Dry catalyst impregnation in a double cone blender

Powder Technology 221 (2012) 57–69

Contents lists available at SciVerse ScienceDirect

Powder Technology journal homepage: www.elsevier.com/locate/powtec

Dry catalyst impregnation in a double cone blender: A computational and experimental analysis Francis S. Romanski, Atul Dubey, Arthur W. Chester, M. Silvina Tomassone ⁎ Rutgers Chemical and Biochemical Engineering, 98 Brett Rd. C-118 Piscataway, NJ 08854, United States

a r t i c l e

i n f o

Available online 9 December 2011 Keywords: Catalyst impregnation Discrete element method Granular mixing

a b s t r a c t The dry impregnation of catalysts is widely used in industrial catalyst preparation, however, until recently, it has not been possible to model this system computationally. In this work, a novel algorithm for the spray and inter-particle transfer of fluid onto and within a rotating bed of granular catalyst support was explored using discrete element method (DEM). The simulations were validated by experiments utilizing a geometrically identical double cone blender fixed with a single nozzle impregnator. The effects of liquid flow rate and fill level were explored at a fixed rotation rate of 25 rpm. Specifically, three flow rates of 1.5, 2.5 and 5 L/h were selected and evaluated at a 30% and 45% fill fraction by volume. Mixing analysis and fluid concentration distributions were used both experimentally and computationally to investigate the propagation of fluid throughout the bed with the goal of modeling and improving industrial content uniformity. It was discovered that low flow rate systems and lower fill fractions resulted in better mixing and content uniformity throughout the bed. Results obtained from our model show good agreement with experiments, and therefore it was demonstrated that a novel fluid transfer algorithm incorporated into DEM could be used to accurately model dry catalyst impregnation, therefore introducing a new tool for optimizing catalyst manufacturing. © 2011 Elsevier B.V. All rights reserved.

1. Introduction Catalyst impregnation is one of the most crucial steps in the preparation of heterogeneous catalysts. In this process, metal salts or complexes are dissolved in an aqueous solution and contacted with a porous oxide catalyst support such as alumina (Al2O3) or silica (SiO2) [1–4]. In a traditional impregnation process, metal solutions are sprayed using a liquid nozzle over a rotating granular blender. During the typical 30–60 minute processing time, the sprayed fluid is absorbed and the metal ions are adsorbed from the solution onto the high surface area support. Subsequently, the catalyst support is subject to drying and further pretreatment in order to transform the metal from its precursor state into its active form [3]. Impregnation is divided into two distinct types: wet and dry. Wet impregnation consists of a solid catalyst support immersed in liquid media, resulting in a two phase system. However, during dry impregnation a volume of liquid less than or equal to the pore volume of the catalyst support is sprayed onto the catalyst support. As a result, the entire process remains a granular system. Dry impregnation is also commonly referred to as pore filling impregnation. The dry impregnation process typically occurs in large blenders designed to mix the granular bed both during and after spraying. As the liquid penetrates into the dry powder, the particles absorb the

⁎ Corresponding author. Tel.: + 1 732 445 2972. E-mail address: [email protected] (M.S. Tomassone). 0032-5910/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.powtec.2011.12.018

fluid, the metal is adsorbed, and the liquid content of the particles increases until a relative saturation is reached based on the pore volume of the catalyst support (typically 20 to 50 wt.%). Contrary to what is desired in dry impregnation, the liquid content may increase beyond the pore volume in localized areas, often resulting in liquid bridging and increased cohesive forces between wet support particles. This may ultimately result in poor mixing and inadequate content uniformity. Furthermore, wet particles have a significantly higher density than dry particles, and a density gradient may develop within the granular bed if the fluid spray rate is too high. This density gradient may not only have a negative impact on mixing, but also may lead to granular segregation [5]. Mixing and segregation are further compounded by the notorious mixing phenomena associated with powder blenders including dead zones, segregation, and a general lack of content uniformity [6,7]. Unfortunately, a change in the mixing or fluid distribution as a consequence of the aforementioned phenomena will significantly and adversely affect the homogeneous distribution of the metal precursor into the solid [6,8,9]. Consequently, poor fluid distribution will not only lead to manufacturing issues such as poor quality control and non-uniform distributions of metal, but also longer processing times; which are very expensive in highvolume or precious metal impregnations. While dry impregnation is a ubiquitous technique in catalyst manufacturing, the process is very seldom studied at the macroscopic scale. Most research in this field has been allocated to the surface chemistry of the impregnation, the metal content inside individual support particles, or transport models isolated to a very small window

58

F.S. Romanski et al. / Powder Technology 221 (2012) 57–69

of the overall process [1–3,10]. As a result, there are several open questions remaining for dry impregnation such as: i) how mixing and flow are affected when particles contain a certain degree of moisture or are saturated with liquid, ii) what factors affect the homogeneity of the liquid distributed within the bed, iii) the extent and distribution of dead zones or localized supersaturation for a given impregnator configuration, iv) the effect of tunable process parameters such as fill level, spray rate, and rotation speed, and finally v) the distribution of metal impregnated in the support. Fortunately, many of the problems associated with dry catalyst impregnation can be investigated by applying modern computational techniques in conjunction with experimentation. In particular, discrete element method (DEM) has been increasingly used to study granular systems, albeit more often in the pharmaceutical industry. Specifically, DEM has been used to evaluate granular mixing, flow, die-filling, granulation, milling, and compression for pharmaceutical products [11–15]. Although the catalyst manufacturing industry uses many analogous processes, the modeling of catalyst manufacturing processes using DEM remains unexplored. In this work, we focus on the development of a novel algorithm for fluid transfer onto a granular bed of rotating catalyst support using a DEM modeling. The algorithm was developed to allow for the transfer of fluid to and between particles following the saturation of each individual support particle. In addition, a non-constant cohesion parameter increases as a function of liquid content on a per particle basis. Since the model described has not been previously developed, the goal of this work was to adequately represent a dry impregnation process using several key process parameters and evaluate the model for accuracy when compared to equivalent experiments. The computational and experimental setups consisted of a double cone blender filled at two distinct fill levels, 30% and 45% by volume of porous catalyst support. Three flow rates of liquid were evaluated for content uniformity and mixing throughout the bed for each of the aforementioned fill levels using a single spray nozzle located at the center of the bed. Each simulation was done in conjunction with a series of parametrically and geometrically equivalent experiments for validation. It was desired to establish a working methodology and algorithm for a dry impregnation with the goal of modeling, optimizing, and evaluating the dry impregnation process for optimal processing conditions and the most favorable fluid content

uniformity without the need of performing numerous and often very expensive experiments. The remainder of this paper is organized as follows: Section 2 describes the equipment and methodology for the experimental techniques and the DEM simulations, as well as the algorithm development, and key user-defined simulation parameters. Section 3 describes the results and discussion of the model in direct comparison to the experimental results. The final section of the paper evaluates conclusions and illustrates areas for improvement and further application of this work. 2. Materials and methods 2.1. Summary of the chemicals used and experimental equipment 4.76 mm γ-Alumina spheres (3/16 in.) were kindly donated from Saint-Gobain Norpro (Stow, OH, USA). The granular spheres contained a surface area of 200 m 2/g and a pore volume of 0.389 cc/g. All water used was de-ionized using the Millipore system. Copper (II) nitrate trihydrate and 0.1 M hydrochloric acid were obtained from Fischer Scientific (USA) and used as received; experiments including copper solutions were performed with a 1 M aqueous solution, where 241.60 g of Copper (II) nitrate trihydrate was dissolved in one liter of deionized water and mixed magnetically until complete dissolution occurred. Double cone blending experiments were conducted using Patterson-Kelly 10-quart rotating double cone blender. An impregnator was retrofitted into the system using Swage-lock ¼ in. fittings and a ¼ in. NPT nozzle adapter. The spray nozzles were the MW-105, MW-145, and MW-275 Microwhirl™ nozzles purchased from BETE Fog Nozzle, Inc (USA); nozzles were operated at three flow rates: 1.5, 2.5, and 5 L/h. The flow was controlled using a Cole-Parmer Masterflex™ peristaltic pump retrofitted to the Swagelock fittings. The mixing vessel was 24 cm in diameter and 30 cm in height. The inclination angle of the cone was 45°. Experiments were conducted with the vessel loaded at two distinct fill volumes of 30% and 45% by volume corresponding to 1.75 kg and 3 kg of alumina spheres, respectively. Fluid was sprayed in an 8 cm diameter circle at the center of the vessel, corresponding to 1/3 of the rotational axis under spray (~ 11% of the surface area). A schematic of the experimental setup is shown in Fig. 1.

Fig. 1. Schematic and dimensions of Patterson-Kelly 10-quart double cone apparatus retrofitted with a single center-located spray nozzle used in the experiments; the dimensions are identical to the geometry constructed computationally.


2.2. Experimental procedure Prior to impregnation, alumina catalyst support was heated in a Fisher Scientific (Fisher Scientific, USA) oven at 300 °C and allowed to cool to room temperature inside a glass desiccator (Fisher Scientific, USA) prior to each experiment. It is important to note that this step was designed to allow the experiments to start at roughly 0% moisture, as was the case for the simulations. All liquid solutions were placed in a sealed flask and pumped into the impregnator nozzle using the aforementioned peristaltic pump. The blender was operated at a constant speed of 25 rpm. The 4.7 mm spheres were removed for analysis at 7 points across the access of rotation at 3.5 cm intervals from center using a bucket-style powder thief (typically 5–10 g per sample). All samples were stored in air-tight glass vials prior to analysis for moisture. Moisture was analyzed by heating the samples to 300 °C for 6 h and analyzing the associated mass change. Moisture concentrations were normalized by the weight of the sample. Copper concentration for metal-based impregnations was established by soaking each sample in hydrochloric acid for 24 h. The aliquot was then tested for copper concentration using an Ocean Optics USB4000 UV-spectrophotometer (Ocean Optics, Inc., Dunedin, Florida, USA) at a wavelength of 290 nm (calibration R 2 = 0.998). Each experiment was conducted in quadruplicate, the results were averaged, and error was calculated as the standard deviation of the mean. Mixing was characterized using the relative standard deviation equation shown below in Eq. 1, where S is the index of segregation, C is the concentration of fluid, and n is the number of samples. s RSD ¼ c sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n∑C 2 −ð∑C Þ2 S¼ nðn−1Þ

ð1Þ

tangential contact forces and ∑F Body denotes the sum of all body forces. This resultant force is computed for each particle with a time-step on the order of 10 μs, and thus, the new particle positions are computed by numerically solving the equations of motion. Using Newton's law the position of a spherical particle i that has j number of contacts with its surroundings is related to the resultant force by Eq. (3). h i mi x€i ¼ ∑j F nij þ F tij þ ∑K F body h i I i θ€i ¼ ∑j Ri F tij þ ∑l τbody

F ¼ −kn δn −γ n δ_ n δn : n

ð2Þ

In this equation, FTotal is defined as the force on a particle due to all interactions with other particles and/or boundary elements, as well as the effect of external force fields such as gravity, cohesion, or electrostatic interactions. The term ∑F Contact accounts for all the normal and

3=2

1=4

ð4Þ

In the above, δn is the deformation (particle overlap), kn refers to the Hertzian normal stiffness coefficient, γn is the normal damping coefficient, and δ_ n is the rate of deformation. There are several other types of contact models available for use in DEM simulations [17–19], however the Hertz–Mindlin model is widely used due to its efficiency and applicability to engineering problems. The normal stiffness coefficient is knobtained by Eq. (5): Kn ¼

F Total ¼ ∑F Contact þ ∑F Body

ð3Þ

In Eq. (3), mi represents the mass of the particle of radius Ri, xi is its position, x€i its acceleration, and its Ii moment of inertia. Fijn and Fijt are the normal and tangential components of the contact force on particle i due to itsjth contact respectively. The term ∑k F body accounts for all body forces acting on the particle using a summation index k. In this study, gravity is considered to be the only body force exerted on the particles (k = 1, ∑k F body ¼ mi g). The rotational components of mo€ and sum tion are the angular displacement θ, angular acceleration θ, ∑k τ body of all torques due to body forces using summation index l. The contact forces are calculated using the Hertz–Mindlin contact model, where the normal force F n from a contact that resulted in a normal overlap δn is given by Eq. (4):

2.3. Computational methodology A commercially available software, EDEM™ (DEMSolutions Inc.) was used to design the model and algorithm for fluid transfer. Simulations were conducted on a series of Intel® Core™2 Quad 2.5 GHz processors. Each simulation took approximately 2 weeks to obtain 10 min of dry impregnation data. DEM simulates a granular material while treating each particle as a discrete individual unit, as opposed to continuum models which treat the material as bulk [16]. In these simulations, Newton's second law of motion was used to compute each particle's velocity and position by integrating its acceleration. The distance between the centers of each pair of particles, modeled as spheres, and between particles and boundaries was computed at every time step. A contact was detected if the distance between the centers of the particles (in case of a particle–particle contact) was less than the sum of the particle radii, or the distance between a boundary and the center of a particle (in case of a particle–boundary contact) was less than the particle's radius. A very small overlap (δ) was allowed in each of the normal and tangential directions. Two types of forces were considered: contact forces due to particle–particle or particle–boundary collisions and body forces due to gravity; these forces are summarized in Eq. 2 below. The boundary can be any physical object in the system, such as the walls and baffles. The forces were resolved into normal and tangential components, which are independent of each other.

59

qffiffiffiffiffiffiffiffi 4 E Reff : 3 eff

ð5Þ

In the above equation, Eeff is the effective Young's modulus of two colliding entities (two particles or a particle and a wall). For entities with Poisson's ratios v1 and v2, Young's moduli E1 and E2, Eeff is given by: 2

2

1 1−v1 1−v2 ¼ þ : Eeff E1 E2

ð6Þ

In the case of particle–particle collision, Reff is defined as the effective radius of the contacting particles. If two contacting particles have radii R1and R2 the effective radius is obtained by: Reqv ¼

R1 R2 : R1 þ R2

ð7Þ

In case of a particle–wall collision, the effective radius is simply the particle radius. With the knowledge of the normal stiffness coefficient and a chosen coefficient of restitution ε, the normal damping coefficient is calculated as: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi # u " u5 mkn : γn ¼ 2t InðεÞ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3 In2 ε þ π2

ð8Þ

Following the work of Mindlin and Deresiewicz [20], the tangential force F t is calculated in a similar method as its normal counterpart. The tangential contact force also consists of elastic and damping components. When a tangential overlap of δt is detected and there is a

60


corresponding normal overlap of δn due to the same contact, then the tangential force is calculated as:

simulated catalyst support particle, leading to a net mass increase. The mass flow rate of the fluid is defined as Qspr:

F ¼ −kt δt −γt δ_ t δn :

Q spr ¼ Nv ¼ N

t

1=4

ð9Þ

In the above equation, kt is the tangential stiffness coefficient, and γt is the tangential damping coefficient. The tangential stiffness coefficient is calculated [21] by: 1=2 1=2

kt ¼ 8Geff Req δn

ð10Þ

where Geff is the effective shear modulus, which, for two entities with shear moduli G1 and G2, gives: 1 2−v1 2−v2 ¼ þ : Geff G1 G2

ð11Þ

The tangential displacement (or overlap) δtis calculated by t time-integrating the relative velocity of tangential impact, vrel between two colliding entities (either interparticle or particle–wall contact): →

→

t

rel

δ ¼ ∫ ν tdt:

ð12Þ

In addition to the above contact dynamics, additional features were developed in this model in order to allow for fluid absorption, accumulation and transfer among the particles. The fluid spray components are modeled as discrete droplets, which are sprayed from above the rotating bed and are absorbed upon contact with the simulated catalyst medium particles upon contact. The corresponding contact causes the simulated fluid droplet to essentially disappear while simultaneously transferring the mass of the fluid droplet to the

m p

ð13Þ

where N is the number of fluid droplets, v is the volume, m is the mass and p is the density of each droplet. Analogous to the experimental conditions, the particles in this study are modeled to absorb fluid up to 35% of their weight. After saturation of the catalyst particle occurred, additional fluid allows the support particle to be considered supersaturated, and as a result, transferred excess fluid to any nonsaturated particle that it comes into contact with. The amount of fluid transferred between two particles i and j in every contact when one of them was supersaturated, is defined as: Q Tr ¼ κ mi −mj :

ð14Þ

In the above equation, κ is a proportionality constant which reflects by the rate of fluid transfer, mi and mj are the respective mass of each of the particles; for this work, κ was defined as 0.01. Hence the amount of fluid transferred per contact is a function of the difference in the wetness of the contacting particles. The following parameters were used in this study to accurately represent the experimental system: particles contained a shear modulus of 2 × 10 6 N/m 2, a Poisson's ratio of 0.25, a density of 1500 kg/m 2, and a monodisperse diameter of 4.7 mm. For particle–particle interactions, the coefficient of static friction was taken as εS = 0.5, with a coefficient of rolling friction εr = 0.01, and a coefficient of restitution εoes 0.1. The impregnator walls were modeled as steel with a shear modulus of 80 GPa, a density of 7800 kg/m 3, and a Poisson's ratio of 0.29. The particle–wall interactions were set to a 0.5 coefficient of static friction, 0.01 for the coefficient of rolling friction and a 0.1

Fig. 2. Discrete element method snapshot of simulated double cone apparatus containing 4.7 mm catalyst support particles at 30% loading; fluid spray at center of the bed is visible.


61

Fig. 3. Sampling in the DEM model of the double cone impregnation, seven symmetric boxes were used along the axis of rotation.

coefficient of restitution. A snapshot of the DEM simulation is shown in Fig. 2; this snapshot displays a double cone blender filled at 30% volume with the conical fluid spray at the center of the bed visualized. The 30% fill level was modeled using 40,000 particles, while the 45% fill level was modeled using 60,000 particles. Approximately 726,000, 1.2 M and 1.7 M fluid droplets per second were sprayed when simulating a flow rate of 1.5 L/h, 2.5 L/h and 5 L/h respectively. Sampling in the bed was conducted with a similar methodology to the experiments by constructing seven sampling boxes 3.5 cm apart. The particles in each box were analyzed for fluid content and subsequently compared to the experimental results. Sampling computationally, as well as experimentally, was done along the axis of rotation, which has previously been shown to be the area with the

poorest mixing and content uniformity in double cone blenders [6]. An image of the sampling boxes is shown below in Fig. 3, and an example of a simulation snapshot is shown in Fig. 4, where the color represents the concentration of fluid as a function of the amount of fluid sprayed. 3. Results and discussion 3.1. Fluid content distribution Fluid concentration was evaluated throughout the granular bed during a double cone dry impregnation by measuring the fluid content both computationally and experimentally at seven distinct zones in the

Fig. 4. Simulation snapshot from top-down view of double cone blender after 1 min of spray at 2.5 L/h (medium value) containing a 45 vol.% fill level of 4.7 mm spherical support pellets; color scale represents concentration of fluid per particle as a function of total fluid sprayed at 1 min.

62


granular bed along the axis of rotation. The center point was denoted at point zero and three points on either side of the axis were evaluated at 3.5 cm intervals axially in both systems. The extreme left was denoted by the value −3, and the extreme right +3. Sampling along the axis of rotation was selected due to the poor axial mixing known for double cone geometry blenders. Three spray rates, 1.5, 2.5, and 5 L/h, were evaluated for two different fill levels, 30% and 45%. A 45% fill level was

chosen to avoid any dead zones associated with fill levels higher than 50%, and 30% is the approximate volume required to fill the lower cone at a stationary position. The DEM simulations were constructed geometrically and parametrically equivalent to the experiments. The first simulations were conducted using the 45% fill level at the three aforementioned flow rates; the fluid concentration along the axis is shown in Fig. 5, where each line corresponds to the fluid content at

Fig. 5. Simulated fluid concentration of 45% fill level impregnation as a function of axial position and time; axial position is marked left to right by − 3 (far left), − 2, − 1, 0 (center), 1, 2, and 3 (far right) — temporal evolution begins at bottom and increases vertically in one minute intervals; three increasing fluid spray rates are shown as A) 1.5 L/h, B) 2.5 L/h, C) 5 L/h.

Fig. 6. Experimental fluid concentration of 45% fill level impregnation as a function of axial position and time; axial position is marked left to right by − 3 (far left), − 2, − 1, 0 (center), 1, 2, and 3 (far right) — temporal evolution begins at bottom and increases vertically in one minute intervals; three increasing fluid spray rates are shown as A) 1.5 L/h, B) 2.5 L/h, C) 5 L/h.


63

each minute for a total of 10 min starting with 1 min at the bottom. Fig. 5A displays the resulting fluid concentration at the 1.5 L/h flow rate, with Fig. 5B and C as the 2.5 and 5 L/h cases, respectively. It is important to note that the temporal evolution is represented by individual series vertically, where each series represents 1 min of real-time for both the experiments and the simulations for a total of 10 min. The 1.5 L/h flow rate resulted in a very flat fluid content distribution, indicating a well-controlled and well-mixed dry impregnation. The two

alternate cases of 2.5 and 5 L/h exhibit some anisotropy in the initial few minutes of the simulations, but are improved to a more uniform profile by the completion of the ten minute impregnation. The equivalent experiments for the 45% fill level are shown in Fig. 6, where the flow rates are again 1.5, 2.5, and 5 L/h for Fig. 6A, B, and C, respectively. Again, each minute is displayed as a separate line starting with minute one at the bottom. Each line represents the mean value of four experiments; the error bars represent one

Fig. 7. Simulated fluid concentration of 30% fill level impregnation as a function of axial position and time; axial position is marked left to right by − 3 (far left), − 2, − 1, 0 (center), 1, 2, and 3 (far right) — temporal evolution begins at bottom and increases vertically in one minute intervals; three increasing fluid spray rates are shown as A) 1.5 L/h, B) 2.5 L/h, C) 5 L/h.

Fig. 8. Experimental fluid concentration of 30% fill level impregnation as a function of axial position and time; axial position is marked left to right by − 3 (far left), − 2, − 1, 0 (center), 1, 2, and 3 (far right) — temporal evolution begins at bottom and increases vertically in one minute intervals; three increasing fluid spray rates are shown as A) 1.5 L/h, B) 2.5 L/h, C) 5 L/h.

64


standard deviation of the mean, and the error bars were only displayed for minute one, minute five, and minute ten for visual clarity. As expected, there was better fluid content uniformity at the lowest spray rate of 1.5 L/h. Interestingly, the pattern exhibited by the fluid concentration did not entirely reach a uniform value by the conclusion of the ten minute experiment, thus indicating that the axial mixing in the 45% fill level was not entirely adequate for compensating the incoming fluid spray at the center of the bed. Next, the 30% fill level was evaluated computationally; the results are presented in Fig. 5 where each line from the bottom corresponds to 1 min of real-time simulation results. The fluid concentration profiles for this 30% case are shown in Fig. 7. It is important to note that the relative flow rates were significantly higher in the 30% fill level in relation to the amount of support present, and as a result, fluid content increases as a function of time at a faster rate. Moreover, the 5 L/h case shown in Fig. 7C actually reaches pore volume saturation at the 8 minute mark. Despite the higher relative rate of fluid transfer, there existed a noticeable improvement in the content uniformity throughout the bed, particularly after the first few minutes when compared with the simulations from the 45% fill level, indicating a better axial mixing performance. Therefore, it was concluded that the mixing improvement seen at the 30% fill level compensated adequately for the increase in fluid spray rate per unit mass. This is further explored in the mixing section of this work. The experimental equivalence for the 30% fill level is shown in Fig. 8. This setup produced the most noteworthy experimental results. At the 1.5 L/h flow rate shown in Fig. 8A, the patterns exhibited

are very similar to the patterns shown for the simulations. However, at the higher 2.5 and 5 L/h flow rates, the pattern appears more non-traditional, and a corresponding increase in error is noted. Most striking, however, was the 5 L/h case, where a statistically significant, and very apparent right–left segregation pattern was evident, and was not clearly represented by the simulations. It was hypothesized that this segregation was caused by the high flow rates of fluid in the center of the bed producing a density gradient within the bed large enough to cause segregation. This type of segregation has been shown to be a common issue with granular mixers when different density particles are mixed [5], however, until now, it has not been reported for a dynamic, non-constant density system such as the one shown in this work. While it would appear that operating at this level of flow would be detrimental to the process, it is important to note that the segregation pattern and poor content uniformity quickly were adjusted to a more uniform distribution by the conclusion of the ten minute impregnation. This outcome would be more than adequate if the metal solution exhibited a slow kinetic adsorption, but would be severely detrimental to the process if the metal ions quickly adsorb to the support. In this case, although the liquid may be uniformly distributed, the metal would not be. Therefore, it was concluded that in most cases, lower flow rates in combination with the 30% (or lower) fill level will lead to the best overall content uniformity. As a result of the unique segregation findings shown experimentally, future work incorporating simulations needs to be implemented to capture density-based segregation phenomena in a non-constant density system.

Fig. 9. Five minutes of impregnation data at the 45% fill level and a flow rate of 5 L/h for A) fluid content in grams of fluid per gram of catalyst support and B) copper content in grams of copper per gram of catalyst support.


65

Fig. 10. Overlay of 3 min of experimental data and simulation data as a function of position for A) 1.5 L/h at 30% fill level, and B) 5 L/h at 30% fill level; dotted lines represent simulation data, error bars represent one standard deviation of the mean.

During an industrial scale dry impregnation, an aqueous metal solution is often sprayed as the fluid. As a result, several experiments were conducted using a copper (II) nitrate solution as the fluid phase in order to ensure that the previously obtained fluid patterns for water would translate into similar profiles for metal. The copper concentration was measured post-impregnation by stripping the copper from the alumina support using hydrochloric acid, and testing the resulting aliquot using UV-spectrophotometry. This experiment was tested using an extreme case of 5 L/h on the 45% fill level, with three replicates, where the error represents one standard deviation of the mean. These results are shown in Fig. 9, where Fig. 9A displays the concentration of water per unit catalyst, and the corresponding amount of copper impregnated is shown in Fig. 9B. As shown, the trends established by the water content match very closely with the trends exhibited by the resulting copper impregnated on the catalyst

support. This indicates that the performed experiments and developed model would accurately depict an actual metal impregnation in addition to only water transfer. Overall, the fluid content profiles per minute in the simulations appear to be more uniform and flat when compared to the experiments. However, the profiles have a very promising quantitative agreement when observed closely. Fig. 10 overlays the experimental data with the simulation data for two cases: Fig. 10A displays the 30% fill level and the low flow rate of 1.5 L/h, while Fig. 10B shows the 30% fill level at the high 5 L/h flow rate where the segregation phenomenon was clearly evident. In both figures, the simulations are represented by dotted lines, and the error of the experimental data is shown as one standard deviation of the mean for four replicates. At the lowest flow rate, the patterns exhibited by the simulations and the experiments are very similar. However, the peak point

66


Fig. 11. Comparison of overall fluid content within the granular bed as a function of spray rate; dotted lines are simulation data, error bars represent one standard deviation of the mean; A) 30% fill level, B) 45% fill level.

at the center is a bit higher in the experiments, suggesting that the transport of fluid from the center of the bed to the sides was slightly faster in the simulations than seen experimentally; this can be finely tuned on a per-application basis using the fluid transfer constant κ. Alternatively, the 5 L/h at 30% fill level case exhibited segregation phenomena, and therefore, was more closely examined. The qualitative differences between the patterns are self-evident, however it is worthwhile to notice that the model quantitatively accurately captures the amount of fluid present in the support, as well as a slight level of anisotropy. Therefore, further refinement of the model, including adjusting the constant κ for the rate of fluid transfer in Eq. (14) will remain a focal point for future studies. For the quantitative comparison of the DEM model and the experimental dry impregnations, the total fluid concentration was calculated per minute at the three aforementioned flow rates and at each of the two fill levels. The 30%, and the 45% fill levels are shown in Fig. 11A and B, respectively. In the graphs, the simulation values for fluid concentration are displayed as a dotted line, while the experimental values are displayed as a solid line. The error bars represent one standard deviation of the mean value for four replicates. It is important to note that for the 30% case (Fig. 11A), the 5 L/h flow rate results in the catalyst support reaching the pore volume shortly after 8 min, therefore, only the first 8 min is shown. There clearly exists a significant quantitative agreement between the overall fluid content

within the granular catalyst bed for both fill levels, as well as for all three flow rates. This is further shown with a parity plot shown in Fig. 12, where the experimental and simulation derived values for fluid concentration are shown. These results suggest that the DEM model, as well as the novel fluid transfer algorithm, were able to successfully describe the dry impregnation process. This opens the doors for further refinement and optimization of the dry impregnation process computationally, and enables modeling of alternate geometry mixers (e.g. V-blenders, rotating drums), extreme fill levels and flow rates, and additionally, the incorporation of mixing improvements such as baffles. 3.2. Mixing analysis Mixing within the dry impregnation was evaluated by taking the relative standard deviation of the fluid content at the seven axial points described in the previous section. Intuitively, the goal was to achieve high content uniformity, and well mixed system, which is generally exhibited by a relative standard deviation (RSD) of b0.10. It is important to note that mixing has been shown to not be a function of rotation speed in a double cone blender. Therefore, it was assumed that rotation speed was not a significant factor affecting the mixing results [6,22]. However, mixing under spray as a function of spray rate and fill level has not previously been quantified. In order


67

Approximately 6 min or 150 rotations were required to reach a well-mixed system at the 30% fill level, and more than 10 min or 250 rotations were required at the 45% fill level. An alternative method for the computational evaluation of mixing within the dry impregnation system was to color-tag the particles in various configurations and subsequently evaluate the relative standard deviation of the color-tagged particles, a significant advantage resulting from the development of a DEM model system. Three configurations were evaluated for mixing, including side–side, which represents the axial mixing of the system, top–bottom, and front– back mixing. The three starting configurations are shown at the top of Fig. 14. Each configuration was simulated for 5 min of spray under 1.5 and 5 L/h spray-rates for the 30% fill level. The relative standard deviations of the seven axial bins for each system are shown in Fig. 14. The striking difference between the side–side versus the other two configurations was a strong indicator of poor axial mixing

Fig. 12. Parity plot comparing the experimental values and the computed values for fluid concentration in the catalyst support particles.

to accomplish this goal, the relative standard deviation of the fluid concentration within the seven predetermined points along the axis were computed and compared for the two fill levels and the different flow rates. The results of each experiment/simulation were averaged between experiments and are shown in Fig. 13, where the error bars represent one standard deviation between four experiments. Clearly, it is shown that the 30% fill level reached a well-mixed RSD of less than 0.10 at a significantly faster rate than the 45% fill level. As described in the previous section, the 30% fill level was shown to mix significantly better despite the larger ratio of spray-rate to mass of catalyst support. It should be noted that no fluid was sprayed in the 45% case after 10 min, yet a non-trivial amount of additional time was required to reach the same level of RSD exhibited by the 30% case. While it was known previously that lower volumes of granular material mix faster in a double cone geometry blender, this effect had not been previously established for a system under fluid spray and particle–particle fluid transfer. It is important to note that no statistically significant differences to the overall mixing of the powder bed were observed between the different spray flow rates.

Fig. 13. Relative standard deviation of the fluid concentration at all sample points for experimental dry impregnations at the 45% and 30% fill levels, error bars represent one standard deviation of the mean; spray was ceased at t = 10 min for both fill levels.

Fig. 14. Simulation derived mixing analysis using relative standard deviation as a function of color-tagged particles throughout the granular bed for A) 1.5 L/h at 30% fill level, and B) 5 L/h at 30% fill level; color-tagged initial conditions are pictured for the side– side (axial mixing), front-back, and top-bottom arrangements.

68


generally inherent of double cone blenders. Specifically, the system required nearly 5 min, or 125 rotations for the side–side configuration to settle into a b0.10 relative standard deviation, while the top–bottom and front–back systems exhibited a b0.10 RSD after less than 25 rotations (1 min of operation). Moreover, the mixing in the two graphs representing the low and high flow rates of 1.5 and 5 L/ h uncovered very similar profiles, indicating that the higher flow rate did not significantly affect the mixing of the pre-colored particles, which is in agreement with the experimental mixing study. It should be noted that a similar trend was found at the 45% fill level, but is not shown. The key improvements to the model to accurately parallel the experimental data are to i) include a polydisperse catalyst support media, as no real system contains perfect, identical 4.7 mm spheres, ii) improve the fluid mass transfer rate between particles, and iii) simulating more extreme examples of high flow rate fluid transfer in order to determine the presence of granular segregation as a function of particle density gradients. 4. Conclusions In this work, a novel fluid-transfer algorithm was implemented into a discrete element model of a rotating granular catalyst bed to model a dry catalyst impregnation process. The DEM model was evaluated and compared with parametrically and geometrically equivalent experiments using two fill levels (30% and 45%), and three liquid flow rates (1.5, 2.5, and 5 L/h), a constant rotation speed of 25 rpm, and a single spray nozzle centrally located. It was found that the simulation and corresponding algorithm was able to accurately represent the double cone dry impregnation system for the overall fluid concentration within the catalyst support bed as a function of position and time. Consequently, several key open questions regarding the mixing and content uniformity of a dry impregnation were answered. Specifically, it was found that lower fill levels and lower spray rates increased the content uniformity of the resulting particles. Furthermore, it was found, non-intuitively, that only a limited number of rotations following the completion of spray were required to adequately mix the granular bed following spray. Industrially, this would result in a decrease in traditional processing times that are often rotated for longer and unnecessary periods of time post-spray. Finally, it was further illustrated that axial mixing within a double cone blender is poor, especially when liquid is sprayed at the center of the bed suggesting that alternative spray configurations such as on the sides of the vessel would result in better content uniformity. It was determined that side–side segregation phenomena was more evident in the experiments than in the simulations, and therefore, it was hypothesized that this segregation may be a result of any combination of cohesion, liquid bridging, and most likely, density gradient effects within the granular bed. Future studies utilizing this model include a polydisperse catalyst support media, refine the fluid mass transfer rate between particles, and simulate more extreme examples of high flow rate fluid transfer in order to further explore granular segregation. It is the authors' hope that this work has shown that the application of newly developed simulation tools may be extremely valuable to the catalyst industry for improving product content uniformity and limiting processing times. Notation C Eeff ε εS εr εres FTotal FContact FBody

fluid concentration, gfluid/gcatalyst Young's modulus, N/m 2 coefficient of restitution coefficient of static friction coefficient of rolling friction coefficient of restitution total force, N contact force, N body force, N

Geff γt γn Ii kn kt κ m mi,j n N Qsqr QTr Reff Ri RSD ρ S δn δ_ n v t vrel € Χ xi

shear modulus, N/m 2 tangential damping coefficient normal damping coefficient moment of inertia Hertzian normal stiffness coefficient tangential stiffness coefficient proportionality constant mass of fluid droplet, g mass of support particle, g number of samples number of fluid droplets mass flow rate of fluid, g/s fluid transfer coefficient, g effective radius, m radius of particle, mm relative standard deviation density of fluid droplet, g/m 3 index of segregation particle overlap rate of deformation, 1/s volume of fluid droplet, m 3 relative velocity of tangential impact, m/s particle acceleration, m/s 2 particle position

Acknowledgments The authors would like to acknowledge Exxon Mobil for loaning us the Patterson-Kelly double-cone mixer, Saint-Gobain Norpro for donating alumina catalyst support, and the Catalyst Manufacturing Science and Engineering Consortium at Rutgers University for funding and support. References [1] A. Lekhal, B.J. Glasser, J.G. Khinast, Influence of pH and ionic strength on the metal profile of impregnation catalysts, Chemical Engineering Science 59 (5) (2004) 1063–1077. [2] A. Lekhal, B.J. Glasser, J.G. Khinast, Impact of drying on the catalyst profile in supported impregnation catalysts, Chemical Engineering Science 56 (15) (2001) 4473–4487. [3] X. Liu, J.G. Khinast, B.J. Glasser, A parametric investigation of impregnation and drying of supported catalysts, Chemical Engineering Science 63 (18) (2008) 4517–4530. [4] A.J. van Dillen, R.J.A.M. Terörde, D.J. Lensveld, J.W. Geus, K.P. de Jong, Synthesis of supported catalysts by impregnation and drying using aqueous chelated metal complexes, Journal of Catalysis. 216 (1–2) (2003) 257–264. [5] A. Alexander, F.J. Muzzio, T. Shinbrot, Segregation patterns in V-blenders, Chemical Engineering Science 58 (2) (2003) 487–496. [6] A.W. Chester, J.A. Kowalski, M.E. Coles, E.L. Muegge, F.J. Muzzio, D. Brone, Mixing dynamics in catalyst impregnation in double-cone blenders, Powder Technology 102 (1) (1999) 85–94. [7] M. Lemieux, G. Léonard, J. Doucet, L.A. Leclaire, F. Viens, J. Chaouki, et al., Largescale numerical investigation of solids mixing in a V-blender using the discrete element method, Powder Technology 181 (2) (2008) 205–216. [8] D. Brone, C. Wightman, K. Connor, A. Alexander, F.J. Muzzio, P. Robinson, Using flow perturbations to enhance mixing of dry powders in V-blenders, Powder Technology 91 (3) (1997) 165–172. [9] P.E. Arratia, D. N-h, F.J. Muzzio, P. Godbole, S. Reynolds, A study of the mixing and segregation mechanisms in the Bohle Tote blender via DEM simulations, Powder Technology 164 (1) (2006) 50–57. [10] L. Barthe, S. Desportes, M. Hemati, K. Philippot, B. Chaudret, Synthesis of supported catalysts by dry impregnation in fluidized bed, Chemical Engineering Research and Design 85 (6) (2007) 767–777. [11] A. Mehrotra, F.J. Muzzio, Comparing mixing performance of uniaxial and biaxial bin blenders, Powder Technology 196 (1) (2009) 1–7. [12] B. Chaudhuri, A. Mehrotra, F.J. Muzzio, M.S. Tomassone, Cohesive effects in powder mixing in a tumbling blender, Powder Technology 165 (2) (2006) 105–114. [13] A. Faqih, B. Chaudhuri, A.W. Alexander, C. Davies, F.J. Muzzio, M. Silvina Tomassone, An experimental/computational approach for examining unconfined cohesive powder flow, International Journal of Pharmaceutics 324 (2) (2006) 116–127. [14] S.S. Manickam, R. Shah, J. Tomei, T.L. Bergman, B. Chaudhuri, Investigating mixing in a multi-dimensional rotary mixer: experiments and simulations, Powder Technology 201 (1) (2010) 83–92.

F.S. Romanski et al. / Powder Technology 221 (2012) 57–69 [15] E. Sahni, R. Yau, B. Chaudhuri, Understanding granular mixing to enhance coating performance in a pan coater: experiments and simulations, Powder Technology 205 (1–3) (2011) 231–241. [16] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies, Geotechnique 29 (1) (1979) 47–65. [17] H. Kruggel-Emden, E. Simsek, S. Rickelt, S. Wirtz, V. Scherer, Review and extension of normal force models for the discrete element method, Powder Technology 171 (3) (2007) 157–173. [18] J. Shäfer, S. Dippel, D.E. Wolf, Force schemes in simulations of granular materials, Journal de Physique I. 6 (1) (1996) 5–20.

69

[19] H. Kruggel-Emden, M. Sturm, S. Wirtz, V. Scherer, Selection of an appropriate time integration scheme for the discrete element method (DEM), Computers and Chemical Engineering 32 (10) (2008) 2263–2279. [20] R.D. Mindlin, H. Deresiewicz, Elastic spheres in contact under varying oblique forces, Journal of Applied Mechanics 20 (1953) 327. [21] R.D. Mindlin, Compliance of elastic bodies in contact, Journal of Applied Mechanics 16 (1949) 259–268. [22] D. Brone, F.J. Muzzio, Enhanced mixing in double-cone blenders, Powder Technology 110 (3) (2000) 179–189.