Np [−] | 83767 |
Mechanical properties | |
---|---|
Young modulus E [MPa] | 3 |
Poisson ratio ν [−] | 0.22 |
Restitution coefficient e [−] | 0.8 |
Friction factor μ [−] | 0.542 |
The following boundary conditions have been selected. At the bottom of the reactor, i.e. the inlet, the velocity corresponding to the experimental flow rate to be reproduced has been imposed, together with the adopted composition reported in the Catalytic tests section. No slip conditions have been imposed at the reactor walls as well as a zero gradient condition for the species and for pressure. At the top of the reactor, i.e. the outlet, atmospheric pressure has been imposed as well as fully developed profiles for both velocity and species. The temperature has been set equal to 296 K, i.e. room temperature, in the whole domain. Indeed, isothermal simulations have been performed since the maximum experimentally determined temperature difference in the reactor is below 5 K.
Each reactive CFD–DEM simulation has been carried out for at least 10 residence times with a first-order chemical kinetic, whose parameter is evaluated in the Results and discussion section, to achieve the pseudo-steady state. To deal with the intrinsic oscillations of the composition at the outlet of the reactor, caused by the bubbling of the fluidized bed, the following procedure has been adopted to obtain the reported CFD conversion data. Each numerically predicted conversion of oxygen has been computed by time averaging the oxygen mass fraction at the outlet of the reactor, obtained by means of cup-mix average, for at least three residence times at the pseudo-steady state.
With respect to the particle density (ρp), the fixed bed bulk density of the alumina particles is equal to 0.84 g cm−3. After the packing of the three different beds of 12 g of particle, the heights have been measured obtaining a value equal to 2.60 cm for the normal configuration, 2.43 cm for the dense configuration and 2.67 cm for the loose one, and the pressure drop vs. velocity tests have been run.
Fig. 3 reports the measured pressure drops as a function of the inlet superficial velocities. The symbols are the experimentally measured pressure drops for both the dense (blue squares) and loose (red circles) configurations, while the solid lines represents the pressure–velocity trends obtained by fitting the Ergun equation in case of the dense (blue line) and loose (red line) pressure drop points (i.e. the ones below the 70% of the minimum fluidization velocity). A value equal to 0.357 and 0.392 for dense and loose packing respectively. Finally, the particle density has been computed from the two bed void fractions. A particle density of 1.36 g cm−3 is obtained in both the configurations, confirming the reliability of the adopted procedure.
At last, an average alumina particle diameter of 300 μm has been found (with d50 equal to 290.1 μm). The whole distribution has been reported in the ESI† (section S1).
Fig. 4 Deposited Pt weight percentage as a function of the exposure time of the alumina to the Pt precursor (a) and the number of the performed ALD cycles (b). |
Fig. 5 shows an example of the comparison between SEM images of bare alumina particles (reported on the top) and the Pt loaded one (on the bottom) randomly chosen from the respective batches. The images indicate that no significant differences are present between the two batches. Consequently, no relevant particle breakage or deformation are caused by the vibrated fluidization adopted in the ALD experiments (average circularity and roundness of the particles measured with ImageJ39 software equal to 0.96 and 0.92 for bare alumina and 0.96 and 0.88 for coated one, thus close to 1, obtained in case of perfect spheres).
Fig. 5 Comparison of the SEM image of two samples from two different batches: bare alumina (top) and alumina loaded with Pt (bottom). |
Given the consistent Pt loadings obtained after the ALD procedure, we proceeded with the characterization of the catalytic particles. In particular, we characterized the size of the Pt nanoparticles obtained on the alumina surface.
Fig. 6 shows an example of the TEM images of the fragments of the alumina particles loaded with 2.3% Pt (Fig. 4a, red circle). The TEM image allows for detecting the Pt nanoclusters (Fig. 6, dark spots) and thus to quantify their area by means of the ImageJ39 software. Then, the Pt nanoparticle diameters have been computed from the nanoparticle area assuming their shape as perfect hemispheres. Table 2 reports the computed Pt nanoparticle average diameter for all the Pt loadings highlighted in Fig. 4, as long as their consequent specific surface area av in m2 per kg of supported catalyst (kgcat) computed according to eqn (5).
(5) |
Fig. 6 Example of the TEM images of the Pt/Al2O3 catalyst. The dark spots are the Pt nanoclusters deposited during the ALD experiments. |
Pt loading ωPt [% w/w] | Average diameter dPt [nm] | Specific catalytic surface [m2cat kg−1] |
---|---|---|
1.0 | 1.99 | 1406 |
2.3 | 2.52 | 2221 |
(6) |
(7) |
Fig. 7 reports as symbols the measured oxygen conversions at different inlet velocities, also corresponding to the ones used in the fluidized catalytic tests, while the red dashed line reports the oxygen conversions resulting from the fitting of the kinetic constant by means of the measured conversions (symbols) and the 1D heterogeneous model applied with an oxygen first order expression kinetics and the Reichelt correlation, leading to a kinetic constant at room temperature (k[296]) equal to 2.18 ± 0.03 [m3 kgcat−1 s−1] (95% confidence interval). After the evaluation of the kinetic constant, the Damköhler number has been computed for all the experimental data used for the fitting and value ranging from 0.55 and 0.8 have been obtained. As expected the experimental data are not in full chemical regime due to the high reactivity of the Pt catalyst, highlighting the needs of an heterogeneous model to properly evaluate the kinetic constant.
This derived k[296] has been then used as input in the multiscale simulation, together with the Reichelt correlation for the gas–particle mass transfer, for modeling the lab-scale fluidized bed.
An excellent agreement has been obtained (maximum error of 5% and 4% for 1% and 2.3% Pt respectively) over a wide range of operating velocities which leads to different fluidization regimes, ranging from the bubbling bed resulting at the lowest simulated flow rate (i.e. 0.5 L min−1) to the slugging regime observed at high flow rates (above 0.7 L min−1). Additionally, by reducing the feed velocity the importance of the mass transport resistances becomes more and more relevant and the multiscale framework is always capable to correctly reproduce the mutual interactions between kinetic and species transport.
Furthermore, the excellent agreement achieved in case of the 2.3% Pt catalyst, confirms the reliability of the derived kinetic constant. Indeed, this kinetic constant have been derived directly from the 1% one by considering just the ratio of measured specific surfaces, and no-additional fitting have been needed.
Hence, the framework has demonstrated reliable at different inlet velocities, and in particular, at different fluidization and chemical regimes, and consequently it can be adopted to assist the experimental campaign by adding fundamental understanding of the investigated reactor unit and process, as discussed in the following section.
With respect to the analysis of the multiphase fluid dynamics of the system, is possible to identify the different fluidization regimes obtained in the reactor as a function of the operating velocity. At the lowest one, the typical bubbling behaviour of a Geldart B28 powder can be observed. In particular, the reactor is characterized by small bubbles generated close to the gas inlet, e.g. the one at the bottom right of the plane, and constantly growing in diameter along the fluidized bed.
By increasing the velocity at 16.1 cm s−1, a transition between the bubbling and the slugging regime is observed. Indeed, on the one hand, bubbling behaviour is still clearly distinguishable from the first two snapshots where a bubble is rising and growing on the left side of the plane. On the other hand, this situation is alternated with the generation of a gas piston at the bottom of the fluidized bed. However, it rapidly turns into a large growing elongated bubble frames (III–VI), since the velocity is still not sufficiently high for the insurgence of a full slugging regime. With respect to the 20.7 cm s−1 velocity, the slugging phenomenon is observable. Pistons of gas generate at the bottom of the reactor and move upward in the bed (frames I–III), even if there is still the formation of a large elongated bubble occupying the major part of the bed height (frames IV–VI). Finally, the transition from the bubbles of gas to the cluster of particles arises at the 34.5 cm s−1. In this condition, a piston of gas is generated at the bottom of the reactor, pushing upward a piston of solid with a lateral raining of particles, as reported in the I frame. The solid piston thins during the upward movement due to the lateral particle raining, until a thin layer of solid is present at the top of the gas piston and a fluidized bed is observable at the bottom (frame III–IV). This behaviour is alternated with a vigorously fluidized dense phase breaking into solid spots, i.e. the blue regions representing the cluster of particles (frame V–VI).
The first-principles multiscale model thus can provide additional insights into the reactor environment that can be hardly or cannot be achieved through experiments. Indeed, on the one hand, it can be used to help or even replace expensive device that are able to monitored the movement of the solid phase inside the unit, such as X-ray tomography. On the other hand, it can allow for the understanding of how the fluid dynamic structures influences the reactivity of the system. Here, for example, the amount of reactants and products within the bubbles and the emulsion can be quantified without experimental invasive methods (e.g. in situ probing) enabling to identify the presence of transport resistances. Indeed, the reactant species predominantly flow through the bed in gas bubbles (yellow regions in the void fraction maps of Fig. 9), and they must reach emulsion (blue regions in the void fraction maps of Fig. 9) before arriving to the catalyst surface through the particles boundary layer. In this view, an investigation on the importance of this mass transfer resistance can be performed by observing the oxygen mass fraction maps reported for each velocity and frame in Fig. 9. This mass transfer resistance becomes more and more important by increasing the inlet flowrate. Starting from the lowest velocity, the transfer resistance between the bubble and the emulsion, is evident from the oxygen gradient between the two regions, as highlighted by the light red spots of oxygen at the same position of the bubble. Similarly to the 11.5 cm s−1 velocity, oxygen rich bubbles are still present due to bubble–emulsion mass transfer limitations, and higher oxygen concentrations in the bubble and more pronounced oxygen gradients can be observed, coherently with the increment of the bubbles diameters. In the case of the 20.7 cm s−1 velocity, high oxygen concentrations close to the feed one are observed in the bubbles due to the decreased surface to volume ratio of the bubbles which reduces the species exchange between the gas bubbles and the emulsion phase. However, as for the other velocities, the typical mixing inside the emulsion phase is still present with a homogeneous oxygen distribution observed in this region for all the snapshots avoiding the presence of another possible transport resistance related to the position of the catalytic particles inside the emulsion. Finally, at the 34.5 cm s−1 velocity, a relevant by-pass of the bed is observed for the gas phase, leading to streams of reacted oxygen coming from the dense phase and clusters alternate periodically with stream of almost unreacted oxygen.
Beyond the qualitative analysis of the macroscale mass transfer resistances, e.g. the bubble–emulsion ones, additional insights are provided by the framework with respect to the interplay of the different mass transfer resistances and catalytic reactions. In particular, the importance of three mass transfer resistances can been evaluated. The first one which accounts for the species transport between the bubble and the emulsion (i.e. bubble–emulsion resistance), the second one considers the species gradient between the emulsion and the catalytic particle surface (i.e. gas–particle resistance), while the third one results from the combination of the other two resistances (i.e. overall resistance). The importance of each resistance has been quantified by means of the ratio between the oxygen mass fraction at the end point of the species transfer and the oxygen fraction at the starting point of the species transfer, according to the following equations:
(8) |
(9) |
(10) |
Fig. 10 shows a reactor cross section characterized by relevant bubble area for the 11 cm s−1 velocity. The active particles located in the considered section for the selected time frame have been reported and colored on the basis of the importance of the considered mass transfer resistance (ϕgp and ϕbe). The gas phase of the cross section is indeed colored on the basis of the oxygen mass fraction evidencing the mass transfer limitations observed in Fig. 9. Indeed, a red oxygen rich area can be identified at the right side of the cross section coherently with the bubble position observable in the solid distribution map reported aside of the cross section. Then, a relevant oxygen gradient is present until the oxygen poor area in dark blue is reached far from the bubble.
First, the ϕoverall have been computed for all the active particles, obtaining an average value of 0.29, thus confirming the importance of mass transfer resistances in the investigated system, but excluding at the same time a fully external mass transfer regime coherently with the experimental observations in Fig. 8, where a change in the Pt loading has produced a non-negligible change in the oxygen conversion for the same operating conditions.
Then, each single mass transfer mechanisms have been analysed. In the upper cross section image of Fig. 10, the particles are colored on the basis of the gas–particle ϕ number, whereas in the bottom image of the cross section, the particles are classified on the basis of the bubble–emulsion ϕ number. Two different spatial trends are observed for the two ϕ numbers. As expected, ϕgp is independent from the location of the active particle with respect to the bubble. This contribution is mainly influenced by the specific surface of the particle, leading to value close to 1 for the smallest particles (e.g. 150 μm) until 0.5 for the one with the largest ones (e.g. 500 μm). Thus, the value of ϕgp increases by decreasing the particle diameter. On the other hand, a clear pattern emerges for ϕbe. Its value is strongly dependent on the distance between the center of the bubble and the location of the particle in the emulsion and, in particular, it decreases by increasing the distance. Indeed, the active particles located near to the bubble–emulsion interface are characterized by ϕbe close to 1, while the ϕbe value of the ones present in the center of the emulsion can be also equal to 0.2.
As a whole, the work has proven the capability of the adopted numerical investigation, after an experimentally-driven evaluation of the mechanical and chemical properties of the catalytic particles, in analysing the phenomena inside the reactor, which can be extended to arbitrary complexity geometries and kinetic mechanisms given the multiscale nature of the developed and adopted framework. Consequently, this paves the way for the analysis of novel fluidization concepts, e.g. pulsed or confined beds, in the context of heterogeneous catalytic processes and for the refinement of the multiscale models of industrial fluidized units.
Footnotes |
† Electronic supplementary information (ESI) available: Particle size distribution, reactive CFD–DEM multiscale framework. See DOI: https://doi.org/10.1039/d3re00152k |
‡ The authors contributed equally to this work. |
This journal is © The Royal Society of Chemistry 2023 |