Cluster characteristics of continuous size distributions and binary mixtures of Group B particles in dilute riser flow

Chemical Engineering Journal 178 (2011) 348–358

Contents lists available at SciVerse ScienceDirect

Chemical Engineering Journal journal homepage: www.elsevier.com/locate/cej

Cluster characteristics of continuous size distributions and binary mixtures of Group B particles in dilute riser flow Jia Wei Chew a , Drew M. Parker a , Ray A. Cocco b , Christine M. Hrenya a,∗ a b

Department of Chemical and Biological Engineering, University of Colorado at Boulder, Boulder, CO 80309, United States Particulate Solid Research Incorporated, Chicago, IL 60632, United States

a r t i c l e

i n f o

Article history: Received 26 May 2011 Received in revised form 5 October 2011 Accepted 6 October 2011 Keywords: Cluster Fiber optic probe Video images Wavelet decomposition CFB riser

a b s t r a c t Gas–solids circulating fluidized bed (CFB) experiments have been carried out, with a focus on understanding the impact of polydispersity on cluster characteristics in a dilute riser. Two categories of polydispersity were studied: continuous particle size distributions (PSDs) of varying distribution widths, and binary mixtures of varying compositions. Video images show that particle clusters exist even in these very dilute systems (solid loading m = 0.03–0.29). Local measurements were acquired using two independent instruments – a fiber optic probe and a high-speed video camera – with the cluster trends obtained from both are in qualitative agreement with each other. Cluster characteristics extracted from the measurements include appearance probability, duration and frequency. Results show that: (i) axial position is the strongest influence on radial profiles of cluster duration and frequency, but has negligible effect on cluster appearance probability, (ii) profile shapes are invariant with height, although magnitudes of cluster duration and frequency changes with height, and (iii) effects of both the widths of continuous PSDs and the compositions of binary mixtures are observed at the riser bottom for cluster duration and at the riser top for cluster frequency, though are insignificant for the appearance probability. © 2011 Elsevier B.V. All rights reserved.

1. Introduction Extensive application of the circulating fluidized bed (CFB) technology started in the 1980s [1], and have since become an increasingly important unit operation, with examples of principal applications including fluid catalytic cracking (FCC) [2,3], gasification and combustion [4–6]. Such gas–solid suspensions are inherently unstable, leading to transient meso-scale structures manifesting as clusters of particles [7–16]. Because clusters affect the hydrodynamics (e.g., gas–solids contact time, elutriation rate) and hence impact reactor performance (e.g., productivity, rate of reaction, heat transfer), an enhanced understanding of the behavior of such flow structures in the CFB riser is necessary. Previous experimental and modeling studies focused on clustering behavior in monodisperse systems, but since polydispersity is ubiquitous, their associated effects warrant attention [17–20]. Accordingly, the focus of this experimental study is on the impact of polydispersity on the cluster characteristics of Geldart [21] Group B particles. Polydispersity is prevalent in industrial processes. For example, coal feedstock and fluid cracking catalyst (FCC) contain a wide range of sizes, and biomass gasification involves two or more materials of different particle size and material density. However,

∗ Corresponding author. Tel.: +1 303 492 7689; fax: +1 303 492 4341. E-mail address: [email protected] (C.M. Hrenya). 1385-8947/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.cej.2011.10.020

a predictive understanding of the impact of polydispersity remains elusive [17–20], which can lead to poor performance and undesirable flow behavior [18,22,23]. The only work to date on the impact of polydispersity on clusters [24] indicates that, similar to trends shown for monodisperse materials [25], cluster characteristics vary most with the local position in the riser relative to operating condition and/or material type (i.e., monodisperse materials of different particle size and/or material density, or different types of polydispersity). Yet, this study [24] also shows that the effect of polydispersity is clearly noticeable on cluster measurements, as neither of the two binary mixtures investigated mimics the constituent monodisperse components [24]. Also, the operating conditions were shown to exert influence on the cluster behavior of the continuous PSD but not on the binary mixtures, although the cause may be either due to the wider range of operating conditions or an inherent difference between the two categories of polydispersity (as has been documented before for other fluidization characteristics [26–32]). What has not been investigated to date is the effect of varying widths of continuous PSDs and compositions of binary mixtures on cluster characteristics. The current dilute CFB riser study is divided into two categories: continuous PSDs and binary mixtures (two species with different particle size, dave , and material density, s ). Since characterization of the clusters over limited portions of the riser may lead to contradictory trends [25], this work aims to generate a complete cluster dataset spanning multiple regions throughout the riser.

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

349

Nomenclature Aj Aj,ave CFB dsm Dj EDj EJ,all EDj /EJ,all fm fn Gs h H h/H j J m PSD r R r/R Us Ut g s 

jth scale of wavelet-decomposed approximation average of the jth scale of wavelet-decomposed approximation circulating fluidized bed Sauter-mean particle diameter jth scale of wavelet-decomposed detail energy of Dj energy of all the wavelet-decomposed approximations and details normalized wavelet energy mass-based frequency distribution number-based frequency distribution solid mass flux height at which measurement was taken total riser height dimensionless height wavelet-decomposition scale maximum scale of wavelet-decomposition solid loading particle size distribution radius at which measurement was taken total riser radius dimensionless radius superficial gas velocity particle terminal velocity gas density solid density standard deviation of PSD Fig. 1. Schematic diagram of experimental set-up.

Specifically, the objectives are to experimentally investigate the impact of changing the widths of continuous PSDs and compositions of binary mixtures on cluster characteristics. Two independent instruments, namely a fiber optic probe and a highspeed video camera, were used to obtain local cluster information. As part of this work, the answers to two questions are sought. First, keeping in mind that dilute suspensions as in this work are known to be more homogeneous (i.e., less tendency to cluster) [7,10,13], do the cluster characteristics differ from that reported for a denser riser [24,25]? Second, in view of the different qualitative nature of continuous PSDs and binary mixtures reported for other flow characteristics of bubbling fluidized beds [26–30] and CFB risers [31,32], do clusters behave differently between these two categories of polydispersity in a dilute CFB riser? The presence of clusters for the dilute systems studied here (solid loading, m = 0.03–0.29) is documented via both highspeed imaging and fiber optic probe. Local cluster characteristics extracted from the measurements include appearance probability, duration and frequency. Results show that: (i) riser axial position is a key influence on cluster duration and frequency, but has negligible effect on cluster appearance probability, (ii) profile shapes are invariant with height, although magnitudes of cluster duration and frequency changes with height, (iii) effects of both widths of continuous PSDs and compositions of binary mixtures are observed at the riser bottom for cluster duration and riser top for cluster frequency, though are insignificant for appearance probability. 2. Experiment description 2.1. Experimental apparatus Experiments were conducted in a circulating fluidized bed (CFB), shown in Fig. 1. The riser section is a PlexiglasTM column (18.4-cm

in inner diameter and 4-m tall) with a blind-tee exit at the top. The distributor at the bottom of the riser is a Mott Corporation 316 stainless steel sintered porous plate, of Media Grade 100 and thickness of 2.4 mm. A Yaskawa V7 variable frequency drive, controlled by a National Instruments LabVIEW system (version 7.1), directs a Fuji Electric VFD5 regenerative blower. The superficial air velocity (Us ), which is reported at local atmospheric conditions (air with a 0.97 kg/m3 density and 1.85 × 10−5 Pa s viscosity), was determined via a Lambda Square Oripac 4150-P orifice plate flow meter, which is located upstream of the plenum. The relative humidity (RH) in the CFB is enhanced by an Air-O-Swiss model AOS 7144 humidifier placed at the inlet of the blower. To reduce electrostatics [33–36], the relative humidity is maintained above 40% for all experiments. The operating air temperature and relative humidity in the plenum are measured by means of an Omega HX93AV-RP1 probe, with a temperature range of −4 to 171 ◦ C and relative humidity range of 0–100%, inserted into the plenum. A cyclone located after the riser exit enables collection of the solids, which are returned back to the riser through an aerated L-valve. The pressure drops across the orifice plate flow meter, across the distributor plate, and along the entire riser are measured using Orange Research 20100 Series low-differential pressure transmitters with the ±0.2% accuracy option. All temperature, relative humidity, superficial velocity and pressure data are recorded throughout the experiments with the LabVIEW data acquisition system. Seven sample ports are available approximately equally spaced along the riser height for insertion of the fiber optic probe or video camera borescope, which are used for characterizing clusters. Up to three sample ports each 90◦ azimuthally apart on the horizontal plane are available at each of these axial positions to detect potential flow asymmetries.

350

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

Table 1 Experimental parameters for continuous PSDs. Material

Sand

s (kg/m3 ) dsm (␮m) Ut ,dsm (m/s) [45] /dsm (%) Us (m/s) m

2650 196 1.6 10, 25, 40, 65 1.7 0.13–0.29

2.2. Materials and operating conditions Two categories of polydisperse materials within the Geldart Group B classification are of interest: continuous particle size distributions (PSDs) and binary mixtures. The relevant material properties for the continuous PSDs, namely material density (s ), Sauter-mean diameter (dsm ), width of PSD (denoted /dsm , where  is the standard deviation of the mass-based PSD), and singleparticle terminal velocity (Ut ) are listed in Table 1. As for the binary mixtures (Table 2), the mass-based average particle size, dave , is more relevant. Also listed in Tables 1 and 2 are the operating conditions, namely superficial gas velocity, Us , and solid loading (m = Gs /Us g , where m is solid loading, Gs is total mass flux and g is air density). For all of the experiments, several parameters are kept constant for direct comparison among the mixtures, namely (i) CFB inventory (8 kg of particles) since bed depth in a fluidized bed is correlated with elutriation rate [37] (part of a corollary investigation [32]) and cluster size [38], and (ii) Us within each of the two sets (i.e., continuous PSDs and binary mixtures) of experiments since the driving force for elutriation is linked to Us –Ut or Us /Ut [39–44]. The parameters varied are widths of the continuous PSDs (/dsm ) and compositions of the binary mixtures (Tables 1 and 2, respectively). Finally, since a dilute riser is the focus of this work, solid loadings are in the ranges of 0.13–0.29 for the continuous PSDs (Table 1) and 0.03–0.25 for the binary mixtures (Table 2). Pertaining to the continuous PSDs (Table 1), the distributions are prepared using sand particles with a wide range of particle sizes from Agsco Corporation. The equation defining the mass-based frequency distribution (fm ) of Gaussian distributions is



fm,

Gaussian (x) =

1 (x − x) exp − √ 2 2  2

2



(1)

where  is the standard deviation of the mass-based PSD, x is particle diameter, and x is the arithmetic mean of the mass-based PSD. Analogously, the mass-based lognormal distribution is defined as



fm,

1 (ln(x) − )2 exp − √ log normal (x) = 2 2 x 2



(2)

whereby the natural logarithm of this distribution is a Gaussian PSD with arithmetic mean  and standard deviation   . In this work, the Sauter-mean diameter dsm is kept constant and the widths (denoted /dsm ) are varied. The dsm is used as opposed to other characteristic diameters since it is the most physically relevant. Namely, Sauter-mean diameter is the ratio of volumebased to surface-based diameters, the former is proportional to the gravitational force and the latter is proportional to the drag force. Table 2 Experimental parameters for binary mixtures. Material s (kg/m ) dave (␮m) Ut (m/s) [45] Mass % Us (m/s) m 3

Glass

Polystyrene (PS)

2500 165 1.2 0, 25, 50, 75, 100

1050 327.5 1.6 100, 75, 50, 25, 0 1.5 0.03–0.25

Fig. 2. Continuous PSDs invtigated. Lines denote PSDs as per exact lognormal equation (Eq. (2)), while discrete points denote experimental distribution prepared with sieves.

To compute the Sauter-mean diameter (dsm ) [43], the mass-based distributions (fm ) in Eqs. (1) and (2) are first converted to the corresponding number-based distributions (fn ), and then dsm is  determined as x3 fn / x2 fn . To prepare the continuous PSDs, sand acquired from Agsco Corporation was classified into different sizes using various Fisherbrand U.S. standard brass sieves. The narrowest distribution investigated is /dsm = 10%, in which particles largely fall within one sieve cut (180–212 ␮m). The widest distribution examined is /dsm = 65%, above which would require a considerable amount of Group A particles, which is avoided since the focus here is on Group B materials. Since Gaussian distributions (Eq. (1)) are restricted to /dsm < 30% (greater values of /dsm involve unphysical negative diameters), lognormal distributions (Eq. (2)) are used to allow for wider PSDs to be investigated. Also, since Gaussian and lognormal distributions are similar in shape for /dsm < 30% [45], the two narrowest distributions investigated (namely, /dsm = 10%, and 25%) can be referred to as either Gaussian or lognormal, but will be referred to as lognormal here onward. Fig. 2 shows the exact lognormal distributions (Eq. (2)) as solid lines and corresponding sieve fractions used in the experiments as discrete points. Here, fm refers to the mass fraction of each sieve cut normalized with respect to the bin width (i.e., x) of a given sieve cut. The choice of Sauter-mean diameter, dsm , is made high enough such that particles still largely belong to Geldart Group B even for the widest PSD (Fig. 2), while low enough such that the corresponding single-particle terminal velocity (Ut ) is lower than the maximum limit of superficial air velocity (Us,max ) that the blower can generate to ensure elutriation. Because Us,max is 1.7 m/s, the largest sand particle that can be ideally elutriated is of 202 ␮m. Due to the restriction of sieve sizes available, however, the dsm was set to 196 ␮m, which is the average size between two sieve cuts (180 and 212 ␮m) and has a Ut of 1.6 m/s (Table 1). For all the experiments involving continuous PSDs, the value of Us = 1.7 m/s was used. Finally, because it was observed that sand attrits over time, a fresh PSD was prepared after every three cumulative hours of experimental runs in order to preserve the integrity of the PSD. Specifically, preliminary checks on 8 kg of particles which fall between the 180 and 212 ␮m sieves showed that only 85 mass% (i.e., greater than acceptable tolerance of 10 mass%) of the bed retained the original size range at the end of 3 h.

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

351

Fig. 3. Fiber optic voltage traces at Us = 1.7 m/s, h/H = 0.43 and r/R = 0.96 for (a) background (i.e., absence of solids), and (b) continuous PSD with o/dsm = 40%, and (c) waveletbased threshold superimposed on continuous PSD with o/dsm = 40%.

Binary mixtures (Table 2) were prepared in various compositions by mixing glass and polystyrene (PS) particles. Glass beads purchased from Grainger were sieved to obtain only sizes that fall between the 150 and 180 ␮m sieves, while PS beads purchased from of Glen Mills Incorporated were sieved to restrict sizes to those between the 300 and 355 ␮m sieves. The terminal velocity (Ut ) of the glass beads is lower than the implemented superficial gas velocity (Us ), whereas the terminal velocity of PS is higher than the superficial gas velocity. Accordingly, the binary mixtures investigated encompass the entire range of compositions, including (i) monodisperse PS (i.e., 100 mass% PS), (ii) 25 mass% glass (and 75 mass% PS), (iii) 50 mass% glass, (iv) 75 mass% glass, and (v) monodisperse glass (i.e., 100 mass% glass). The value of superficial gas velocity used in the suite of binary mixture experiments is chosen to be high enough to enable elutriation of the monodisperse PS but low enough to restrict the total elutriation flux of the monodisperse glass to a practical level. It was found experimentally that Us = 1.5 m/s allowed for some elutriation of the monodisperse PS while also restricting the level of elutriation of monodisperse glass to reasonable levels. 2.3. Instrumentation and method of analysis Two independent instruments were used to characterize clusters in the CFB riser, as described below. 2.4. Fiber optic probe and wavelet decomposition The fiber optic probe, fabricated by Particulate Solid Research Incorporated (PSRI), is a stainless steel tubing (outer diameter = 0.013 m) housing two fibers, each with a diameter of 2000 ␮m. One fiber is used to transmit infrared from a light-emitting diode (LED) source, while the other is the receiver conduit. The two fibers are aligned to converge towards the probe tip in order to reduce the sampling volume, thereby reducing unnecessary scattering of the light source and enhancing sensitivity [46,47]. In this work, fiber optic data was collected at 1000 Hz [25,48] for 30 s at each local riser position, namely seven axial positions and six radial positions from center to wall. Two measurements were taken at each position. The key operating principle of the fiber optic probe is that it translates the light received by the receiver fiber into a voltage value, which gives an indication of solids concentration in the riser. It is worthwhile to note that the fiber optic probe records a higher voltage when the receiver fiber detects more light, either when (i) the infrared emitted from the transmitter fiber is blocked, as occurs when the probe is surrounded by a denser (particle-rich) phase or (ii) the receiver fiber receives more light through the surroundings, as occurs when a highly reflective surface is across the probe tip. Accordingly, to improve sensitivity of the receiver fiber to the infrared from the emitter fiber, the lights in the lab were turned off when fiber optic data was acquired, although some stray

light was still present. Darker surroundings are implemented for two reasons: (i) the riser is made of Plexiglas, which implies the riser interior is flooded with light and (ii) a dilute system warrants increased sensitivity of the probe. As examples of the fiber optic data, voltage traces obtained in the riser of the CFB are shown for the cases of background (i.e., absence of solids) in Fig. 3a and in the presence of solids flow in Fig. 3b, both as functions of time. Not surprisingly, the average voltage of the background trace (Fig. 3a) is higher than the solids trace (Fig. 3b), which is due to the presence of a reflective metallic plug on the sample port on opposite wall from the probe. In some previous works, the background signal was taken either in a black box [49] or on a matte black surface [50] to allow for subsequent calibration of voltage to solids concentration. Such a calibration is not used here due to the error-prone nature of the calibration methods [49–51]. Instead, transient information on relative concentration is obtained directly from the raw signal. Specifically, the amplitude of fluctuations about 5.6 V in the background signal (Fig. 3a) is quite small due to the presence of only background noise. On the other hand, the amplitude of fluctuations in Fig. 3b is much greater (ranging from 5.45 to 5.65 V), which results from the dynamic nature of solids concentration (i.e., clustered and non-clustered regions). Thus, the transient voltage traces are used for cluster identification, as detailed below. The threshold used for cluster identification should have a physical significance, and wavelet decomposition [52,53] facilitates this via the demarcation of the different scales (namely, micro-scale, meso-scale, macro-scale) in fluidized bed systems [25,48,54–57]. More explicitly, the micro-, meso-, and macro-scales have been identified as corresponding physically to background noise, clusters, and equipment, respectively [25,48,54–57]; a more detailed explanation of the implementation of this method is described elsewhere [25]. In brief, wavelet decomposition [52,53] provides a means of extracting different frequency ranges of the transient fiber optic voltage signal by repeatedly breaking down the signal into higher-frequency details (D) and lower-frequency approximations (A). At the first scale of decomposition (Scale 1), the signal of N Hz is divided into the first scale of approximation (A1 ) and the first scale of detail (D1 ), whereby A1 contains the lower half of the frequency range and D1 contains the higher half. With a further increase in scale from j to j + 1, each approximation Aj is subsequently decomposed into low-frequency Aj+1 and high-frequency Dj+1 signals. The higher the scale of decomposition, the lower the frequency information is, and vice versa. Accordingly, Fig. 4a and b is the plots of normalized wavelet energy (EDj /EJ,all ) versus wavelet decomposition scale, generated using the following equations:

N    Dj (t) EDj = t=1

2

(3)

352

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

Fig. 4. Normalized wavelet energy plots for lognormal PSD with Us = 1.7 m/s, a/dsm = 10%, h/H = 0.30, and (a) r/R = 0 and (b) r/R = 0.83. Subplots (a) and (b) are obtained from voltage data using the fiber optic probe, while (c) and (d) are obtained from area % data using video images.

N    Aj (t) − Aj,ave  EAj =

2

(4)

t=1

EJ,all = EA,J +

J 

EDj

(5)

j=1

where E is the energy of the signal, N is the total number of data points, J is the maximum scale of decomposition (in this work, J = 13 is used for data acquired at 1000 Hz) [25], j is the scale of decomposition, Dj is the detail signal at the jth scale, Aj is the approximation signal at the jth scale, Aj,ave is the average value of the Aj signal, and t is the time index of collected data. As detailed in a previous work [25], scale 11 in Fig. 4a and b is used as threshold for cluster identification since it marks the delineation between meso- and macro-scale [24,25,48]. The wavelet toolbox [58] in Matlab is used for the wavelet decomposition and associated analysis. Correspondingly, Fig. 3c shows the wavelet-based threshold superimposed on the transient fiber optic voltage signal, and the clusters are identified as data points above the threshold. With wavelet decomposition of the transient waveform, various cluster characteristics can be obtained including (i) the appearance probability of the cluster, which is the number of data points above the threshold divided by total number of data points, (ii) the duration of the cluster, which is the average number of consecutive data points above the threshold, and (iii) the frequency of cluster appearances, which is the number of clusters (i.e., number of segments of consecutive data points above the threshold) detected per unit time. 2.5. Video and image analysis High-speed video images were collected using a Kodak Motion Corder Analyzer Model SR-Ultra (PS-110) attached to a borescope attached to enable measurement within the riser [38]. The borescope is an Olympus R100-038-000-50 Industrial Rigid Borescope, with a depth of field of 5 mm to infinity. Some modifications were made to the borescope to enhance the quality of the images. To correct for the distance between the borescope tip

and the focal length, the borescope is fitted with a 6-mm diameter optical spacer (Melles Griot), which prevents particles closer than the focal length from adversely affecting the images. The spacer is further secured using a stainless steel guard collar since the sand and glass particles used in this work are highly abrasive. A xenon light source with an Olympus Liquid-Filled Light Guide is used to supply light through the borescope probe. Since increased illumination enhances image quality, two floodlights were additionally positioned external to the riser and directed at the borescope tip. Borescope measurements were taken at the same axial and radial positions as per the fiber optic probe. Video images were collected at 2500 frame per second (i.e., 2500 Hz) for a duration of 2 s at each local riser position. The file size for 2 s of video data is 800 MB. Accordingly, in light of the number of measurements required and the memory limitations of the laptop used, only 2 s of video is acquired per measurement as opposed to the 30 s sampling time for the fiber optic probe. VLC media player is used to convert the video files from avi to mp4, and then QuickTime 7 Pro converts the mp4 files to jpg image sequences. Subsequently, an algorithm written in Mathematica 7.1 facilitates analysis of the jpg image sequences. First, each raw jpg image (i.e., unprocessed video image) is fine-tuned to improve the contrast between solids and background. Afterward, the image is binarized (i.e., pixels above a specified threshold brightness are designated as 1, and pixels below as 0). In the corresponding images, the solids appear white and the background appears black (Fig. 5). Examples of four raw images obtained at random times and at three heights are illustrated in Fig. 5a (first column of Fig. 5). In addition, Fig. 5b and c depicts the progressively processed images after contrast is enhanced and after binarization, respectively. After the images are binarized, the area percent occupied by solids is computed by dividing the total number of white pixels with the total number of pixels per image. Fig. 6 shows examples of area percent versus time plots, which are similar in nature to the fiber optic voltage traces in Fig. 3b; i.e., clusters appear as upward spikes. The trace obtained at dimensionless axial height (h/H, where h is the height at which measurement is taken and H is the total height of the riser) of 0.30 (Fig. 6a) has an overall higher area percent than at h/H = 0.90 (Fig. 6b), which is expected since the riser

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

353

Fig. 5. At three riser heights (h/H = 0.30, 0.55, 0.90) at Us = 1.7 m/s, samples at four random time instances of (a) raw, (b) contrast enhanced, then (c) binarized video images for lognormal distribution with o/dsm = 10% at r/R = 0.

Fig. 6. Area % versus time traces from video images for lognormal PSD with o/dsm = 65% and Us = 1.7 m/s at r/R = 0 and (a) h/H = 0.30 and (b) h/H = 0.92.

has a higher solids concentration at h/H = 0.30 (Fig. 5). Thus, with a suitable threshold, clusters can be identified. Hence, similar to the fiber optic traces, wavelet decomposition was carried out on the area % traces (Fig. 6), and the resulting EDj /EJ,all plots are shown in Fig. 4c and d. These plots are noisier and dissimilar in shape than those for the fiber optic voltage traces in Fig. 4a and b. Due to the shorter duration of the area percent data (2 s as opposed to 30 s for fiber optic traces), no clear demarcation between micro-, meso-, macro-scales is demonstrated in the EDj /EJ,all plots (Fig. 4c and d). Similarly, if only 2 s of data obtained from the fiber optic is used for the corresponding wavelet analysis (as opposed to the total of 30 s which is available), the resulting EDj /EJ,all plots (not shown) also take on a noisy appearance similar to Fig. 4c and d. These results suggest that 2 s is an insufficient duration for cluster analysis via wavelet decomposition, and that a longer duration such as the 30 s used in this work for the fiber-optic measurements is necessary if wavelet decomposition is to be used. Since wavelet decomposition is infeasible for the short duration of video data, another way to determine the threshold for cluster identification is necessary. An inspection of the cluster characteristics obtained using the fiber optic probe (as seen in Figs. 8 and 10 below, for example) reveals that the cluster appearance probability (when integrated across cross-section and averaged across heights) is approximately 50% for all cases. Accordingly, a 50th

percentile threshold is used for cluster identification in the video images (previous works have also used a constant threshold for cluster identification [59], and results compared well to the wavelet based analyses [48]). Because the main objective here is to compare the qualitative nature of the cluster trends between two independent instruments (i.e., fiber optic probe and video camera), the specific choice of such a threshold is not as critical. In contrast to the wavelet decomposition method used for the fiber optic probe in which the threshold is dynamic and changes with time (Fig. 3c), a time-independent 50th percentile threshold is imposed on each of the area percent traces obtained from video (Fig. 6). However, the setting of a 50th percentile value as threshold implies that the cluster appearance probability is 50% (input to analysis), and hence only cluster duration and frequency will be obtained as outputs from analysis of video images, whereas the appearance probability is an additional output when wavelet analysis is used for the fiber optic data. 3. Results and discussion 3.1. Continuous PSD Before presenting the cluster measurements obtained by the fiber optic probe and video camera, it is worthwhile to first present

354

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

additional imaging evidence of clusters in dilute flows, especially since dilute suspensions are known to be more homogeneous, i.e., less likely to cluster [7,10,13]. It is perhaps not surprising that dilute conditions in this work (Tables 1 and 2) leads to clusters, as the clustering phenomenon has been reported for a similarly dilute work at m = 0.1 [60]. Photos of clusters taken with a SLR camera (Canon 40D with EFS 15–85 mm f/3.5–5.6 IS UM and EF 100–400 mm f/4.5–5.6L IS USM lenses) are presented in Fig. 7, wherein darker colors imply higher solids concentration (i.e., clusters). The SLR has a larger field of view than the video camera/borescope (Fig. 5), and hence allows for a macroscopic visualization of the clusters. For the photos shown here, the SLR camera is located at the wall and thus is non-intrusive (unlike fiber optic probe and video camera techniques). Comparing Fig. 7a and b (which has a time lag of 0.1 s between them) reveals an upward-moving cluster disappearing from field of view (dissipating and/or moving into page). Fig. 7c and d (also with a time lag of 0.1 s in between) portray an upward-moving cluster. Finally, Fig. 7e and f (similar time lag of 0.1 s) portray a downward-moving cluster that is disappearing from field of view. These photos provide evidence of clusters even under the most dilute conditions examined here (m = 0.13 in Fig. 7). Prior to inspecting the cluster results obtained with the fiber optic probe for the continuous PSDs (Table 1), the matrix of subplots in Fig. 8 warrants explanation. Each of the three columns of subplots depicts one of the cluster characteristics obtained from the wavelet analysis (i.e., appearance probability, duration and frequency), with the y-axis ranges kept constant within each column for straightforward comparison of axial trends throughout the riser. Each of the seven rows represents one of the seven axial positions in the riser, and the corresponding dimensionless axial height (h/H)

is labeled on the right of each row. Note that all measurements were taken in the (dilute) freeboard region above the surface of the (dense) bubbling bed, which is present just above the distributor plate. The x-axes represent dimensionless radius (r/R, where r is the radial position at which measurement is taken and R is the radius of the riser), and are kept constant in the range of 0 and 1 for all subplots. Each subplot contains four profiles—one for each of the four continuous PSDs investigated (Table 1). Comparisons of the cluster trends obtained here and previous works that also report cluster characteristics throughout the entire riser [24,25] will be made below, though it is important to point out several key differences between the current and previous efforts. Note that 1000 Hz fiber optic data is analyzed to obtain the cluster characteristics here (Fig. 8), while previous work presented characteristics derived from 100 Hz data. Nevertheless, differences in qualitative trends of cluster behaviors are still possible. Accordingly, three differences between the current work and previous should be noted: (i) lower solid loading of m < 0.3 are examined in this work, whereas in the previous works m = 5.9–16.0, (ii) different operating conditions are used to control the system, namely Us and Us –Ut are kept constant in this work, whereas Us and Gs are kept constant in previous work, and (iii) perhaps most importantly, this work examines the impact of varying widths of continuous PSDs and compositions of binary mixtures on cluster characteristics, unlike previous work. Hence, the focus here is on the impact of polydispersity – specifically, the widths of continuous PSDs and the compositions of binary mixtures – on cluster characteristics. Similar to previous work [24,25], it is observed in Fig. 8 that riser height plays a key role in the radial profiles of two of the three cluster characteristics, namely the cluster duration (Fig. 8b), and the cluster frequency (Fig. 8d). On the other hand, dissimilar

Fig. 7. SLR camera photos of clusters at riser wall for lognormal PSD with o/dsm = 10% at h/H = 0.80. The subplots (a) and (b), (c) and (d), and (e) and (f) are characterized by a 0.1 s time lag between each pair.

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

355

Fig. 8. Cluster characteristics, namely, (a) appearance probability, (b) duration, and (c) frequency of various widths of continuous PSDs at various axial positions.

to previous work [24,25], cluster appearance probability does not appear to be affected by local riser position. In addition, in contrast to the previous work [24,25] wherein profile shapes change with riser height for all three cluster characteristics (e.g., transforming from a U-shape at the bottom to an inverted U-shape at the top), profile shapes are relatively invariant with height in Fig. 8. In other words, although the magnitude of a given cluster characteristic may change with axial position, the shape of the profile remains similar. With regard to magnitude, it is observed that although the cluster appearance probability does not vary much with an increase in riser height (Fig. 8a), the magnitude of the duration decreases tenfold (Fig. 8b) and that of the cluster frequency increases tenfold (Fig. 8c). Regarding the impact of changing widths of continuous PSDs, Fig. 8a (first column of Fig. 8) shows that the profiles in each subplot are very similar, implying that widths of continuous PSDs have negligible impact on cluster appearance probability. Yet for the cluster duration (Fig. 8b), profile variations are observed lower in the riser (h/H < 0.43), hence indicating the effect of PSD width is prominent at this lower location. In addition, Fig. 8b presents an interesting contrast between current work and previous work since variations in duration profiles are seen lower in the riser here (Fig. 8b), but are seen higher in the riser in previous work [24,25]. Finally, as for cluster frequency (Fig. 8c), profile variations are observed higher in the riser (h/H > 0.43), which agrees with previous work [24,25]. Thus, Fig. 8c illustrates that the impact of PSD width on cluster frequency is apparent only at these higher locations. For cross-validation of the two independent instruments used for cluster characterization, Fig. 9 compares cluster profiles obtained with the video camera (Fig. 9a and c) vis-à-vis those

obtained with the fiber optic probe (Fig. 9b and d) for the lognormal PSD with /dsm = 10% at various riser heights. (Recall that the appearance probability of clusters is not available for the video camera results since the chosen 50th percentile threshold implies a cluster appearance probability of 50%.) It is worthwhile to note that absolute magnitudes of the cluster characteristics are not expected to agree, since the two instruments involve different measurements (area percent of solids for video images and voltage for fiber optic probe), though both are indicators of relative changes in solids concentration. Also, the nearest wall (r/R = 0.96) measurement is physically not possible for video camera (Fig. 9a and c) because of the spacer on the borescope. Comparing the data obtained from the two instruments reveals that duration decreases with height in both cases (Fig. 9a and b) while frequency increases with height in both cases (Fig. 9c and d). On the other hand, trends in the radial direction are not as clear-cut, since the video camera data appear noisier in this direction than for fiber optic probe. The increased noise for the video data can be traced to the shorter data acquisition duration (constrained by the huge file size generated), which also led to noisier EDj /EJ,all plots (Fig. 4c and d). Nonetheless, the correspondence of the axial trends in duration and frequency of the two instruments is an important validation [23] step since axial position is the primary factor affect cluster behavior (Fig. 8). It is worthwhile to note that although previous works have utilized wavelet analysis of fiber optic data for the characterization of clusters [24,25,48,59,61], its direct comparison with video imaging shown here is the first such validation to be reported. From a practical standpoint, the fiber optic probe is a more feasible method for cluster characterization on two counts: (i) file size—2 s of video file (800 MB) is more than 100 times the size of 30 s of fiber data

356

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

Fig. 9. For lognormal PSD with o/dsm = 10%: cluster duration obtained via (a) video camera, and (b) fiber optic probe; cluster frequency obtained via (c) video camera, and (d) fiber optic probe.

Fig. 10. Cluster characteristics, namely, (a) appearance probability, (b) duration, (c) frequency of various compositions of binary mixtures at various axial positions.

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

(7 MB), and (ii) data analysis time—on the same computer used for data analysis, 2 s of video takes 5 h of computation time, while 30 s of fiber data takes less than a minute. Thus, all three independent instruments – SLR camera (Fig. 7), fiber optic probe (Fig. 8), and video camera (Fig. 5) – provide corroborating evidence of the existence of clusters in dilute flow (m = 0.13–0.29 for continuous PSDs and m = 0.03–0.25 for the binary mixtures). In addition, cluster trends have been verified via measurements with two independent instruments, namely the fiber optic probe and video camera. 3.2. Binary mixture Analogous to Fig. 8, Fig. 10 displays the cluster characteristics for the suite of binary mixture experiments conducted. While the focus of Fig. 8 is on understanding the impact of the widths of continuous PSDs, Fig. 10 aims to illustrate the impact of the composition of binary mixtures (Table 2) on cluster behavior. The axial trends in Fig. 10 are very similar to those in Fig. 8, indicating that riser height is also the primary influence on the magnitude of only two of the three cluster characteristics (cluster duration and frequency in Fig. 10b and c, respectively), though the profile shapes themselves do not change much with height. With respect to cluster appearance probability (Fig. 10a), profile variations among different compositions are generally insignificant, indicating the composition effect of binary mixture is negligible on this cluster characteristic. As for cluster duration (Fig. 10b), similar to continuous PSDs (Fig. 8b), the duration decreases with height and variations among profiles (compositions) are most prominent at the bottom-most axial position. Finally, the composition effects on the cluster frequencies of binary mixtures are similar to width effects of continuous PSDs in that frequency increases with height, and variations among profiles (compositions) increases with riser height (Fig. 10c). Although the binary mixtures (Fig. 10) represent a different category of polydispersity than the continuous PSD (Fig. 8), and previous works have noted different qualitative behaviors between the two [26–32], the impact of the two forms of polydispersity on cluster characteristics in a dilute riser appear to be similar at the noted operation conditions (Tables 1 and 2). 4. Summary An experimental suite with a focus on understanding the clustering phenomena of polydisperse Geldart Group B particles in a dilute CFB riser (m = 0.03–0.29) has been performed. Two categories of polydispersity were investigated. In one category, the widths of continuous PSDs are varied while keeping the Sauter-mean diameter (dsm ) constant. In the second category, the compositions of binary mixtures consisting two species differing in average particle diameter (dave ) and material density (s ) are varied. Within each category of polydispersity, the superficial gas velocity (Us ) was kept constant. Images with large field of view show the presence of clusters even in these very dilute systems, and local (axial and radial) cluster trends spanning the entire riser via both fiber optic probe and video camera show qualitiative agreement. Results indicate that riser axial position is the primary influence on two of the three cluster characteristics (namely, duration and frequency), whereas the effect of PSD width (continuous) or composition (binary) is comparatively minor. With regard to the effect of riser axial position, results are consistent with previous work on a denser CFB riser [24,25] (m = 5.9–16.0) wherein the riser axial position was the dominant influence on cluster duration and frequency. On the contrary, whereas riser axial position was also a key influence on cluster appearance probability in previous work [24,25], the effect of riser axial position on

357

cluster appearance probability is negligible in this study. It is also shown that the magnitude of cluster appearance probability does not change with height, whereas cluster duration decreases and cluster frequency increases with height. Pertaining to the effect of polydispersity (i.e., effect of PSD width (continuous) or composition (binary)), although polydispersity has a negligible effect on appearance probability, it impacts cluster duration at the riser bottom and cluster frequency at the riser top. Comparing to the denser riser in a previous work [24,25], results are also consistent with regard to the insensitivity of cluster appearance probability to polydispersity, and the increase in variations among radial cluster frequency profiles with height. Notably, profile variations are due to the effect of varying widths of continuous PSDs or varying compositions of binary mixtures in this work, whereas those in previous work [24,25] are due to material type (i.e., monodisperse materials of different particle size and/or material density, or different types of polydispersity at fixed continuous PSD width and fixed binary mixture compositions) and/or operating condition. Still along the lines of effect of polysidpersity, one discrepancy with previous work [24,25] is that, whereas variations (due to polydispersity) between cluster duration profiles are most apparent at the riser bottom in this work, variations (due to effect of material type and/or operating condition) between cluster duration profiles are instead most noticeable at the riser top in previous work [24,25]. Interestingly, the two categories of polydispersity investigated here exhibit similar behaviors in terms of axial cluster trends (namely, cluster appearance probability profiles invariant with height, variations among cluster duration profiles decreases with height, and variations among cluster frequency profiles increases with height), which represents a deviation from previous work that reported differences in other fluidization characteristics (e.g. species segregation and elutriation characteristics) between continuous PSDs and binary mixtures [26–32]. The comprehensive cluster dataset on the impact of polydispersity shown here (together with a corollary effort [32] on elutriation and segregation behaviors on the same system) provides valuable insights on flow phenomena in a dilute CFB riser. Such experimental datasets are expected to be valuable for the validation of the numerous kinetic-theory based models [20] for binary mixtures (for example, see [62,63]) and continuous PSDs [64]. It is also expected that this dataset, along with similar sets in a largerscale system [24,25], will provide validation data for clustering predictions, thereby leading to greater physical insight of the trends observed here. Acknowledgements The authors gratefully acknowledge Particulate Solid Research Incorporated (PSRI) for the gifting of the fiber optic probe, and loaning of the borescope. In addition, authors are grateful to Prof. Heinrich Jaeger (University of Chicago) for the loan of the highspeed video camera. Finally, the authors would also like to thank the Department of Energy (DE-FC26-07NT43098) and National Science Foundation (CBET-0650893) for providing funding for this work. References [1] J.R. Grace, A.A. Avidan, T.M. Knowlton, Circulating Fluidized Beds, 1st ed., Blackie Academic & Professional, London; New York, 1997. [2] A.A. Avidan, M. Edwards, H. Owen, Fluid catalytic cracking—past and future challenges, Reviews in Chemical Engineering 6 (January–March (1)) (1990) 1–71. [3] A.A. Avidan, Fluid catalytic cracking, in: J.R. Grace, A.A. Avidan, T.M. Knowlton (Eds.), Circulating Fluidized Beds, 1st ed., Blackie Academic & Professional, London; New York, 1997, pp. 466–488. [4] C. Brereton, Combustion performance, in: J.R. Grace, A.A. Avidan, T.M. Knowlton (Eds.), Circulating Fluidized Beds, 1st ed., Blackie Academic & Professional, London; New York, 1997, pp. 369–416.

358

J.W. Chew et al. / Chemical Engineering Journal 178 (2011) 348–358

[5] Y.Y. Lee, Design considerations for CFB boilers, in: J.R. Grace, A.A. Avidan, T.M. Knowlton (Eds.), Circulating Fluidized Beds, 1st ed., Blackie Academic & Professional, London; New York, 1997, pp. 369–416. [6] P. Basu, Combustion and Gasification in Fluidized Beds, CRC Press, Boca Raton, FL, 2006. [7] R. Jackson, The mechanics of fluidized beds. I. The stability of the state of uniform fluidization, Transactions of the Institution of Chemical Engineers 41 (1963) 13–21. [8] T.B. Anderson, R. Jackson, Fluid mechanical description of fluidized beds—stability of state of uniform fluidization, Industrial & Engineering Chemistry Fundamentals 7 (1) (1968) 12–21. [9] J. Yerushalmi, D.H. Turner, A.M. Squires, Fast fluidized-bed, Industrial & Engineering Chemistry Process Design and Development 15 (1) (1976) 47–53. [10] J.R. Grace, J. Tuot, Theory for cluster formation in vertically conveyed suspensions of intermediate density, Transactions of the Institution of Chemical Engineers 57 (1) (1979) 49–54. [11] G.K. Batchelor, A new theory of the instability of a uniform fluidized-bed, Journal of Fluid Mechanics 193 (1988 August) 75–110. [12] Y.P. Tsuo, D. Gidaspow, Computation of flow patterns in circulating fluidizedbeds, AIChE Journal 36 (June (6)) (1990) 885–896. [13] C.M.H. Brereton, J.R. Grace, Microstructural aspects of the behavior of circulating fluidized-beds, Chemical Engineering Science 48 (July (14)) (1993) 2565–2572. [14] D. Gidaspow, Multiphase Flow and Fluidization: Continuum and Kinetic Theory Descriptions, Academic Press, San Diego, CA, 1994. [15] M. Horio, Hydrodynamics, in: J.R. Grace, A.A. Avidan, T.M. Knowlton (Eds.), Circulating Fluidized Beds, 1st ed., Blackie Academic & Professional, London; New York, 1997, pp. 21–85. [16] K. Agrawal, P.N. Loezos, M. Syamlal, S. Sundaresan, The role of meso-scale structures in rapid gas–solid flows, Journal of Fluid Mechanics 445 (2001 Oct 25) 151–185. [17] S. Sundaresan, Some outstanding questions in handling of cohesionless particles, Powder Technology 115 (1) (2001) 2–7. [18] F.J. Muzzio, T. Shinbrot, B.J. Glasser, Powder technology in the pharmaceutical industry: the need to catch up fast, Powder Technology 124 (1–2) (2002) 1–7. [19] J.S. Curtis, B. van Wachem, Modeling particle-laden flows: a research outlook, AIChE Journal 50 (November (11)) (2004) 2638–2645. [20] C.M. Hrenya, Kinetic theory for granular materials: polydispersity, in: S. Pannala, M. Syamlal, T. O’Brien (Eds.), Computational Gas–Solids Flows and Reacting Systems: Theory, Methods and Practice, IGI Global, Hershey, PA, 2011, Chapter 3. [21] D. Geldart, Types of gas fluidization, Powder Technology 7 (5) (1973) 285–292. [22] E.W. Merrow, Estimating startup times for solids-processing plants, Chemical Engineering 95 (15) (1988) 89–92. [23] T.A. Gauthier, Current R&D challenges for fluidized bed processes in the refining industry, International Journal of Chemical Reactor Engineering 7 (2009), Article A22. [24] J.W. Chew, R. Hays, J. Findlay, T.M. Knowlton, S.B.R. Karri, R.A. Cocco, C.M. Hrenya, Cluster characteristics of Geldart Group B particles in a pilotscale CFB riser. II. Polydisperse systems, Chemical Engineering Science, doi:10.1016/j.ces.2011.09.011, in press. [25] J.W. Chew, R. Hays, J. Findlay, T.M. Knowlton, S.B.R. Karri, R.A. Cocco, C.M. Hrenya, Cluster characteristics of Geldart Group B particles in a pilotscale CFB riser. I. Monodisperse systems, Chemical Engineering Sciencee, doi:10.1016/j.ces.2011.09.012, in press. [26] A.C. Hoffmann, E.J. Romp, Segregation in a fluidized powder of a continuous size distribution, Powder Technology 66 (2) (1991) 119–126. [27] D. Gauthier, S. Zerguerras, G. Flamant, Influence of the particle size distribution of powders on the velocities of minimum and complete fluidization, Chemical Engineering Journal 74 (3) (1999) 181–196. [28] C.L. Lin, M.Y. Wey, S.D. You, The effect of particle size distribution on minimum fluidization velocity at high temperature, Powder Technology 126 (3) (2002) 297–301. [29] J.W. Chew, J.R. Wolz, C.M. Hrenya, Axial segregation in bubbling gas-fluidized beds with Gaussian and lognormal distributions of Geldart Group B particles, AIChE Journal 56 (12) (2010) 3049–3061. [30] J.W. Chew, C.M. Hrenya, Link between bubbling and segregation patterns in gas-fluidized beds with continuous size distributions, AIChE Journal 57 (2011) 3003–3011. [31] J.W. Chew, R. Hays, J. Findlay, S.B.R. Karri, T.M. Knowlton, R.A. Cocco, C.M. Hrenya, Species segregation of binary mixtures and continuous size distribution of Group B particles in riser flow, Chemical Engineering Science 66 (2011) 4595–4604. [32] J.W. Chew, D.M. Parker, C.M. Hrenya, Elutriation and species segregation characteristics of polydisperse mixtures of Group B particles in a dilute CFB riser, AIChE Journal, submitted for publication. [33] J. Ciborowski, A. Wlodarski, On electrostatic effects in fluidized beds, Chemical Engineering Science 17 (1) (1962) 23–32. [34] G. Tardos, R. Pfeffer, A method to measure electrostatic charge on a granule in a fluidized-bed, Chemical Engineering Communications 4 (6) (1980) 665–671.

[35] J. Guardiola, V. Rojo, G. Ramos, Influence of particle size, fluidization velocity and relative humidity on fluidized bed electrostatics, Journal of Electrostatics 37 (1–2) (1996 May) 1–20. [36] G. Hendrickson, Electrostatics gas phase fluidized bed polymerization reactor wall sheeting, Chemical Engineering Science 61 (4) (2006 Feb) 1041–1064. [37] G.L. Osberg, D.H. Charlesworth, Elutriation in a fluidized bed, Chemical Engineering Progress 47 (11) (1951) 566–570. [38] R. Cocco, F. Shaffer, R. Hays, S.B.R. Karri, T. Knowlton, Particle clusters in and above fluidized beds, Powder Technology 203 (1) (2010 Oct 25) 3–11. [39] M. Leva, C.Y. Wen, Elutriation, in: J.F. Davidson, D. Harrison (Eds.), Fluidization., Academic Press Inc., New York, N.Y., 1971, pp. 627–650. [40] D. Geldart, Elutriation, in: J.F. Davidson, R. Clift, D. Harrison (Eds.), Fluidization, Academic Press, London, 1985, pp. 383–411. [41] D. Kunii, O. Levenspiel, Fluidization Engineering, Butterworth-Heinemann, Massachusetts, 1991. [42] L.S. Fan, C. Zhu, Principles of Gas–Solid Flows, Cambridge University Press, Cambridge, UK, 1998. [43] M. Rhodes, Introduction to Particle Technology, John Wiley & Sons Ltd., West Sussex, England, 1998. [44] S.M. Tasirin, D. Geldart, Entrainment of FCC from fluidized beds—a new correlation for the elutriation rate constants K-i infinity*, Powder Technology 95 (March (3)) (1998) 240–247. [45] S.R. Dahl, C.M. Hrenya, Size segregation in rapid, granular flows with continuous size distributions, Physics of Fluids 16 (January (1)) (2004) 1–13. [46] R. Chrisman, R. Cocco, J. Cleveland, inventors, Method and apparatus for determining flow rates and concentrations in multi-phase streams, U.S., 1994. [47] R.A. Cocco, J. Cleveland, R. Harner, R. Chrisman, Simultaneous in-situ determination of particle loadings and velocities in a gaseous medium, AIChE Symposium Series 308 (1995) 147. [48] T.Y. Yang, L.P. Leu, Multiresolution analysis on identification and dynamics of clusters in a circulating fluidized bed, AIChE Journal 55 (March (3)) (2009) 612–629. [49] H. Zhang, P.M. Johnston, J.X. Zhu, H.I. de Lasa, M.A. Bergougnou, A novel calibration procedure for a fiber optic solids concentration probe, Powder Technology 100 (2–3) (1998 Dec 1) 260–272. [50] Y. Matsuno, H. Yamaguchi, T. Oka, H. Kage, K. Higashitani, The use of optic fiber probes for the measurement of dilute particle concentrations—calibration and application to gas-fluidized bed carryover, Powder Technology 36 (2) (1983) 215–221. [51] P.M. Herbert, T.A. Gauthier, C.L. Briens, M.A. Bergougnou, Application of fiber optic reflection probes to the measurement of local particle-velocity and concentration in gas–solid flow, Powder Technology 80 (September (3)) (1994) 243–252. [52] S.G. Mallat, A theory for multiresolution signal decomposition—the wavelet representation, IEEE Transactions on Pattern Analysis and Machine Intelligence 11 (July (7)) (1989) 674–693. [53] S.G. Mallat, A Wavelet Tour of Signal Processing, Academic Press, San Diego, 1998. [54] B.R. Bakshi, H. Zhong, P. Jiang, L.S. Fan, Analysis of flow in gas–liquid bubblecolumns using multiresolution methods, Chemical Engineering Research & Design 73 (August (A6)) (1995) 608–614. [55] J. Ren, J. Li, Wavelet analysis of dynamic behavior in fluidized beds, Fluidization IX (1998) 629–636. [56] N. Ellis, L.A. Briens, J.R. Grace, H.T. Bi, C.J. Lim, Characterization of dynamic behavior in gas–solid turbulent fluidized bed using chaos and wavelet analyses, Chemical Engineering Journal 96 (1–3) (2003 Dec 15) 105–116. [57] G.B. Zhao, Y.R. Yang, Multiscale resolution of fluidized-bed pressure fluctuations, AIChE Journal 49 (4) (2003 Apr) 869–882. [58] M. Misiti, Y. Misiti, G. Oppenheim, J. Poggi, Matlab Wavelet Toolbox Users Guide Version 2, The MathWorks, Inc., Natick, MA, 2002. [59] C. Guenther, R. Breault, Wavelet analysis to characterize cluster dynamics in a circulating fluidized bed, Powder Technology 173 (3) (2007 Apr 30) 163–173. [60] M. Horio, H. Kuroki, 3-Dimensional flow visualization of dilutely dispersed solids in bubbling and circulating fluidized-beds, Chemical Engineering Science 49 (August (15)) (1994) 2413–2421. [61] F.A. Afsahi, R. Sotudeh-Gharebagh, N. Mostoufi, Clusters identification and characterization in a gas–solid fluidized bed by the wavelet analysis, Canadian Journal of Chemical Engineering 87 (June (3)) (2009) 375–385. [62] V. Garzó, J.W. Dufty, C.M. Hrenya, Enskog theory for polydisperse granular mixtures I. Navier–Stokes order transport, Physical Review E 76 (September (3)) (2007) 031303. [63] V. Garzó, C.M. Hrenya, J.W. Dufty, Enskog theory for polydisperse granular mixtures II. Sonine polynomial approximation, Physical Review E 76 (September (3)) (2007) 031304. [64] J. A. Murray, C. M. Hrenya, V. Garzó, Enskog Theory for Polydisperse Granular Mixtures. III. Comparison of dense and dilute transport coefficients and equations of state for a binary mixture, Powder Technology, doi:10.1016/j.powtec.2011.09.030, in press.