You are on page 1of 15

Powder Technology 202 (2010) 24–38

Contents lists available at ScienceDirect

Powder Technology
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / p ow t e c

Detection of agglomeration and gradual particle size changes in circulating


fluidized beds
Malte Bartels, John Nijenhuis, Freek Kapteijn, J. Ruud van Ommen ⁎
Delft University of Technology - Department of Chemical Engineering, Delft Research Centre for Sustainable Energy, Julianalaan 136, 2628 BL Delft, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: The monitoring of fluidized bed processes is valuable for the optimization of operational efficiency and
Received 21 May 2008 avoiding unscheduled shut-downs in industrial practice. We here focus on two important applications for
Received in revised form 13 February 2010 the industrially widely applied circulating fluidized beds (CFBs): the detection of small, gradual changes in
Accepted 26 March 2010
particle size and the early detection of agglomeration. Based on high-frequency (200 Hz) pressure
Available online 2 April 2010
fluctuation measurements we have applied the attractor comparison method in different CFBs to investigate
Keywords:
its suitability as a monitoring tool for the two proposed applications.
Circulating fluidized beds Attractor comparison has shown to be generally sensitive to small, gradual particle size changes in a lab-scale
Solid fuel conversion CFB. The sensitivity normally depends on the measurement location (downcomer, return system, riser) and
Agglomeration is often larger in the downcomer section. From the results, detection limits for the particle size have been
Particle size identified. In an industrial installation the sensitivity and applicability of the method for measurements in
Monitoring the riser has been confirmed. With the help of model agglomerates it has been shown that attractor
Early warning comparison can also detect the presence of small shares of agglomerates on lab-scale; here, filtering the
pressure fluctuation data before the analysis has shown to improve the method.
The suitability of the method for agglomeration detection has been confirmed with an agglomeration
incident in a hot lab-scale facility; in this case the approaching defluidization has been detected with an early
time of about 30 min based on measurements in horizontal return-leg and lower riser.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction from the fuel, especially potassium (K) but also sodium (Na), can be a
source of agglomeration. The content of this critical inorganic material
Fluidized bed reactors are applied in a wide variety of processes in can vary much between fuels; especially in the case of certain types of
industry, comprising e.g. catalytic cracking, gas-phase polymerization biomass (e.g. straw) as well as some low-rank coal types their content is
and drying. They are also used in solid fuel conversion, where often rather high. When both alkalis and silica are present in the bed
combustion is currently the most important application in terms of they can form alkali silicates that are characterized by a lower melting
capacity, but also gasification is becoming increasingly important. point than the individual components. This lower melting point is often
Amongst other reasons especially geo-political considerations as well lower than the operating temperature. As a consequence, the sand
as the strongly increasing crude oil price have made coal attractive particles become coated with an adhesive layer. Sand particles with a
again lately. For the same reason, but also in light of environmental sticky surface will then form agglomerates due to the formation of
considerations, biomass usage is currently quickly increasing. Fluid- permanent bonds upon collisions (e.g. [3,4]). If this agglomeration
ized beds are generally very flexible with respect to the utilized fuel process is not recognized, it will eventually result in partial or total
(e.g. [1]), which makes changes to other solid fuel sources as well as defluidization of the reactor, and as a consequence a lengthy and
co-combustion of biomass and other fuels relatively easy to expensive unscheduled shutdown.
implement. Consequently, fluidized beds have become widely The detection of agglomeration with commonly available process
applied, mainly in rapidly developing countries in south-east Asia [2]. measurement frequencies of 1 Hz or less has shown to be not reliable
Still, solid fuel conversion in fluidized bed processes is facing and, more important, detection is not early enough to take corrective
technical difficulties; specifically agglomeration can be a major measures in time. Various different measurements and analysis
operational problem. Usually, the conversion of the solid fuel is carried methods have been proposed in the literature; for a review see Ref.
out with silica sand as the bed material. Inorganic alkali components [5]. Here, we focus on the “attractor comparison” method, briefly
described in the following; for a more detailed description of the
⁎ Corresponding author.
method, see Ref. [6]. The state of a fluidized bed at a certain time can
E-mail addresses: m.bartels@tudelft.nl (M. Bartels), j.r.vanommen@tudelft.nl be determined by projecting all variables governing the system into a
(J. Ruud van Ommen). set of delay-vectors in a multidimensional space (“state space”). The

0032-5910/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.powtec.2010.03.035
M. Bartels et al. / Powder Technology 202 (2010) 24–38 25

contrast to agglomeration, where avoiding unnecessary shut-downs is


the motivation, on-line particle size monitoring can be used to keep
the bed parameters within tighter limits, e.g. for optimized heat
transfer, increasing the overall efficiency of the process. We consider
on-line particle size monitoring as an important application, also with
opportunities in many other fluidized bed processes.
The goal of this paper is to present how the attractor comparison
method can be applied for the early detection of agglomeration and as a
particle size monitoring tool in circulating fluidized beds. In order to
obtain a general picture on the applicability of this method, we have
included several setups of different size and with different solids return
mechanisms in our work. Furthermore, we address some of the important
Fig. 1. Reconstruction of an attractor from a pressure fluctuation time series. issues regarding the implementation of attractor comparison in practice.

subsequent delay vectors can be regarded as points in an m-


2. Experimental
dimensional state space yielding a reconstructed “attractor” (Fig. 1).
In the actual implementation of the method first a reference attractor
Three different setups are used (Table 1). The cold lab-scale setup
is constructed based on pressure fluctuation data from a well-
was operated with two different downcomer configurations.
fluidized bed, called the reference state. Subsequently, the attractor
The cold-flow lab-scale CFB is located at the Product & Process
of the current pressure fluctuation data from the fluidized bed is
Engineering section of Delft University of Technology. The lower part
reconstructed and compared to the reference attractor. This compar-
of the setup with L-valve and a CFB schematic are shown in Fig. 2, also
ison is based on a statistical test developed by Diks et al. [7], which
indicating the measurement positions for pressure fluctuations.
evaluates the dimensionless squared distance S between both
With a moving bed downcomer and L-valve configuration the
attractors: S = pQ̂ffiffiffiffiffiffiffiffiffi, where Q̂ is the estimator of the squared
Vc ðQ̂ Þ setup has been operated with 30 kg bed material (d50 = 235 μm).
distance between two smoothed distributions of the two attractors Table 2 gives an overview of the experiments with the L-valve
and Vc(Q̂ ) is its variance. For stationary hydrodynamics, S has an configuration, including the operating conditions. To evaluate the
expectation of zero and a standard deviation of one. An S-value larger influence of mass changes, 3 and 6 kg bed material of the same particle
than 3 indicates with at least 95% confidence that the two attractors size have been added (run L 1). This corresponds to a relative bed mass
have been generated by a different system, meaning that the increase of 10% and 20%, respectively. Material has been added
hydrodynamics has changed, e.g. because of agglomeration. Attractor through a hopper into the downcomer, thereby slightly increasing the
comparison has been shown to be sensitive to agglomeration in solids bed height in the downcomer. Subsequently, experiments with
bubbling fluidized beds on lab-scale as well as on industrial scale [6,8]. the addition of 3 and 6 kg particles of a larger particle size
However, the S-value is a global measure of the hydrodynamics of the (d50 = 400 μm) have been carried out (run L 2). This corresponds to
fluidized bed and the change in hydrodynamics obviously does not an averaged particle size increase from 235 μm to 242 μm and 249 μm,
necessarily have to originate from agglomeration, but could also respectively. For a lower reflux two additional experiments have been
originate from changes in the operating parameters or other process carried out, in which a certain share of the bed material has been
disturbances. With respect to operating conditions, attractor compar- substituted by particles of d50 = 400 μm (run L3) and very large
ison was shown to be insensitive to relative changes of ±10% in bed particles of d50 = 1900 μm, i.e. model agglomerates (run L 4). For all
mass and fluidization velocity in bubbling fluidized beds [6]. calculations of the attractor comparison method, the initial bed mass
Agglomeration detection in fluidized beds with attractor comparison of 30 kg (100%) bed mass has been taken for the reference attractor.
has been focussed on bubbling fluidized beds (BFBs) in the past. However, The lower part of the setup with loop seal configuration is shown
especially for large-scale energy conversion processes circulating in Fig. 3. With the loop seal configuration the setup has been operated
fluidized beds (CFBs) are frequently applied in industry, mainly because with 35 kg bed material due to the increased volume of the modified
of their larger capacity. The share of CFBs on worldwide installed capacity downcomer. The relative mass changes of 10 and 20%, however, are
of fluidized bed energy conversion has increased to an estimated 85% in kept the same here. In this case, 20% bed mass have ultimately been
2004 according to the International Energy Agency, mainly due to a removed again after the addition in two steps, thereby returning to
dramatic capacity increase in Asia in the beginning of this century [9]. the original amount of bed mass. Bed mass addition is carried out as in
This development also motivates the investigation on whether and how the L-valve configuration. The same two sand types have been used as
attractor comparison can be applied for monitoring CFBs. for the CFB in Fig. 2. The reflux has been measured during operation at
Besides the early warning of agglomeration also the monitoring of specific times. The measurement has been carried out with a separate
particle size is considered an important application in practice. In valve and transparent section in the upper downcomer, for which the

Table 1
Overview of the experimental setups and operating conditions.

Setup Diameter Riser height Fluidization velocity riser Mass reflux Particle size Load Experiments
[m] [m] [m/s] [kg/(m2s)] [μm] [MWth]

Cold lab-scale CFB, 0.083 ∼4 5 40 235 (d50) n/a Increase of particle size by mixture of 2
L-valve configuration different sized sand types:
Cold lab-scale CFB, 0.083 ∼4 5 40; 65 235 (d50) n/a 1) 235 + 400 μm
loop seal configuration 2) 235 + 1900 μm (only L-valve)

Hot lab-scale CFB, 0.083 ∼ 5.5 ∼4.4 Not measured 500–800 ∼ 0.1 (max.) Forced to agglomerate by adjustment of
L-valve configuration (range) fuel type and temperature

Hot industrial CFB, 4.2 ∼ 33 ∼3 Not measured ∼ 250 (d50) 75–100 Monitoring of particle size during operation
loop seal configuration (max. = 109) and simultaneous bed sampling
26 M. Bartels et al. / Powder Technology 202 (2010) 24–38

Fig. 2. Photo and schematic drawing of the cold-flow setup with L-valve for solids return. Locations of pressure fluctuation sensors are indicated in white (DC = Downcomer,
RD = Return-Leg (Riser/Downcomer), R = Riser; numbers refer to distance (mm) from distributor plate). Black arrows indicate solids flow.

material building up in that section for a short time after closing the adjusted to its normal value in case it had changed significantly. For
valve has been determined (after that the valve is opened again). This experiment LS 3 and LS 4 the operating conditions were changed to a
measurement did not affect the results of the attractor comparison larger reflux and lower bubbling bed velocity to investigate whether
method. In certain cases the mass reflux has been re-adjusted by the performance of the method changes for a different operating
adapting the gas flow to the according value as given in Table 2. For condition.
the gas flows in this configuration, the total riser gas flow has been Note that for both the moving bed configuration and the loop seal
corrected for the gas flow coming from the bubbling bed as this gas configuration a small stagnant zone in the lower downcomer is
also enters the riser due to the pressure conditions; the total riser flow present and estimated to contain 2–3% of the total bed mass. This
including the bubbling bed gas flow results in a velocity of 5 m/s. effect has not been taken into account here, the true changes in bed
Table 3 gives an overview of the experiments with the loop seal mass and particle size therefore will be slightly higher. For both
configuration, including the operating conditions. In all experiments configurations, a dense bed was present in the lower section of the
two times 10% bed material has been added and subsequently 20% has riser (up to approx. 0.25 m height).
been removed again to return to the original bed mass. The resulting The hot lab-scale CFB (Fig. 4) is located at the Energy Technology
median particle size changed from d50 = 235 μm to d50 = 242 μm and section of Delft University of Technology. With regard to the
d50 = 249 μm upon the addition. Next, the mass reflux has been re- dimensions this setup is very similar to the cold lab-scale setup and
also has a moving bed downcomer with an L-valve for the solids
return. Pressure fluctuations have been measured at different
Table 2 locations in riser and downcomer; however, many positions got
Overview of the experiments for the lab-scale CFB with L-valve configuration. blocked during operation due to insufficient purging. Only the two
Run Riser velocity Mass reflux Imposed change Steps [wt.%] positions used in this study, in the horizontal return-leg and lower
[m/s] [kg/(m2s)] riser, are indicated in Fig. 4. The lower riser position is within a dense
L1 5 40 Addition of 235 μm +10/+ 20 bed region, with some tendency for slugging. The hot lab-scale CFB
particles (total excess) has been operated as a gasifier. Bed agglomeration has been forced by
L2 5 40 Addition of 400 μm the choice of fuel containing relatively high amounts of alkali, in
particles
combination with operating temperatures sufficiently high for sticky
L3 5 25 Substitution of 1.4/2.9/5.7/8.6/14.3
bed material with (total substitution) low-melting silicates to be formed. In this case straw in pelletized
400 μm particles form was used, operating the setup in the range of 750–760 °C.
L4 5 ∼25 Substitution of 1.4/2.9/5.7 The industrial CFB boiler is operated by Sachtleben Chemie GmbH,
(changing) bed material with (total substitution) located in Duisburg, Germany. It has a loop seal for solids return. Two
1900 μm particles
measurement positions were located in the riser (diameter 4.2 m) at a
M. Bartels et al. / Powder Technology 202 (2010) 24–38 27

Fig. 3. Cold-flow setup with loop seal configuration. Locations of the pressure fluctuation sensors are indicated in white (DC = downcomer, BB = bubbling bed, R = Riser; numbers
refer to distance (mm) from distributor plate). Black arrows indicate solids flow.

height of ∼900 mm from the distributor plate within the lower, dense 3. Results and discussion
bed region. Both measurement locations were located at the wall,
separated 90° relative to each other in radial direction. Measurements 3.1. Cold lab-scale CFB with L-valve configuration
in the downcomer/loop seal were not possible in this case. During the
course of several weeks the pressure fluctuations have been recorded In Fig. 5 a short representative pressure time series from the
and simultaneously the particle size has been determined three times downcomer and the riser are shown. The pressure fluctuations from
per day by sampling and sieving. both riser positions look very similar, with the higher position (R625)
Pressure fluctuations were measured using piezoelectric sensors of having a slightly smaller amplitude due to the signal attenuation along
type Kistler 7261 in all cases. For the cold-flow setup the 4 mm ID the flow upwards in the riser. The downcomer is operated as a moving
measurement tubes were covered with wire mesh gauze (mesh bed and does therefore not show any bubbling behavior. The visual
diameter 144 μm), for combustion and gasification the measurement inspection of the pressure time series shows that the downcomer
tubes were purged with nitrogen to avoid blockage by particles. The pressure fluctuations resemble the riser pressure fluctuations in shape.
pressure signal was low-pass filtered at 200 Hz and recorded at They are significantly decreased in amplitude and have a small time
400 Hz to avoid aliasing effects according to the Nyquist criterion delay as compared to the riser pressure fluctuations. For this reason,
(sampling frequency ≥ 2 lowest present frequency). Here, this the pressure fluctuations measured in the downcomer are suspected
frequency range is generally considered a “high-frequency” range, to mainly originate from the riser pressure fluctuations, which are
compared to the usual sampling frequencies of process data of about attenuated by the moving sand bed between riser and downcomer as
1 Hz or lower. The parameterization of attractor comparison based the pressure waves propagate from the riser to the downcomer.
on these measurements has been optimized for each experimental
setup and is presented in Table 4. For a detailed description of the 3.1.1. Bed mass changes
parameters see Ref. [6]. In order to investigate the sensitivity of the attractor comparison
method to changes in particle size, we decided to gradually increase
the particle size by adding larger sized sand to the downcomer during
Table 3 operation. In order for this to be a valid approach, the influence of
Overview of the experiments for the lab-scale CFB with loop seal configuration.
changes in total mass on the attractor had to be investigated by adding
Run Riser Mass reflux Bubbling Imposed Steps bed material of the same size (d50 = 235 μm). The S-value has then
velocity [kg/(m2s)] bed velocity change [wt.%] been calculated for the comparison between the reference case
[m/s] [Multiple of
(100%) and each of the cases with increased bed mass (110% and
Umf, fine
(5.3 cm/s)] 120%) with same particle size, as shown in Fig. 6. Adding 10% and 20%
relative bed mass does not significantly influence the S-value. Only for
LS 1 5 40 5 Addition of +10/+20/0
235 μm (total excess)
the measuring position R189 in the riser, the S-value occasionally but
particles not structurally exceeds the threshold value of 3.
LS 2 5 40 5 Addition of
400 μm 3.1.2. Gradual particle size changes
particles
In the next step, sand with a bigger particle size (d50 = 400 μm)
LS 3 5 65 3 Addition of +10/+20/0
235 μm (total excess) has been added to the system (Fig. 7 a–c). Also here, 3 kg and 6 kg, or
particles 10% and 20% relative bed mass, have been added in one experiment,
LS 4 a/b 5 65 3 Addition of resulting in mean particle sizes of 242 μm and 249 μm, respectively.
(duplicating) 400 μm Again, 100% bed mass with d50 = 235 μm has been taken as the
particles
reference.
28 M. Bartels et al. / Powder Technology 202 (2010) 24–38

Fig. 4. Schematic of the hot lab-scale CFB. Locations of the pressure fluctuation sensors are indicated in white. Black arrows indicate solids flow.

The S-values from the downcomer and riser measuring positions fluctuations measured in the downcomer are originating from the
are sensitive to the changing particle size; the sensitivity is clearly riser and have been altered, or filtered, in some way. Pressure
strongest for the measurements in the downcomer. The minimal fluctuation measurements in the downcomer can therefore also yield
change in particle size that can be reliably detected is estimated information about the hydrodynamic state of the riser.
approximately 10 μm for the riser and below 7 μm for the downcomer, Subsequently, a measurement location in the horizontal return-
based on a median particle size of 235 μm. Please note that the mass pipe between downcomer and riser been added and the response of
reflux has not been monitored during these experiments and could
have changed slightly upon the addition of larger size sand (and same
size sand in the previous case), although this has not been visually
observed. Note that for the loop seal configuration (see subsequent
sections) the mass reflux has been measured.
The reason for the greatly differing sensitivities between riser and
downcomer is not yet clear. The downcomer is operated as a moving
bed where no bubble-formation takes place. Together with the raw
pressure fluctuation data (Fig. 5) it is suspected that the pressure

Table 4
Attractor comparison parameterization for the different setups (bandwidth d = 0.5 in
all cases).

Setup Reference Evaluation Embedding Segment


time window time window dimension length [s]
[min] [min] [# points]

Cold lab-scale CFBs 10 5 40/20 2


Hot lab-scale CFB 4 2.5 100 2
Fig. 5. Example pressure time series at different positions for the cold lab-scale CFB with
Industrial CFB 15 15 20 3
L-valve solids return.
M. Bartels et al. / Powder Technology 202 (2010) 24–38 29

Fig. 6. Influence of bed mass changes on the S-value. The reference is taken at 100% bed
mass in both cases; the S-values are shown for an increase of 10% (a) and 20% (b) in bed
mass for the downcomer DC126, riser R189 and R625 (run L 1).

the S-value on incased particle size has been investigated. In contrast


to the previous experiments, the particle size increase has not been
carried out by addition of larger particles, but by substituting shares of Fig. 7. Response of the S-value towards changing particle size for the downcomer
the bed material with larger particles to rule out any influence of a DC126, riser R189 and R625 (run L 2). 100% bed mass (Reference): d50 = 235 μm, 110%
bed mass: d50 = 242 μm, 120% bed mass: d50 = 249 μm.
changing bed mass. The results of attractor comparison are shown in
Fig. 8. Interestingly, the horizontal return-pipe (position RD126) has
shown to be the most sensitive position, even more sensitive than the
downcomer position; the detection limit for the increase in particle
size would in this case be around 2 μm. The riser is hardly sensitive to
the particle size changes.

3.1.3. Filtering
Previously, the hypothesis has been stated that the pressure
fluctuations measured in the downcomer basically are a modified
version of the pressure fluctuation signal from the riser, i.e. both are
correlated. Given the stronger sensitivity of the attractor comparison
method based on pressure fluctuations measured in the downcomer,
it could be possible to modify the pressure fluctuation signal from the
riser so that the attractor comparison method will exhibit a larger
sensitivity. Therefore, it has been investigated whether or not an
increased sensitivity to particle size changes could also be realized by Fig. 8. Response of the S-value for increasing particle size via the substitution of bed
applying some numerical filter to the pressure fluctuation data from material with larger sized sand of d50 = 400 μm. Operating conditions: Riser velocity
the riser before carrying out attractor comparison. 5 m/s, constant mass reflux of 25 kg/(m2s) (run L 3).
30 M. Bartels et al. / Powder Technology 202 (2010) 24–38

We have observed that the difference in signal power (amplitude)


between riser and downcomer is getting larger with increasing
frequency in the frequency range of 0 to about 20 Hz. This implies that
the sand bed between the riser and the downcomer attenuates
(damps) the pressure waves travelling from riser to downcomer,
particularly for those higher frequencies, i.e. it behaves as a low-pass
filter. Several low-pass filters with cut-off frequencies in the range of
2–10 Hz were applied to the riser data (run L 2) before carrying out
attractor comparison, to see whether it is possible to “imitate” this
hypothesized hardware filter of the bed and increase sensitivity to
particle size changes. However, the sensitivity only decreased, which
shows that the higher-frequency signal components (above 10 Hz)
indeed contribute to the sensitivity of the method. Although this does
not disprove the low-pass filter character of the bed, it can be Fig. 10. Response of the S-value in the horizontal return-leg for changing riser gas
velocities. Low-pass filtering (cut-off 30 Hz) can significantly reduce the sensitivity to
concluded that the sensitivity of attractor comparison based on the velocity changes, here for decreased velocities. (Operating conditions: Riser velocity
riser measurements cannot be increased using a simple (single-band) 5 m/s, constant mass reflux of 25 kg/(m2s)).
low-pass filter. Also different band-pass filters of 10 Hz band widths
(in the range from 0 to 100 Hz) did not increase sensitivity. High-pass
filters (based on similar experimental data) have shown to be not For very low mass refluxes, and therefore lean riser densities,
suitable either [10]. High-pass filters have the disadvantage that the attractor comparison often yielded erratic results and cannot be
attractor comparison analysis on such data yields often very erratic reliably applied to detect particle size changes. This phenomenon can
results, i.e. much less reproducibility and consistency, despite be expected since the pressure fluctuation amplitude decreases with
generally higher sensitivities. This effect is not surprising, considering decreased riser density. Therefore the contribution of the pressure
the fact that the total power of the pressure fluctuation signal is fluctuations from the voids/clusters in the riser to the overall signal is
dominated by lower frequency components (up to 10 Hz). Therefore becoming smaller relative to other undesired effects, such as e.g.
also the capability of the pressure sensor to resolve the higher temperature or relative humidity changes or vibrations of the column
frequency components is limited by definition which in turn means itself. For very lean riser densities we have indeed observed that the
that noise and other physical effects (e.g. column vibration) will then analysis can be completely dominated by a large high-frequency peak
play a more dominant role. It could be worthwhile to investigate (∼80 Hz) observed in the power spectrum of the pressure fluctua-
whether is it possible to design a more sophisticated filter that can tions, which we suspect to originate from a characteristic column
achieve an increased sensitivity. One possible starting-point would be vibration.
to experimentally investigate or model the filter-characteristics of a
slowly moving sand bed in order to design a filter, potentially a multi- 3.1.4. Bi-modal particle size changes (model agglomerates)
band frequency filter. Besides the gradual mono-modal particle size changes we have
We have also investigated the effect of low-pass filtering for run L also investigated the replacement of fractions of the bed material with
3 and several other experiments (not shown here). In general, we much larger particles with a mean diameter of 1900 μm. The resulting
observed that low-pass filtering generally decreases the sensitivity to particle size distribution is bi-modal and is therefore seen as a
particle size changes, although that decrease is often just very small. simulation of the formation of small agglomerates. The results for the
More important, however, it also has shown to reduce the sensitivity substitution of bed material in three steps are shown in Fig. 11. Upon
to changes in bed mass and changes in riser gas velocity at the same each substitution of a share of the bed mass with the model
time. This is an important factor when considering that the method agglomerates characteristic steps in the S-value are observed,
should be robust against distortions in these parameters during especially for the riser. This shows that the method is also sensitive
operation. This effect of sensitivity reduction has shown to be to the presence of the model agglomerates. However, also another
strongest for the horizontal return-pipe. Overall, applying a 30 Hz effect is present here, which results in strong changes in S and is
low-pass filter to the riser positions and a 10–40 Hz band-pass filter clearly visible for the first ∼90 min in the riser. We have observed a
(or alternatively a 30 Hz low-pass filter) to the downcomer and relatively strong peak in the power spectrum at about 80 Hz that is
return-pipe has shown to be optimal. An example for the return leg slowly shifting, potentially from a changing mass reflux. This shift is
for an increase in particle size is shown in Fig. 9 and a change in riser
gas velocity is shown in Fig. 10.

Fig. 9. Response of the S-value in the return leg for increasing particle size via the
substitution of bed material with larger sized sand of d50 = 400 μm. Low-pass filtering Fig. 11. Response of S-value for increasing particle size via the substitution of bed
(cut-off 30 Hz) only slightly decreases the sensitivity to particle size. Operating material with model agglomerates of d50 = 1900 μm. Mass reflux plotted with 95%
conditions: Riser velocity 5 m/s, constant mass reflux of 25 kg/(m2s) (run L 3). confidence intervals. (Operating condition: Riser velocity 5 m/s).
M. Bartels et al. / Powder Technology 202 (2010) 24–38 31

Fig. 13. Response of the S-value to changes in the total mass in the downcomer and
Fig. 12. Response of the S-value for increasing particle size via the substitution of bed bubbling bed (run LS 1). Mass addition and removal are indicated with vertical bars.
material with model agglomerates of d50 = 1900 μm, based on 30 Hz low-pass filtered
pressure fluctuation data (Operating conditions: Riser velocity 5 m/s).
addition (similar as in Fig. 12) and does not indicate sensitivity to the
increased total mass. For the riser positions the S-value is shown in
considered to be the reason for the strong, unexpected changes in S. Fig. 14. Also for the riser positions the S-value stays around zero; only
We applied a 30 Hz low-pass filter to the data before the attractor the position at 189 mm occasionally increases above the value of 3,
comparison, the results as shown in Fig. 12. After filtering, the but this is also the case before the addition and there is no correlation
unknown effect is not observed anymore and for the second and third with the mass changes in the system. Higher S-values are observed at
substitution with model agglomerates the method clearly follows the the very beginning of the measurements and are related to the start-
increasing presence of model agglomerates. The sensitivity of the riser up. Moreover, one can again see that during the removal of bed mass
obviously decreased, as the influence of the unknown effect is now all measuring positions show a temporary increase of the S-value, but
removed. The detection limit for the added model-agglomerates this is no permanent effect. This effect is assumed to originate from a
appears to be around 2–3 wt.%. For the downcomer, there is a spike temporary drop in the riser density due to the decreased amount of
observed upon addition, as for the unfiltered data, but the S-values material transported into the riser during mass removal from the
only minimally decreased as compared to the unfiltered data. For the lower downcomer. This is confirmed by a temporary decrease in the
return-leg, the S-values did decrease somewhat. This generally shows pressure drop in the riser at the same time.
that low-pass filtering is very beneficial for improving the selectivity These results confirm that attractor comparison is not sensitive to
of the method in the riser and the return-leg, specifically in this case relative changes of the bed mass of up to 20%. Only relatively quick
where another overlaid disturbing effect is present. In contrast with mass changes, here about 10% of the bed mass within a few minutes,
the gradual increase of particle size, the sensitivities of all measuring can temporarily increase the S-value at certain measurement
positions now are very similar. It therefore appears that the positions.
hydrodynamic effect from the presence of agglomerates is different
from the gradually increased particle size; the mechanism for this 3.2.2. Gradual particle size changes (2 different operating conditions)
effect has not been investigated. The increase in particle size has been realized via the addition of
larger size sand together with subsequent removal of excess bed
3.2. Cold lab-scale CFB with loop seal configuration mass. The mass reflux had actually changed due to the increase of the
particle size but has been re-adjusted, as shown in Fig. 15. This
The cold lab-scale setup has been modified with a loop seal decrease in mass reflux is suspected to either originate from the lower
replacing the L-valve for solids recirculation for two reasons. First, we moving bed downcomer compartment, where the sand flow can be
wanted to obtain a stronger decoupling between riser and down- hindered as a result of some unfavourable settlement of larger
comer as compared to the L-valve configuration to investigate the particles with the gas flows being constant, or from the bubbling bed,
previously proposed hypothesis of the moving sand bed acting as a where the increased particle size could slow down transport through
hardware filter for pressure fluctuations originating from the riser and the bed. Other experiments (not presented here) showed a clear
thereby increasing the sensitivity of attractor comparison to small variation in mass reflux changes upon addition of larger sized sand;
particle size changes. Second, a loop seal configuration mimics large
industrial installations better than a moving bed L-valve configuration
and is therefore more relevant for the application of the method in an
industrial environment.

3.2.1. Bed mass changes


The loop seal configuration is compared to the L-valve configura-
tion for the same operating conditions, at a riser velocity of 5 m/s and
a mass reflux of 40 kg/(m2s). During this experiment (run LS 1), the
mass reflux did not change. For the addition of two times 10% bed
mass of the same particle size and subsequent removal of 20% bed
mass, the responses of attractor comparison for the downcomer and
bubbling bed positions are shown in Fig. 13. S does not react on the
changes in bed mass in the downcomer and the bubbling bed. The
only exception is the downcomer position at a height of 850 mm,
where one can see a spike in the S-value during the actual addition. As Fig. 14. Response of the S-value to changes in the total mass in the riser (run LS 1). Mass
this peak quickly disappears, it is clearly related to the moment of addition and removal is indicated with vertical bars.
32 M. Bartels et al. / Powder Technology 202 (2010) 24–38

construction is principally clearly less prone to such settlement effects


and subsequent mass reflux changes.
The influence of gradual particle size changes has also been
investigated for a second set of operating conditions, at increased
mass reflux and decreased fluidization velocity in the bubbling bed.
Visual observations of the previously presented operating conditions
(LS 1 and 2) showed that the riser density was lower and that the riser
showed less tendency for slugging compared to the L-valve
configuration at the same set riser velocity and mass reflux. Here,
the mass reflux has been adapted to compensate this effect and
operate at higher riser densities, in the range of the experiments with
L-valve configuration.
The reason for this difference in riser density is suspected to
originate from the different reflux mechanisms for L-valve and loop
Fig. 15. Development of the mass reflux G(s) with 95% confidence intervals during seal. For the loop seal configuration a share of ∼10% of the total riser
particle size changes from d50 = 235 μm to d50 = 242 μm and d50 = 249 μm (run LS 2).
Additions and removal of bed mass and adaptation of G(s) are indicated with vertical
gas flow is delivered by the bubbling bed. This results in a secondary
bars. gas injection in the dense part of the riser, centered at a height of
12.5 cm from the distributor plate. In bubbling fluidized beds it has
been shown that secondary gas injection in the bed increases the
this indicates that the resulting hydrodynamics upon the addition is Peclet number, corresponding to a decreased back-mixing [11],
not necessarily well reproducible. although is not clear if this effect is also applicable in a circulating
The resulting S-values for the particle size increase related to the bed where gas voids are much less distinct. From the definition of Pe
location in the downcomer and in the bubbling bed are shown in this effect is equivalent to an increase in gas velocity, which would
Fig. 16. The S-value obtained from the highest downcomer position lead to a decreased density and a lower tendency for slugging.
DC850 increases with increasing particle size. However, one can also Another reason for this difference in riser densities could lie in the
notice that the S-value changes due to the final re-adjustment of the physical connection between bubbling bed and riser. The space above
mass reflux. A similar qualitative trend is observed for DC350, but at the bubbling bed surface including the connection to the riser could
much lower S-values; for DC100 the S-value does not change. buffer temporarily higher pressures in the riser, which would
Therefore, the sensitivity of the S-value increases with increasing otherwise result in more slugging-like behavior with increased riser
height of the measuring location in the downcomer. For the bubbling densities.
bed the S-value also does not change, which is consistent with For the adapted operating conditions it has been confirmed that S
findings from van Ommen et al. [6] that in bubbling beds such small is not sensitive to an increase of 10% and 20% in bed mass (run LS 3,
changes in particle size are not detected. not shown here). In contrast to the previous set of operating
The S-values for the riser location pressure measurements are conditions, there are no temporary spikes during mass addition.
shown in Fig. 17. The S-values obtained from the riser respond weakly For the addition of larger sized sand (runs LS 4 a/b) the response of
to the increased particle size (average S ∼ 2); there is no trend of the the S-values for two independent experiments is shown in Fig. 18. We
sensitivity with respect to the measurement position observed. here compare the S-values before the addition of larger sized sand and
In general, a correlation between the S-value and increasing after the final adaptation of the mass reflux, and therefore compare
particle size is observed. In the riser the sensitivity is very small for all two situations with the same overall mass and mass reflux, but larger
locations. In the downcomer the sensitivity increases with higher particle size. First of all, Fig. 18 shows a good reproducibility for both
measurement location. Moreover, one can also observe some experiments. The bubbling bed again shows no significant response
sensitivity to mass reflux changes, considering the changes in the S- whereas all the downcomer positions ultimately react similarly
value due to re-adjustment of the mass reflux. It should be noted that sensitive with S-values of around 6. In the previous case with a
this effect could in principle also have occurred in the L-valve lower mass reflux (run LS 2) there was a strong difference between
configuration (mass reflux not measured). However, the L-valve the sensitivities of the upper and the lower downcomer positions.

Fig. 16. Response of the S-values for changing particle size (from d50 = 235 μm to d50 = 242 μm and d50 = 249 μm), obtained for different locations in the downcomer (run LS 2).
M. Bartels et al. / Powder Technology 202 (2010) 24–38 33

Fig. 17. Response of the S-values for changing particle size (from d50 = 235 μm to d50 = 242 μm and d50 = 249 μm), obtained for different locations in the riser (run LS 2).

Here, no vertical dependency for the sensitivity of the S-value to velocity in the bubbling bed (run LS 4) causes less interference with
particle size is present; moreover, the lower positions DC100 and the downcomer; in addition, the riser pressure fluctuations have a
DC350 are now sensitive enough to detect the applied particle size higher amplitude here due to a higher density/reflux and therefore
changes. A possible explanation for this phenomenon is that the lower propagate stronger into the downcomer. In contrast to the lower

Fig. 18. Response of the S-values for changing particle size (from d50 = 235 μm to d50 = 242 μm and d50 = 249 μm), obtained for different locations in the downcomer and the
bubbling bed. Duplicating experiment, Top: run LS 4 a, Bottom: run LS 4 b.
34 M. Bartels et al. / Powder Technology 202 (2010) 24–38

downcomer positions, the upper position DC850 strongly decreased amplitude) with increased bed density. The sensitivity of the method
in sensitivity as compared to run LS 2; the reason for this decrease is increases with higher mass reflux / riser density, with the highest
not clear. It is also important to notice that in combination with a downcomer position being the only exception to this rule.
lower mass reflux, i.e. before its final adaptation (G(s) b 65 kg/(m2s) in The reductions in mass reflux in these two experiments after addition
Fig. 18), the vertical dependence of the sensitivity is indeed still and removal of sand were both approximately 10–12%. In comparison,
observed. It is only after the re-adjustment back to 65 kg/(m2s) that the reduction in mass reflux for the case of adding the same sized
this vertical dependence disappears. material as a reference case (not shown) has been maximal approxi-
For the riser the S-values for two independent experiments are mately 7%; however, in that case there was no change in the S-value.
shown in Fig. 19. Also for the riser, the reproducibility for both Overall, as already the case for the previous operating conditions
experiments is good. The S-values are in the range of approx. 5–10. one can observe a sensitivity of the S-value to particle size changes
The sensitivity has increased as compared to the previous case of and to mass reflux changes. The resulting S-values in case of a
lower mass reflux and riser density (run LS 2, Fig. 17). Again, there is changing mass reflux here are higher than for the case of adding same
no clear dependency of the sensitivity with respect to height, only sized particles (run LS 3, not shown), but the relative change in mass
R128 is somewhat less sensitive here. This decreased sensitivity could reflux is also higher here. The sensitivity to particle size has also
have to do with entrance-effects of the recycling bed material from increased for the riser and the lower downcomer part, only exception
the downcomer, with R128 being directly opposite to the riser inlet. is the upper downcomer with a sensitivity decrease.
The S-values in the riser now roughly correspond with the S-values In light of the observed sensitivity of attractor comparison with
observed in the riser of the L-valve configuration (Fig. 7 b and c) with particle size changes, it is worthwhile to have a physical interpreta-
the same riser velocity but a reflux of 40 kg/(m2s). The riser densities, tion. For increased particle size at a constant mass reflux the average
however, are roughly similar in both cases. In contrast, for the loop absolute deviation (AAD), a measure of the pressure fluctuation
seal configuration at a reflux of 40 kg/(m2s) both the resulting riser amplitude, increases in the riser whereas in the downcomer and the
density and sensitivity to particle size changes were lower. This bubbling bed it remains constant or only slightly increases. The
indicates that the absolute S-values are related to the resulting riser average cycle time (ACT), a measure for the characteristic time scale of
density rather than the chosen mass reflux. This is understandable due the hydrodynamics, increases for all positions with increasing particle
to the direct relation of more intense pressure fluctuations (increased size. In light of these trends, one can view the resulting riser

Fig. 19. Response of the S-values for changing particle size (from d50 = 235 μm to d50 = 242 μm and d50 = 249 μm), obtained for different locations in the riser. Duplicating
experiment, Top: run LS 4 a, Bottom: run LS 4 b.
M. Bartels et al. / Powder Technology 202 (2010) 24–38 35

correlated, i.e. having a constant phase shift, as a function of frequency. A


coherence of 1 means that the two signals have the same shape and a
constant time shift, although their amplitudes can be different. A
coherence of 0 means that both signals are independent from each
other. With the signals from the three measuring positions the three
different coherences are shown in Fig. 21. The pressure fluctuations from
the different locations of the CFB are strongly correlated at frequencies
below 5 Hz. The high coherence at this lower frequency range generally
can be attributed to global phenomena in the system with a high
propagation velocity, such as gas flow fluctuations, bubble formation,
bubble coalescence, bubble eruption, and bed oscillations [13]. Compared
to the other two combinations, the coherence between the downcomer
and the bubbling bed is higher and decreases only slowly with increasing
frequency. This effect can be expected because of the direct physical
coupling of downcomer moving bed and bubbling bed. In contrast, both
the downcomer and the bubbling bed are more decoupled from the riser
by the gas volume between bubbling bed and riser. One can therefore
generally conclude a frequency-dependent decoupling effect: a strong
decoupling holds for frequencies N5 Hz, for lower frequencies riser and
downcomer are still coupled well.
Fig. 20. Typical raw pressure fluctuations for the lower downcomer position (DC100),
bubbling bed (BB) and riser opposite to the solids return (R189) (run LS 4b). Note the
For run LS 2 with a lower riser density and higher gas velocity in
different vertical scale for the bubbling bed. the bubbling bed (experiment 2) visual inspection of the raw pressure
fluctuations (Fig. 22) already indicates a different situation. The
resemblance between the pressure fluctuation signals is much less
hydrodynamics as somewhat more moving into the direction of compared to the previous case (Fig. 20). The riser signals look
“slugging”. This effect would also be expected due to the presence of qualitatively similar, with slightly decreased amplitude. The bubbling
larger particles (having a larger minimum fluidization velocity) in the bed and downcomer signals, however, look rather different: the
riser in combination with a relatively small diameter riser. downcomer signal resembles to a certain degree that of the bubbling
bed, but has no resemblance with the riser signal anymore.
3.2.3. The decoupling effect of the loop seal The coherences (Fig. 23) confirm the observations from the raw data
With the L-valve configuration, the pressure fluctuations in the concerning the more decoupled behavior between riser and down-
lower part of the downcomer appeared to be strongly related to the comer. As compared to run LS 4 one can observe a decreased coherence
riser fluctuations, resembling them in a dampened form (Fig. 5). The between the riser and both other positions at low frequencies of
time shift in the pressure signals shows that for both L-valve and loop approximately 0–3 Hz. Moreover, one can observe a clear peak at
seal the pressure waves are travelling against the solids flow direction around 5 Hz in all coherence plots; in case of coherence between
upwards the downcomer. downcomer and bubbling bed this effect appears somewhat masked by
The loop seal was anticipated to “decouple” the riser from the an overall high coherence, in the other cases it is clearly visible. From the
downcomer in order to gain more insight in the big difference in power spectra (not shown) this 5 Hz peak is identified to correspond to
sensitivities to changes in particle size (L-valve) and to see how the the dominant frequency in the bubbling bed, but not in the riser at this
application of the method would change. We consider the lower frequency. Together with the high correlation, this means that the 5 Hz
downcomer (DC100), the bubbling bed (BB) and the lower riser signal originates from the bubbling bed and is strongly attenuated as it
(R189) for the investigation of this effect. The remaining positions in propagates into the riser (the coherence is not dependent on
the riser are very similar to the lower riser; the positions higher in the amplitude). The coherence between downcomer and bubbling bed
downcomer resemble the lower downcomer in a dampened form. An has not significantly changed as compared to run LS 4 and is still high
example of a pressure time series for the three locations for the operating due to the more direct coupling via the moving sand bed.
condition with higher riser density and lower gas velocity in the bubbling One should keep in mind that the setup with a loop seal does not
bed (run LS 4b, Figs. 18 and 19) is shown in Fig. 20. The pressure simply decouple the riser pressure fluctuations from the downcomer
fluctuations from the downcomer positions resemble those of the to a higher degree, but also introduces the bubbling bed as an
bubbling bed and the riser to some degree. For a quantitative comparison additional source for pressure fluctuations.
we use the coherence (see e.g. Ref. [12]) here. The coherence is a A change in the degree of decoupling between riser and down-
dimensionless measure that quantifies to which degree two signals are comer could explain the increased sensitivity to particle size changes

Fig. 21. Coherence between pressure fluctuations from the lower downcomer (DC100), bubbling bed (BB) and riser (R189) (run LS 4).
36 M. Bartels et al. / Powder Technology 202 (2010) 24–38

Fig. 24. Development of the differential pressure and the S-value based on the
horizontal return-leg and the lower riser. The vertical line at 1:22 h indicates the point
where the bed defluidized.
Fig. 22. Typical raw pressure fluctuations for the lower downcomer position (DC100),
bubbling bed (BB) and riser opposite to the solids return (R189) (run LS 2). Note the
different vertical scale for the bubbling bed.
reflux as well as disturbances originating from addition and removal
of material. The sensitivity to mass reflux changes is to be expected
for the lower downcomer positions at the high riser density and low because they directly influence the riser density profile. However,
bubbling bed velocity (run LS 4a/b). The pressure fluctuations there was no correlation observed with particle size. It is concluded
measured in the downcomer are more coupled with those in the that simple pressure drop measurements are not suitable for
riser in that case; therefore hydrodynamic changes in the riser will monitoring particle size changes.
have a more significant contribution in the downcomer. This also
implies that the riser is still the main source for information from 3.3. Hot lab-scale CFB with L-value configuration
pressure fluctuations regarding the changing particle size.
Comparing the L-valve configuration (Fig. 7) with the loop seal For the cold-flow setup, attractor comparison is sensitive to small
configuration (Figs. 18 and 19) the sensitivity to particle size changes in changes in particle size and for the presence of larger particles (“model
the riser is very similar in both cases (S-values around 6), whereas the agglomerates”). The next step is to investigate its application during
very high sensitivity in the lower downcomer with L-valve configuration energy conversion under hot conditions. For this purpose we used a lab-
(S-values around 40) is greatly reduced for the loop seal configuration scale, hot circulating fluidized bed setup. The setup has been operated
(S-values around 6). This strengthens the hypothesis that the very high with straw pellets in gasification mode; no recycling or refreshing of bed
sensitivity to particle size changes in the downcomer with L-valve material has taken place during the experiment. By slowly increasing
configuration is indeed a result of the riser pressure fluctuations that are the temperature, agglomeration has been induced in the setup until the
modified (filtered) as they travel from riser to downcomer. bed defluidized. Given the fact that in the downcomer and the return-
Finally, the question remains how the decoupling effect would leg the gas velocities are much lower and mixing is much less vigorous
change when going to larger-scale industrial CFBs. Measurements in an than in the riser, it is most likely defluidization occurs in the moving bed.
industrial size CFB riser, loop seal and downcomer are therefore highly Pressure fluctuation measurements have been carried out at various
desirable to check the scale-dependency of the decoupling hypothesis. positions in riser and downcomer. Unfortunately, many of the
measurement taps were blocked due to insufficient purging and were
3.2.4. Monitoring average riser pressure drop not suitable for the analysis. However, the fact that they got blocked
One could raise the question whether the average pressure drop – confirms the problems related with the developing agglomeration
often a readily available process variable in industrial installations – process. Only two positions were suitable for the analysis, the horizontal
would be sufficient for monitoring particle size and detect agglom- return-leg as well as the lower riser 150 mm above the distributor
eration. In the experiments with the loop seal we have also measured plate. Fig. 24 shows the pressure drop between the downcomer return-
the pressure drop over five consecutive vertical segments in the riser. leg and the lower riser as well as the calculated S-values from the return-
Generally, some correlation was observed with the changing mass leg and the lower riser position. The differential pressure fluctuates

Fig. 23. Coherence between the pressure fluctuations from the lower downcomer (DC100), bubbling bed (BB) and riser (R189) (run LS 2).
M. Bartels et al. / Powder Technology 202 (2010) 24–38 37

considerably, but does dot exhibit any clear trend. At 1:22 h it suddenly
drops to zero at the point of defluidization of the bed. The S-value has
been calculated based on a reference time window just before the
presented data. For the return-leg it first stays rather constant around
zero. At 0:48 h, roughly half an hour before the point of defluidization,
the S-value increases above 3 for the first time. This indicates the onset
of a change in the hydrodynamics, originating from the onset of the
agglomeration process. The S-values from the bottom of the riser are
generally higher once they have crossed 3. Although the strong increase
at 0:48 appears to correspond less to a gradual built-up of agglomerates,
the S-value still yields a correct early warning with the same lead time as
obtained from the S-value from the return-leg.

3.4. Hot industrial CFB with loop seal Fig. 26. The median particle size from sieving analysis (squares) and the filtered S-value
(diamonds) during a period in which the particle size in the industrial combustor
increased.
Attractor comparison has also been implemented in an industrial
CFB boiler. The objective of this part of the work was not agglomeration
detection, but the detection of small changes in the particle size, as those larger than 3 are present, but are mostly explained by other
are relevant for the heat transfer. The particle size has not been temporary process disturbances and were correlated to other effects
artificially modified, but the occurring changes in particle size during via logged process data. The filtering of S-values has shown to
regular operation have been monitored. Such changes can originate significantly reduce the temporary peaks due to other irrelevant
from e.g. changing process conditions or gradual attrition of particles. process events.
The particle size has been determined three times per day via bed During another operating period the particle size slowly increased
sampling and sieving. over the course of about three weeks. At the same time the S-value
The industrial CFB boiler has not been so tightly controlled as the gradually increased, following the trend in particle size (Fig. 26).
lab-scale setups. Especially the boiler load (i.e. the delivered power Within this correlation between the S-value and the median particle
expressed in MW) is changing in a range of about 30% and therefore size one can observe a significant change in S clearly crossing the value
results in changing fluidization velocities. Attractor comparison has of 3 around day 13–15, when the median particle diameter has
been shown to be insensitive to changes in gas velocity within certain increased by approx. 20 μm compared to the reference median
limits (around 10% for bubbling beds [6], around 5% in a lab-scale particle diameter of approx. 240 μm. Please note that the most
circulating bed riser (Fig. 10 and other unpublished results)). As the important result here is a globally increasing trend in the S-value, i.e.
relative changes in this case greatly exceed these limits, we have the S-value correlates with the increasing particle size on longer time-
chosen to work with several reference attractors, each covering a 5% scales. Particularly in industrial units there are various kinds of
load range. In the actual attractor comparison we switched to the process variations and disturbances, such as fuel feed and gas flow
appropriate reference according to the measured load; the load is variations, temporary usage of soot blowers and others. Many cases in
directly correlated to fluidizing velocity. Moreover, in this industrial which the S-values vary unexpectedly can be explained by additional
unit we have chosen to use “filtered” S-values, i.e. taking the minimum process data (not shown here), but some cases cannot be explained
of three consecutive S-values from both measuring positions. Reason with the process data and knowledge available. For the actual
for this additional filtering is the occurrence of several process events implementation of the method additional effort should be put in
during the operation, such as bed mass addition, ash removal and soot increasing the robustness of the method to avoid taking unsuitable
blowing; these events would otherwise cause S to temporarily measures. An important starting point for this will be incorporating
increase beyond 3 more frequently apart from the changes in particle additional process data to account to significant process disturbances.
size. As the objective here was to remove temporary strong process Keeping in mind that the S-value is a measure relative to the
disturbances, we did apply this specific filtering strategy and not chosen reference condition, one could actually not a-priori determine
frequency filtering, as done previously in the cold-flow CFBs. whether the particle size increases or decreases as the S-value only
First, the S-value has been determined for a period of about increases. To distinguish increasing and decreasing particle size
2.5 weeks with rather constant median particle diameter of approx- during operation without simultaneous sampling it is possible to
imately 240–250 μm according to the bed samples (Fig. 25). The S- use several reference attractors (at least two), each at a different
value does not exhibit any trend during this period. Occasional values particle size. One can then match the current attractor with the
corresponding reference from such a database to obtain the currently
present particle size.

4. Conclusions

The attractor comparison method can be used for the detection of


gradual particle size changes as well as the early detection of
agglomeration in various circulating fluidized beds.
In a cold-flow lab-scale circulating fluidized bed (CFB) with an
L-valve for solids return attractor comparison is sensitive to small
changes in particle size. For a median particle size of 235 μm the
detection limit in the riser is a size change of approx. 10 μm. For the
downcomer the method is much more sensitive, with a detection limit
of a size change well below 7 μm for the lower vertical downcomer
section and approx. 2 μm in the horizontal return-leg. Changes in bed
Fig. 25. The median particle size from sieving analysis (triangles) and the filtered S-value mass up to 20% do not disturb the method. Applying a low-pass filter can
(squares) during a period of stationary particle size in the industrial combustor. be beneficial as it greatly reduces sensitivity to fluctuations in bed mass
38 M. Bartels et al. / Powder Technology 202 (2010) 24–38

and gas flow, but only marginally reduces sensitivity to particle size 5.2. Downcomer
changes. The method has also been shown to be sensitive to the
presence of model agglomerates (d50 = 1900 μm). Applying a 30 Hz The analysis based on measurements in the downcomer has shown
low-pass filter has shown to successfully remove sensitivity to changes to be often more sensitive than in the riser. A vertical sensitivity
in mass reflux, especially in the riser. With such a filter, the sensitivities gradient can be present, depending on the operating conditions. It is
from riser, downcomer and return-leg are very similar; the resulting therefore recommended to measure low as well as high in the dense
detection limit is approx. 2–3 wt.% of model agglomerates. bed downcomer, if possible, also to gain more insight on the optimal
With a loop seal for the solids return, attractor comparison is position. In general, it is recommended to further investigate the
sensitive to changes in particle size of 14 μm relative to the median pressure fluctuation signals and the behavior of the attractor
particle size of 235 μm. For the downcomer, there can be a strong comparison method on industrial scale, particularly the influence of
sensitivity gradient, with higher sensitivities at higher measuring the measurement location (riser, downcomer, and return-pipe).
positions; the lower positions in the downcomer do not detect the
imposed changes. At operating conditions with higher mass reflux 5.3. L-valve or loop seal
and riser density, however, this gradient in sensitivity disappears and
all positions become equally suitable. The bubbling bed does not The solids return mechanism in CFBs can be carried out with
detect such small changes in particle size, consistent with [6]. The L-valve or loop seal. In case of an L-valve a measurement position in the
riser is only slightly sensitive to particle size changes when operating horizontal return-pipe is highly recommended; in addition low-pass
with a lower riser density, and more sensitive at higher riser densities. filtering before the analysis of the data might be necessary for robust
Despite a large vertical density gradient in the riser, no significant monitoring. In case of a loop seal configuration small particle size
vertical sensitivity gradients were observed. The method is not changes were not detected in the bubbling bed. Nevertheless, this
sensitive to mass changes up to 20% and not sensitive to relative measuring position is still recommended especially when agglomera-
changes of about 7% in mass reflux. However, the method is sensitive tion should be detected, which is motivated by multiple positive
to a resulting 10–12% decreased mass reflux due to the addition of experiences of agglomeration detection in bubbling beds (e.g. [8]).
larger sized sand, which is thought to originate from settling effects in
the moving bed.
In both the loop seal and L-valve configuration the pressure waves Acknowledgements
are travelling against the solids flow direction in the downcomer. The
application of a loop-seal as compared to an L-valve leads to less The authors thank Roderik Bohlken for his work on the cold CFB
coupling between the downcomer pressure fluctuations and riser within his Master Thesis project, Marcin Siedlecki & Wiebren de Jong
pressure fluctuations. This coupling is quantified by the coherence for their assistance with the hot lab-scale CFB and the Sachtleben
between riser and downcomer pressure fluctuations and is frequency Chemie GmbH for permission to use the presented data from their
dependent: At higher frequencies the coherence is low, whereas for plant.
low frequencies smaller than ∼ 5 Hz (global hydrodynamic phenom-
ena) it is generally high. References
Under actual gasification conditions in a lab-scale CFB with very
[1] P. McKendry, Energy production from biomass (part 3): gasification technologies,
similar geometry as the cold-flow setup, attractor comparison was Bioresource Technology 83 (1) (2002) 55–63.
applied to a real agglomeration case. The S-values from the horizontal [2] J. Koornneef, M. Junginger, A. Faaij, Development of fluidized bed combustion —
return-leg as well as the lower downcomer can be used to detect the an overview of trends, performance and cost, Progress in Energy and Combustion
Science 33 (1) (2007) 19–55.
developing agglomeration process and indicate the approaching
[3] M. Öhman, A. Nordin, B.-J. Skrifvars, R. Backman, M. Hupa, Bed agglomeration
defluidization with an early warning time of about 30 min in advance. characteristics during fluidized bed combustion of biomass fuels, Energy & Fuels
In an industrial CFB boiler with gradually changing particle size 14 (1) (2000) 169–178.
during regular operation, attractor comparison detects gradual [4] H.J.M. Visser, The influence of fuel composition on agglomeration behaviour in
fluidised-bed combustion, ECN (Energy research Centre of the Netherlands)
changes in the median particle size, based on measurements in the report ECN-C–04-054, 2004.
lower dense bed riser. Here the method detects an increase of ∼ 20 μm [5] M. Bartels, W. Lin, J. Nijenhuis, F. Kapteijn, J.R. van Ommen, Agglomeration in
relative to the median bed particle diameter of ∼240 μm. fluidized beds at high temperatures: mechanisms, detection and prevention,
Progress in Energy and Combustion Science 34 (5) (2008) 633–666.
[6] J.R. van Ommen, M.-O. Coppens, C.M. van den Bleek, J.C. Schouten, Early warning
5. Recommendations for Implementation of agglomeration in fluidized beds by attractor comparison, AIChE Journal 46
(2000) 2183–2197.
[7] C. Diks, W.R. van Zwet, F. Takens, J. DeGoede, Detecting the differences between
Based on the results from the different experimental investiga- delay vector distributions, Physical Review E 53 (1996) 2169–2176.
tions, we propose the following guidelines for the implementation of [8] J. Nijenhuis, R. Korbee, J. Lensselink, J.H.A. Kiel, J.R. van Ommen, A method for
the attractor comparison method in CFBs. agglomeration detection and control in full-scale biomass fired fluidized beds,
Chemical Engineering Science 62 (1–2) (2007) 644–654.
[9] Energy technologies at the cutting edge, International Energy Agency (IEA), 2007.
5.1. Riser [10] R. Bohlken, Dynamic Monitoring of Gas–Solid Circulating Fluidized Beds, M.Sc.
Thesis, Delft University of Technology (2007).
[11] D. Christensen, J. Nijenhuis, J.R. van Ommen, M.-O. Coppens, Residence times in
Small particle size changes as well as agglomeration can be
fluidized beds with secondary gas injection, Powder Technology 180 (3) (2008)
detected, with increasing sensitivity at higher riser densities. It will 321–331.
therefore be useful to measure in the lower, dense part of the riser; [12] M.B. Priestley, Spectral analysis and time series, Academic Press LTD, London,
very lean regions in the upper part should not be considered. The 1981.
[13] J. van der Schaaf, J.C. Schouten, C.M. van den Bleek, Origin, propagation, and
sensitivity and detection limits therefore also improve with operation attenuation of pressure waves in gas–solid fluidized beds, Powder Technology 95
conditions of higher riser densities. (1998) 220–233.

You might also like