Powder Technology 291 (2016) 284–298

Contents lists available at ScienceDirect

Powder Technology

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

Numerical and experimental study of pulse-jet cleaning in fabric filters

B.O. Andersen a,⁎, N.F. Nielsen a, J.H. Walther b,c
a
FLSmidth A/S, Airtech, Vigerslev Allé 77, 2500 Valby, Denmark
b
DTU Mechanical Engineering, Technical University of Denmark, Nils Koppels Alle, Building 403, room 216, 2800 Kgs. Lyngby, Denmark
c
Computational Science and Engineering Laboratory, ETH Zurich, Clausiusstrasse 33, CH-8092, Zurich, Switzerland

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

Article history: Pulse-jet cleaning and understanding of the complex physics are essential when designing fabric filters used for
Received 30 July 2015 air pollution control. Today, low-pressure cleaning is of particular interest due to demand for reduced com-
Received in revised form 25 November 2015 pressed air consumption. Pulse-jet cleaned fabric filters have been studied for many years by experimental inves-
Accepted 21 December 2015
tigation and to a limited extent by Computational Fluid Dynamics (CFD). The majority of the studies have focused
Available online 30 December 2015
on high-pressure cleaning systems, and the CFD models presented are so far two-dimensional (2D). In the work
Keywords:
presented here, pulse-jet cleaning of low-pressure fabric filters (2 bar) is studied using a full three-dimensional
Computational fluid dynamics (3D) CFD model. Experimental results obtained in a pilot-scale test filter with 28 bags, in length of 10 m and in
Fabric filter general full-scale dimensions of the cleaning system are used to verify the reliability of the present CFD model.
Baghouse The validated CFD model reveals the strong compressible effects, a highly transient behaviour, the formation of
Pulse-jet cleaning compressible vortex rings and the shock cell phenomenon within the overexpanded supersonic jet. The cleaning
Low-pressure cleaning nozzles and venturi design aid or oppose the pulse-pressure within the bags, and this plays an important role in
Design optimisation the resulting efficiency of removing the dust layer from the bags. The CFD simulation shows that the traditional
straight-bore nozzles provide substantial misalignment of the jet, and the add-on nozzle design offers only lim-
ited improvement. Further, the need for venturis in low-pressure filters and the importance of optimising the
venturi design are demonstrated. The working principle of the venturi is to restrict backflow which is detrimental
to the pressure rise in the bags. Reducing the venturi throat diameter is shown to reduce backflow and improve
the pulse-pressure.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction the flow reversal and high resistance in the fabric cause an internal pres-
sure increase called the pulse pressure. As a result, the bag is briefly in-
Pulse-jet cleaned fabric filters are commonly used for air pollution flated and the dust cake is removed (Fig. 3 right).
control in many industries, e.g. power production and mining. Dust In recent years, interest in low-pressure pulse-jet cleaning has in-
transported by the flue gas is collected on the external surface of the fab- creased in the search for reduced energy consumption. Low-pressure
ric bag, thereby forming a so-called ‘dust cake’. Periodic removal of the filters operate at a tank pressure of 2–3 bar gauge, whereas traditional
dust cake is required due to the continuous build-up of dust. Pulse-jet high-pressure filters operate at much higher pressure, typically 4–7
cleaning is widely used for this purpose as it enables frequent cleaning bar gauge. This relatively new technology is expected to introduce chal-
whilst the filter is operating. lenges not found in traditional high-pressure systems. Of particular in-
In pulse-jet cleaning, a short pulse (50–150 ms) of compressed air is terest is the potential axial misalignment between pulse-jet and
released by a valve and distributed in a purge tube to multiple nozzles venturi/bag.
(Fig. 1). Each nozzle is directed towards a venturi placed above the In low-pressure filters, the lower density of the compressed air is ex-
open end of a single bag. The compressed air expands through the noz- pected to cause higher axial velocity in the purge tube and therefore in-
zle, thereby forming a pulse-jet of primary air. In the near-field region creased jet misalignment.
around the pulse-jet, secondary air entrains the pulse-jet as it travels to- Understanding the jet behaviour and the interaction with the ventu-
wards the venturi (Fig. 2). The pulse-jet acts in countercurrent direction ri and bag is crucial to design effective pulse-jet cleaned fabric filters.
to the flow of flue gas during normal filter operation (Fig. 3). In cleaning The introduction of low-pressure filters poses an even greater challenge
mode, the pulse-jet travels through the venturi and into the bag where to engineers, requiring a detailed insight into the highly dynamic fluid
flow. Obtaining this valuable information is difficult through conven-
⁎ Corresponding author.
tional methods relying on physical measurements, which often fall
E-mail address: bjoa@flsmidth.com (B.O. Andersen). short in terms of spatial as well as temporal resolution. Computational
URL: http://www.flsmidthairtech.com (B.O. Andersen). Fluid Dynamics (CFD) is highly suited for this purpose as it offers the

http://dx.doi.org/10.1016/j.powtec.2015.12.028
0032-5910/© 2016 Elsevier B.V. All rights reserved.
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 285

Fig. 1. Schematic view of the pulse-jet cleaning system.

level of detail needed at a fraction of the cost compared to physical cleaning, the most important being: nozzle diameter, distance between
experiments. nozzle and bag opening, pulse duration, initial tank pressure and vol-
There are a number of studies in the literature on pulse-jet cleaning, ume [1–7]. Bakke [1] identified the jet and venturi concept as essentially
although the majority of the work has focused on high-pressure filters. a jet pump and found that the pulse pressure developed in the bag is in-
Previous research has found that several parameters affect pulse-jet versely proportional to the flow rate entering the bag. Lu and Tsai [2–4]
found that increasing the bag resistance leads to an increase in pulse
pressure. Increasing the nozzle diameter was found to increase the
pulse pressure until a certain limit, after which a larger diameter will
be detrimental; hence an optimum nozzle diameter exists. Similarly,
an optimal distance between nozzle exit and bag opening exists. A lon-
ger distance allows for more entrainment of secondary air, but if the dis-
tance is too long, the jet width will exceed the diameter of the bag or
venturi opening. A mathematical model for optimising this distance
was later developed by Qian et al. [8], who studied high-pressure filters
(6 bar) without venturis.
The role of the venturi is not fully covered in the literature and is a
subject of debate. According to Morris et al. [9], the role of the venturi
is to allow the pulse to travel easily into the bag, while restricting its
escape and thereby increasing the pressure within the bag. The provi-
sion of a high pulse pressure was found to be incompatible with a low
pressure loss through the venturi during filtration. Removal of the ven-
turi caused significant reduction of the pulse pressure. Lanois and
Wiktorsson [10] compared the performance of ‘advanced’ filters at
1–2 bar tank pressure without venturis to ‘traditional’ filters at
4.8–6.2 bar with venturis and found that advanced filters require
lower energy for the equivalent cleaning efficiency. In the study by Lu
and Tsai [4], two venturis were tested against a venturi-less design.
The necessity of the venturi was found to depend on the combined re-
sistance of the bag and dust cake. For high resistance bags, maximum
pulse pressure was obtained with venturi and vice versa. The venturi
role was also studied by Hájek [11], who found a three-fold increase
in the mass flow rate entering the bag when installing a venturi,
claiming increased entrainment of secondary air to be the reason. Fur-
ther, a two-fold increase in pulse pressure was found, demonstrating
the unambiguous advantage of the venturi. Whether the venturi is ben-
eficial is indeed a subject of debate and is perhaps best summarised by
Lu and Tsai [3]: “There are situations where venturis are required to in-
Fig. 2. Illustration of the jet pump principle showing primary air from the nozzle and crease pressure pulse inside the bag, and there are also situations where
secondary air entraining the jet from the surroundings. venturis are not necessary.”
286 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

Fig. 3. Schematic view of the bag, steel cage and venturi assembly showing dust cake build-up during normal filtration (left) and dust dislodgement during pulse-jet cleaning (right).

Li et al. [12] studied the effect of installing a cone-shaped structure 2. Methods

inside the bag just below the bag opening without the use of venturis.
Two different cones were tested against a cone-less design, with both 2.1. Experimental methods
cones leading to an increase in bag pressure as well as peak dust emis-
sion during pulse-jet cleaning. The experimental pilot-scale test filter is shown in Fig. 4 with speci-
The literature offers only a limited number of CFD-based studies on fications listed in Table 1.
pulse-jet cleaning. Hájek [11] studied the role of the venturi by using a The set-up consists of a tank with four internal valves, four purge
simplified CFD model consisting of a single bag and straight-bore nozzle tubes each with 18 equally spaced straight-bore holes connected by a
with and without venturi. A two-dimensional (2D) mesh was applied, reinforced rubber connection. For each purge tube, seven holes and
thereby restricting the jet to be axially aligned with the bag, and thus bags are placed inside the filter casing making up the test section. The
neglecting the possible misalignment between jet and venturi/bag. Un- 11 remaining holes in each purge tube are placed outside the casing
steady RANS and k-ε turbulence modelling was applied. Lo et al. [7] without bags. Each bag is held in place by an internal steel cage as
studied pulse-jet cleaning of a single pleat in a pleated fabric cartridge shown in Fig. 3. The fit between bag and cage is very tight, allowing
by CFD. As in the study by Hájek [11], the geometric simplification en- for less than 5% volume change when the bag is inflated. A venturi de-
forces axial symmetry disregarding the possible axial misalignment be- vice is placed at the opening of each bag, and a tube sheet separates
tween the jet and venturi/bag. Unsteady RANS with the k-ε turbulence the dust zone from the clean zone. All tests are conducted without
model was adopted and isothermal flow assumed despite the non- dust. Instead, the presence of a dust cake is simulated by using special
negligible Mach number and hence the potential cooling due to expan- bags with low permeability, providing exactly the same pressure loss
sion. Experimental measurements were used as inlet condition at the as a conventional bag with dust cake just before cleaning.
nozzle and subsequently for validation of the CFD model. Good agree- Pressure measurements are performed experimentally at seven lo-
ment with experimental results was demonstrated except during pres- cations: in the tank and at three positions in each of bag-1 and bag-7
sure ramp-down, this being ascribed to variable filter permeability from (top, middle and bottom). A pressure transmitter is mounted directly
fabric deformation, and the assumption of constant uniform filter per- in the tank, whereas 3.5 m long PTFE hoses are used between the mea-
meability was questioned. Various CFD models for simulating the filtra- surement location and the transmitter for all six bag measurements.
tion flow through pleated filters were assessed by Feng et al. [13]. For All tests are performed at room temperature and with the filter in
high pleat density, LES, DES or v2f RANS was required to predict pres- off-line mode, i.e. without filtration flow. Pulse-jet cleaning is per-
sure loss and flow field, whereas for low pleat density the k-ε model formed with an initial tank pressure of 2 bar gauge and 50 ms electrical
performed as well as the more computationally costly models. on time (EOT).
The present study concerns CFD and experimental investigation of
the pulse-jet cleaning system. Experimental pressure measurements 2.2. Numerical methods
are used as validation data for the full three-dimensional (3D) CFD
model, which in turn is used to investigate the basic physics of subson- The flow during pulse-jet cleaning is characterised as being both su-
ic/supersonic pulse-jet cleaning, the jet behaviour and the effect of the personic and subsonic. Internally in the purge tube and within the
venturi in relation to a typical pulse-jet cleaned fabric filter. Basic under- pulse-jet cores, the supersonic flow is dominated by inertial forces,
standing of the physics and the jet and venturi behaviour is essential for compressibility and the presence of pressure and shock waves. In the
optimal cleaning of filter bags, and it remains to be ascertained how pulse-jet shear layer and within the bags, the flow is subsonic and tur-
these separate parts affect the cleaning system. Detailed results of the bulent, whereas the flow is laminar through the filter medium. Since
full 3D CFD model include jet alignment, pulse pressure inside the turbulence only plays a role in certain areas, the computationally costly
bags and the effect of the nozzle and venturi design. For further results methods for accurately resolving the turbulent scales (e.g. LES, DES) are
based on the present CFD model, the reader is referred to Andersen [14]. considered unnecessary. Viscous effects are present in, for example, the
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 287

Fig. 4. Schematic illustration of the experimental pilot-scale test filter: (a) valve, (b) compressed air tank, (c) reinforced flexible rubber connecting piece, (d) purge tube with 18 nozzles,
(e) venturi, (f) bag, (g) tube sheet, (h) casing, (P1)–(P7) pressure measurement. See Table 1 for full description of components.

pulse-jet shear layer; hence the inviscid assumption often used for su- segregated or uncoupled, is suitable for shock waves and highly com-
personic flows cannot be applied. The unsteady RANS formulation pressible flows [15,16]. The coupling applies to all three conservation
(URANS) is therefore considered the optimal compromise. The coupled equations as described below.
method of solving the Navier–Stokes (NS) equations, as opposed to the The commercial code STAR-CCM+ is used to simulate the complex
flow in the pulse-jet cleaning system. For detailed information on fluid
flow and computational methods, the reader is referred to the STAR-
Table 1
Components of the experimental pilot-scale test filter as illustrated in Fig. 4.
CCM+ user guide [15] and literature, e.g. [16,17].

Object Type
2.2.1. Governing equations
a Valve Trimec piston valve 3 in
The governing equations are the NS equations including conserva-
b Tank Volume 225 L
c Flexible connection Reinforced rubber tion of: mass (continuity), momentum (Newton's second law) and en-
d Purge tube 18 nozzles, total length ≈3.6 m ergy (first law of thermodynamics). The conservation of mass is
e Venturi 78 mm (throat) written for compressible fluids in Eq. (1). The Cartesian form and the
f Bag Nomitec polyester weave low permeability, 5 Einstein notation are used throughout this article.
L/dm2/min @200 Pa, 127 mm ≈ 5 in
g Tube sheet Steel plate
h Casing Steel plates
∂ρ ∂ðρui Þ
þ ¼0 ð1Þ
Pressure measurement Type ∂t ∂xi
P1 Tank BD Sensors model DMP 331, 0–10 bar, 4–20 mA
P2 Bag-1, top, 0.5 m below FilTEq GmbH model DS2-420 Platine, −60 to 60
tube sheet mbar, 4–20 mA where ρ is the density, t is the time, ui is the velocity vector and xi is
P3 Bag-1, middle, vertical as P2 the spatial vector [16]. The conservation of momentum, neglecting
centre of bag gravitation, is given by [16]:
P4 Bag-1, bottom, 0.5 m above as P2
bag bottom
P5 Bag-7, top, 0.5 m below as P2

tube sheet ∂ðρui Þ ∂ ρu j ui ∂τij ∂p
þ ¼  ð2Þ
P6 Bag-7, middle, vertical as P2 ∂t ∂x j ∂x j ∂xi
centre of bag
P7 Bag-7, bottom, 0.5 m above as P2
bag bottom
where τij is the viscous and turbulent stress tensor and p is the pressure.
288 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

The conservation of energy is given by [17]: of turbulence model is expected to have only minor influence. The k-
  ω SST turbulence model by Menter [21] has also been applied (again
∂h ∂h ∂p ∂p with compressibility modification); however, no significant change in
ρ þ ui ¼ þ ui
∂t ∂xi ∂t ∂xi the result was found and will therefore not be treated further.
ð3Þ
∂ ∂T ∂u Previous numerical studies were limited by axial symmetry, thereby
þ þ τij i disregarding any possible axial misalignment between jet and venturi/
∂xi ∂xi ∂x j
bag [7,11]. In order to simulate the effects of having multiple nozzles
where h is the enthalpy, k is the thermal conductivity and T is the tem- in one purge tube and to fully capture the spatially complex flow field
perature. For compressible flows, the density follows from the continu- in the nozzle and jet, a 3D model is required. Further, experimental val-
ity Eq. (2). The temperature can be computed from the energy Eq. (3). idation is of utmost importance, and the geometry of the numerical
With both density and temperature known, the pressure can be com- model must therefore closely resemble the experimental set-up. The
puted from the equation of state where the ideal gas law is applied [17]: computational domain is depicted in Fig. 5 with the different boundary
conditions highlighted in colors.
p ¼ ρRT ð4Þ The bags are simulated as having zero thickness with a porous inter-
face applied on the cell faces acting as a membrane allowing fluid to
R denoting the gas constant. pass whilst experiencing a pressure loss defined by:

2.2.2. Numerical models Δp ¼ ρβvn ð5Þ

Turbulence is modelled by the realisable k-ε model developed by
Shih et al. [18] using a second-order upwind convection scheme. The where Δp denotes the pressure drop in Pa, ρ is the gas density in kg/m3,
near wall region is treated by the two-layer wall function by Rodi [19]. β is the porous viscous resistance in m/s and vn is the boundary-normal
To account for compressibility, the compressible dissipation dilatational velocity in m/s. This linear expression is suitable during normal opera-
term by Sarkar and Lakshmanan [20] is added to the turbulence trans- tion (filtration) where the flow across the filter is dominated by viscous
port equation. Due to the mainly sonic nature of the flow, the choice forces due to the very low velocity vn ≈ 1 m/min = 0.0167 m/s where

Fig. 5. Computational domain in transparent view. The central part of the filter compartment is omitted for clarity. The boundary types are indicated by colours: orange: outlet, blue:
symmetry, green: porous baffle (bag), cyan: porous baffle (valve), grey: wall. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 289

Fig. 6. Upper: input for controlling the valve porous baffle resistance. Figure showing porous viscous resistance, β in Eq. (5), as a function of time. Lower: time derivative of tank pressure
(experimental) as a function of time. Used to determine the valve opening time.

the pressure drop across the filter is approximately 1000 Pa. Given the [7], the bag resistance varies as the bag is deflected, particularly during
bag peak pressure of 2000–5000 Pa during cleaning, the equivalent in- the pressure ramp-down, which is of less interest in this study. The bags
crease in vn (in reverse direction) is considered sufficiently low to still are modelled as rigid structures neglecting the potential effect of bag
assume viscous dominance. Previous studies show that bag resistance deflection. While bag deflection will affect the pulse pressure, the de-
has a major effect on the peak pulse pressure obtained within the bag flection in the pilot-scale test filter is very small (see Section 2.1);
[1,2,4,7]. Adjusting the value of β to obtain a proper fit with the present hence the assumption is considered valid.
experimental data is therefore highly important. For the bags, β is as- The valve is modelled similarly to the bags, but with temporal varia-
sumed to be constant spatially and temporally. According to Lo et al. tion of the resistance coefficient, β, to represent the characteristics of the

Fig. 7. 3D visualisation of the mesh. For clarity, the geometry is mirrored along the symmetry plane.
290 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

closing phase is of less interest in this study; hence the same value is ap-
plied here. The entire opening time (i.e. phases 1 through 3) is estimat-
ed at 125 ms from the experimental tank measurement, leaving 83 ms
for the fully opened phase. When the simulation time of 125 ms is
reached, the porous interface is converted to a wall hindering any ex-
change of mass. The simulation is continued until 250 ms. Valve resis-
tance values are determined through trial and error.

2.2.2.1. Spatial and temporal discretisation. The presence of shock waves

suggests applying an explicit time-stepping approach. However, as the
duration of the pulse-jet is long relative to the jet speed, this will require
undesirable small time steps, and the implicit method is adopted in-
stead. Between two time steps, inner iterations are performed based
on the default Courant number C = uΔt/Δx = 50, where Δx denotes
the grid spacing, which is converted into an inner iteration step size
by the software [15].
The inner iterations continue until one of the following two condi-
tions are met: (1) the non-dimensional continuity residual has reached
an absolute value of 2 × 10-7 or (2) 15 inner iterations are performed. In
essence, the choice of time scheme and settings is a trade-off between
numerical accuracy, stability and computational speed. First-order tem-
poral discretisation is used to increase stability when shocks are present
[16], although it was found to be quite insensitive to modifications of
solver settings (including time step and inner iteration Courant
number).
The computational domain is composed of unstructured polyhedral
cells with sizes specified in certain regions. High spatial resolution is ap-
plied to regions where large gradients are found as shown in Fig. 7. Four
meshes with varying spatial resolution are constructed as shown in
Fig. 8. Properties and computational times for the four meshes are listed
in Table 4. A constant time-step size of Δt = 1 ×10-5 s is used, resulting
in different maximum convective Courant numbers for the four meshes.
The shock front speed was found to be independent of the mesh res-
olution; hence no refinement is applied internally in the purge tube and
bags. However, the flow field in the jet and nozzle, including the near-
field region, was found to be mesh sensitive, due to the larger gradients
Fig. 8. 3D visualisation of the four mesh levels. Coarse mesh (upper left), Medium mesh present there. The Coarse mesh strongly overpredicts the width of the
(upper right), Fine mesh (lower left), Extra Fine mesh (lower right). View in the axial jet which affects the flow in the venturi and ultimately the pulse pres-
direction of the purge tube. For clarity, the geometry is mirrored along the symmetry
plane.
sure. The Extra Fine mesh provides an overly resolved jet and very
good agreement with experimental results. Based on the quality of the
results compared to the computational effort, the Fine mesh is chosen
actual valve. The operation of the valve is simplified by three phases: for the majority of the further analyses.
(1) valve is nearly closed: high resistance; (2) valve is fully open: low
resistance but non-zero; and (3) valve returns to closed position: high 2.2.2.2. Boundary conditions. Pressure outlets with a total pressure equal
resistance. The progress between the three phases is approximated by to 0 Pa (relative to the reference pressure) are applied in the top box and
a linear variation of the resistance as illustrated in Fig. 6 with timings filter compartment to simulate an infinitely large volume on both sides
determined from the experimental results. Fig. 6 also shows the time of the tube sheet. During pulse-jet cleaning, entrainment of secondary
derivative of the experimental tank pressure measurement. The transi- air from the top box plenum causes a pressure reduction of the enclosed
tion from zero-gradient to peak-gradient is 21 ms, which corresponds to volume in the top box and an increase in filter compartment pressure as
the time it takes the valve to go from fully closed to fully open. The also indicated by Hájek [11]. For small filters, this will greatly influence

Fig. 9. Geometry of the three venturi configurations. From left to right: small venturi (throat 60 mm), large venturi (throat 78 mm) and no venturi (127 mm).
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 291

Fig. 10. Add-on nozzle geometry.

the pulse pressure, but for larger filters and when including the volume numerical model are built to study the flow field and pressure depen-
of adjoining ducts, the effective volume is large and the approximation dency upon the venturi. A configuration essentially without venturi
of infinite volume thus becomes realistic. Inflow may occur on the pres- and two variations of the venturi throat diameter are studied (Fig. 9).
sure outlet boundaries in which case a temperature of 300 K is applied. The baseline geometry features a simple straight-bore nozzle
A symmetry boundary condition is applied along the vertical plane suspected to cause misalignment between jet and venturi due to the
through the centreline of the bags and purge tube. This approximation flow direction in the purge tube and the relatively small wall thickness.
is suitable during the majority of the time domain, where the high ve- A commercially available add-on nozzle (Fig. 10) claiming to improve
locity flow is dominated by inertial forces and the jet is highly stable. alignment is studied in comparison to the simple nozzle.
The possible asymmetry during jet build-up/decay is not considered.
The wall type boundary condition with the no-slip condition is ap- 3. Results and discussion
plied to the remaining surfaces as indicated in grey in Fig. 5. Wall
boundaries are assumed adiabatic. 3.1. Pulse-jet flow characteristics

2.2.2.3. Initial conditions. The computational domain is initialised by ap- The flow field in a pulse-jet cleaning system is inherently different
plying a pressure of 2 bar to the tank volume and 0 bar to the remaining from traditional jet theory. The latter is beyond the scope of the present
parts of the domain, both relative to the reference pressure of 1.01325 study. Several factors complicate the flow field, the most important
bar. A temperature of 300 K is applied throughout the domain. Bidirec- being: strong compressible effects, highly transient behaviour, imper-
tional pressure waves will emanate from the valve from the beginning fect nozzle design and non-uniform upstream flow conditions. For sim-
of the simulation (t = 0) due to the porosity having a finite value. plicity, the jet characteristics are divided into two phases:
The tank volume is widely known to influence the pulse pressure ob-
1. The subsonic phase during jet ramp-up is dominated by large instabil-
tained within the bags [2–4,6,7,22] and is therefore adjusted according-
ities, the formation of a vortex ring and strong dynamic fluctuations
ly to 225 L in the numerical model taking symmetry into account. The
(Figs. 11 and 12).
shape and tank internals are assumed insignificant; for simplicity, a
2. The supersonic phase is quasi-steady with velocities exceeding the
square box is used.
local speed of sound (Fig. 13).
2.2.3. Model applications After the supersonic phase, the jet returns to the subsonic phase dur-
Following the validation by experimental pressure data, the numer- ing ramp-down, although without the formation of a vortex ring.
ical model is modified to study different applications, including venturi A compressible vortex ring structure (Fig. 11) is formed in the shear
and nozzle design. The venturi is known to influence the pulse pressure layer between the jet and the surrounding stagnant air during the sub-
obtained within the bag as well as the pressure loss across the venturi sonic jet build-up when the jet momentum is relatively low and the jet
during filtration [1,3,4,9,11,22–24]. Three configurations of the is subject to a Kelvin–Helmholtz-like instability. This is consistent with

Fig. 11. Formation of a compressible vortex ring structure during the subsonic phase of the jet visualised by an isosurface of the λ2 criterion with isovalue =-1×10-7. For definition of the λ2
criterion, see Chakraborty et al. [32]. Results obtained with the Extra Fine mesh.
292 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

Fig. 13. Shock cells visualised by the local Mach number in the jet when at its maximum
speed. Results obtained with the Extra Fine mesh.

Fig. 12. Vorticity in the y-direction (out-of-plane) and vector field showing the
dynamically oscillating jet in the subsonic phase. Results obtained with the Extra Fine
mesh. (Fig. 14), good agreement between numerical and experimental results
is seen as a result of the valve control scheme applied (Fig. 6), though
the simulation tends to slightly overpredict the emptying of the tank.
the findings by Lim [25] and Murugan et al. [26]. Subsequently, more When the valve opens at t = 0 s, the tank pressure decreases rapidly
vortex rings are formed, but due to the rapid breakdown, these are until around t = 0.04 s. This is a result of the shock wave propagating
less visible and not well-defined. The instabilities cause the jet to fluctu- through the purge tube and being reflected multiple times until the
ate dynamically (Fig. 12). tank and purge tube pressure approach each other. From t = 0.05 s,
The vortex rings and instabilities occur only in the subsonic emptying of the tank takes place at an almost constant rate limited by
phase, when the jet momentum is low. In the supersonic phase, the the pressure difference across the valve and nozzles.
jet velocity exceeds the sonic limit locally and, depending on the spatial In bag-1 and bag-7, the agreement between numerical and experi-
discretisation, a characteristic flow pattern occurs. This is visualised in mental results are found to be satisfactory (Figs. 14 and 15). At the top
Fig. 13 showing the local Mach number when the jet is at its maximum location, a very high pulse pressure of 3800 Pa is reached within a
speed. The phenomenon also known as shock cells or shock diamonds is short time in both bag-1 and bag-7, though the simulation tends to be
the result of interaction between shocks within the supersonic jet due most accurate in bag-7, top, rather than bag-1, top. Wiggles found dur-
to the jet being overexpanded. This causes strong fluctuations of the ve- ing the pressure ramp-up in bag-1, top, are the result of several factors,
locity, temperature, pressure and density within the jet core, which de- including the valve resistance control scheme (Fig. 6) and pressure
crease in strength further from the nozzle as the jet is compressed and wave reflections within the purge tube. Even if present in the experi-
eventually reaches ambient pressure. The pattern is similar to the find- ment, these are not visible in the experimental data due to the 3.5 m
ings by Ooi et al. [27] and Panda and Seasholtz [28], but differs due to measurement hoses, which dampen such rapid fluctuations.
the fact that this jet is not aligned axially with the nozzle. Accurate The top and bottom locations in bag-1 and bag-7 all exhibit distinct
simulation of shock cells requires the use of extremely fine spatial periodic oscillations. The complex frequency patterns are the result of
discretisation, and the Extra Fine mesh has been applied here for illus- oscillation induced into the bag from multiple sources, including:
trative purpose. (1) the pulse-jet pressure wave travelling down through the bag at
the local speed of sound and being reflected at the bottom by the steel
3.2. Validation by pressure measurements plate, and (2) strong fluctuations in the pulse-jet velocity caused by
pressure and shock waves present within the purge tube. This means
The numerical model is validated by comparison of numerical that several frequencies are present in the bag. E.g. in bag-1, top
results and experimental pressure data from the tank and bag-1 (Fig. 14), the period of 0.054 s (18.6 Hz) and the total travel length of
(Fig. 14) and bag-7 (Fig. 15) as a function of time. For the tank pressure 2 × 9.5 m fit well with the speed of sound at room temperature. The
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 293

Fig. 14. Static pressure as a function of time in tank and bag-1 showing experimental and CFD results. The Fine mesh and βbag =35,000 m/s are used.

periodic oscillation is also found at the bottom location, but with a phase At the bottom location, the peak pressure increases to a level slightly
shift of 0.032 s. At the middle location of both bags, this periodic oscilla- lower than at the top location as the pressure wave is reflected in the
tion is less visible. The lack of strong periodicity might be caused by bag bottom where a steel plate is mounted. This is contrary to the find-
wave cancellation and the fact that multiple frequencies are present in ings by Lu and Tsai [2] and Yan et al. [5]. The discrepancy is ascribed to
the bag. the steel plate, the high-resistance bag material used in the present
The peak pressure drops with the axial distance as the pressure study and the fact that no dust is removed; hence the bag resistance re-
wave travels axially downwards through the bag and part of the pulse mains high throughout the duration of the cleaning.
travels outwards through the bag. This is consistent with the findings During the pressure ramp-down (after t = 0.125 s), the numerical
by Lim [2] and Murugan et al. [5]. At the middle location, the peak pres- model overpredicts the pressure in the bags at all locations. This was
sure reaches only ≈ 2000–2300 Pa, which is still sufficient for bag also found by Lo et al. [7], who ascribed the discrepancy to a variable
cleaning. According to Löffler [24], 400–500 Pa is needed for good bag bag resistance due to, e.g., deformation of the fibre structures within
cleaning. the filter medium upon being inflated. Lo et al. [7] suggest that the
294 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

Fig. 15. Static pressure as a function of time in bag-7 showing experimental and CFD results. The Fine mesh and βbag =35,000 m/s are used.

constant resistance coefficient is a poor assumption, which is fully con- misalignment is reduced but the jet core is located in the outermost
sistent with the findings of the present study. Accurate modelling of the quarter of the venturi throat and remains there throughout the duration
bag resistance by applying variable coefficients is required to obtain a of the pulse-jet. This reduces the jet momentum in the axial direction of
better match. This is, however, not considered in the present study. the bag and, most critically, allows for a backflow zone occupying up to
Despite some deviation during pressure ramp-down, the agreement be- half of the throat area.
tween CFD and experimental results is generally good, and the model is For the add-on nozzle, the ability to realign the jet is limited. At t =
considered validated. 0.020 s, the magnitude of the misalignment is comparable to the
straight-bore nozzle, but the direction has changed towards the oppo-
3.3. Nozzle type site side of the venturi. This is due to the non-optimal internal geometry
of the add-on nozzle causing a large separation zone before the nozzle
The two nozzle types mentioned in Section 2.2.3 are tested numeri- exit. At t = 0.039 s, the misalignment has improved slightly compared
cally and compared in Fig. 16 showing the velocity vector field and mag- to the straight-bore nozzle. At t = 0.050 s, the jet is well aligned, and
nitude in nozzle 1. Three time steps are shown: t = 0.020 s when the the jet width fills the venturi throat and thereby eliminates backflow.
misalignment is most pronounced, t = 0.039 s at the maximum jet To quantify the effect of adding the nozzle, the results of the peak
speed and t = 0.050 s when the alignment is best. pressure are listed in Table 2. An overall peak pressure improvement
The straight-bore nozzle is found to cause misalignment between of 6–17% by installing the add-on nozzle is seen, with the biggest im-
the jet and venturi at the nozzles near the tank (see also Fig. 13, provement at the middle and bottom locations where the peak pressure
Section 3.1), whereas the nozzles far from the tank provide good align- is lowest. The mass flow rate into the bag and the overall pressure level
ment. This is caused by a reduction in the mass flow rate and hence ve- after the peak pressure are increased even further (not shown) by the
locity along the length of the purge tube due to the flow out of the add-on nozzle. The cleaning mechanism is not fully understood, but
nozzles. In the following, nozzle 1 is of main interest due to the strong several authors point towards peak pressure and mass flow rate being
misalignment. good indicators of cleaning effectiveness [3,8,24,29–31]. The 6–17%
During the filling of the purge tube (t = 0.020 s), the jet impinges on peak pressure increase and the even bigger increase in the mass flow
the edge of the venturi entrance. Upon filling (t = 0.039 s), the rate are likely to cause improved cleaning, given that the bag is not
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 295

Fig. 16. Velocity vector field and magnitude at nozzle 1. Time progressing from left to right. Upper figures: straight-bore nozzle; lower figures: add-on nozzle.

completely cleaned by the straight-bore nozzle. Further, the peak nozzle was found also to increase the consumption of compressed
pressure increase makes it possible to reduce the tank pressure air by 7% in each pulse, which reduces the saving potential offered
and/or compressed air consumption whilst maintaining sufficient by this nozzle although the correlation is not linear. The increased
cleaning intensity. Despite having equal diameters, the add-on consumption is ascribed to the lower resistance offered by the add-
on nozzle geometry (Fig. 10) compared to the straight-bore nozzle.
Table 2
Nozzle comparison based on bag-1 peak pressure. 3.4. Venturi type
Straight-bore Add-on Increase
The effect of the venturi is analysed by studying the gas velocity in
Nozzle Nozzle [%]
the axial direction through venturi-1 (Fig. 17), the net mass flow rate
[kPa] [kPa]
entering bag-1 (Fig. 18) and the peak pressure in bag-1 (Table 3) for
Top 3.74 3.96 6 the three different venturis mentioned in Section 2.2.3 (Fig. 9). Fig. 17
Middle 2.38 2.60 10 shows that the venturi throat diameter has a direct effect on the gas
Bottom 2.44 2.84 17
flow entering and exiting the bag and thus also on the bag cleaning.
296 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

Fig. 17. Axial velocity and velocity vector field for the three venturi configurations. From left to right: small venturi (throat 60 mm), large venturi (throat 78 mm) and no venturi (127 mm).
Note: Clipping is disabled and some values exceed the velocity scale bar.

For the large venturi (78 mm), the combination of a large throat and in the present study, although the improvement by adding the small
misaligned jet allows for substantial backflow exiting the venturi. or large venturi is less than a factor 2. The small difference between
Decreasing the throat diameter (60 mm) is found to reduce the the small and large venturi points towards both venturi designs being
backflow zone (Fig. 17) and increase the net mass flow rate entering close to a possible optimal throat diameter.
bag-1 (Fig. 18). Consequently, the peak pressure is increased by 7–9%, To evaluate the suitability of a venturi design, the pressure loss dur-
which is a modest improvement. ing filter operation must be taken into account. Ultimately, the choice
Removing the venturi (127 mm) allows for significantly increased should be based on a cost optimisation by estimating the power con-
backflow and reduced net mass flow entering the bag. Consequently, a sumption required for cleaning and filter operation, respectively.
significant reduction of the peak pressure by 30–34% is seen.
In conclusion, the working principle of the venturi is to restrict back- 4. Conclusions
flow which is detrimental to the pulse pressure obtained in the bag.
These findings are consistent with [9], who state that the role of the ven- The flow in a pulse-jet cleaned fabric filter has been investigated nu-
turi is to allow the pulse to travel easily into the bag, while restricting its merically using the commercial CFD code STAR-CCM +. A full 3D CFD
escape and thereby increasing the pulse pressure. Both Morris et al. [9] model simulating the transient pulse-jet cleaning has been developed
and Lu and Tsai [4] found that decreasing the venturi throat diameter in- and validated by experimental pressure measurements. The validated
creases the pulse pressure, which is also the conclusion in the present CFD model has been used to investigate the basic physics of low-
study. Hájek [11] quantified the effect of adding a venturi and found a pressure pulse-jet cleaning (2 bar), jet misalignment and the effect of
factor 2 peak pressure increase. Again, this complies with the findings venturi and nozzle design.

Fig. 18. Mass flow rate through nozzle and venturi-1 as a function of time for the three venturi configurations.
 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298 297

Table 3 been validated by experimental pressure measurements and is capable

Venturi comparison based on bag-1 peak pressure. of capturing details in the flow not previously seen. It has been shown
Venturi Large Small Increase No venturi Increase that 3D unsteady RANS simulation is suitable for optimising the pulse-
[kPa] [kPa] [%] [kPa] [%]
jet cleaning system design in fabric filters.

Top 3.74 4.06 9 2.45 −34

Middle 2.38 2.58 8 1.64 −31
Bottom 2.44 2.61 7 1.71 −30 References

[1] E. Bakke, Optimizing filtration parameters, J. Air Pollut. Control Assess. 24 (12)
(1974) 1150–1154, http://dx.doi.org/10.1080/00022470.1974.10470027.
Experimental measurements of the tank pressure and pulse pressure [2] H.C. Lu, C.J. Tsai, Numerical and experimental study of cleaning process of a pulse-jet
within two bags in a pilot-scale test filter were recorded during pulse- fabric filtration system, Environ. Sci. Technol. 30 (11) (1996) 3243–3249, http://dx.
jet cleaning and used for validation of the CFD model. Good agreement doi.org/10.1021/es960020u.
[3] H.C. Lu, C.J. Tsai, A pilot-scale study of the design and operation parameters of a
between experimental and CFD measurements in the tank was found as
pulse-jet baghouse, Aerosol Sci. Technol. 29 (6) (1998) 510–524, http://dx.doi.
a result of the valve being modelled as a porous interface with resistance org/10.1080/02786829808965587.
varying in time. Key features such as the pulse pressure ramp-up, peak [4] H. Lu, C. Tsai, Influence of design and operation parameters on bag-cleaning perfor-
mance of pulse-jet baghouse, J. Environ. Eng. 125 (6) (1999) 583–591, http://dx.doi.
and oscillations within the two bags were captured well by the CFD
org/10.1061/(ASCE)0733-9372(1999)125:6(583).
model, showing good agreement with experimental measurements. [5] C. Yan, G. Liu, H. Chen, Effect of induced airflow on the surface static pressure of
The CFD model was found to overpredict the pulse pressure during pleated fabric filter cartridges during pulse jet cleaning, Powder Technol. 249
pressure ramp-down. Despite this deviation, the agreement between (2013) 424–430, http://dx.doi.org/10.1016/j.powtec.2013.09.017.
[6] L.M. Lo, D.R. Chen, D.Y. Pui, Experimental study of pleated fabric cartridges in a
CFD and experiment was generally good, and the model is considered pulse-jet cleaned dust collector, Powder Technol. 197 (3) (2010) 141–149, http://
validated. dx.doi.org/10.1016/j.powtec.2009.09.007.
The pulse-jet was found to show highly transient behaviour whilst [7] L.M. Lo, S.C. Hu, D.R. Chen, D.Y. Pui, Numerical study of pleated fabric cartridges dur-
ing pulse-jet cleaning, Powder Technol. 198 (1) (2010) 75–81, http://dx.doi.org/10.
cleaning. During the subsonic ramp-up of the jet, large instabilities 1016/j.powtec.2009.10.017.
were observed in the shear layer between the jet and the surrounding [8] Y. Qian, Y. Bi, Q. Zhang, H. Chen, The optimized relationship between jet distance
stagnant air, leading to the formation of several compressible vortex and nozzle diameter of a pulse-jet cartridge filter, Powder Technol. 266 (2014)
191–195, http://dx.doi.org/10.1016/j.powtec.2014.06.004.
rings. Subsequently, the jet became supersonic, clearly showing shock [9] K. Morris, C. Cursley, R.W.K. Allen, The role of venturis in pulse-jet filters, Filtr. Sep.
cells in the overexpanded jet. 28 (1) (1991) 33–36, http://dx.doi.org/10.1016/0015-1882(91)80039-8.
The effect of the nozzle design was investigated by testing a straight- [10] G. Lanois, A. Wiktorsson, Current status and future potential for high-ratio fabric fil-
ter technology applied to utility coal-fired boilers, Proceedings of the 1st Conference
bore nozzle typical for fabric filters and an add-on nozzle with equal exit
on Fabric Filter Technology for Coal-Fired Power Plants, 1982.
diameter. The straight-bore nozzle was found to cause substantial mis- [11] J. Hájek, Computational fluid dynamic simulations in thermal waste treatment tech-
alignment between the pulse-jet and the bag at the nozzles near the nology — design, optimisation and troubleshooting, Energy 33 (6) (2008) 930–941,
http://dx.doi.org/10.1016/j.energy.2007.11.010.
tank. This was due to the high axial velocity in the purge tube and the
[12] J. Li, S. Li, F. Zhou, Effect of cone installation in a pleated filter cartridge during pulse-
relatively low wall thickness, which is most pronounced for low- jet cleaning, Powder Technol. 284 (2015) 245–252, http://dx.doi.org/10.1016/j.
pressure filters. Axial velocity in the purge tube, and therefore misalign- powtec.2015.06.071.
ment, decreased along the length of the purge tube due to the flow out [13] Z. Feng, Z. Long, Q. Chen, Assessment of various CFD models for predicting airflow
and pressure drop through pleated filter system, Build. Environ. 75 (2014)
of the nozzles. An add-on nozzle designed to correct this showed 132–141, http://dx.doi.org/10.1016/j.buildenv.2014.01.022.
limited ability to realign the jet due to the non-optimal internal design [14] B.O. Andersen, CFD modelling of pulse-jet cleaning in fabric filters(Master's thesis)
of the add-on nozzle. The pulse pressure within the bag was increased Technical University of Denmark, 2014.
[15] CD-adapco. User Guide STAR-CCM+ Version 8.04, 2013.
by 6–17% at the cost, however, of a 7% increase in compressed air [16] J. Ferziger, M. Perić, Computational Methods for Fluid Dynamics, third ed. Springer
consumption. Limited, London, 2002 (ISBN 9783540420743).
The role of the venturi was investigated by studying the gas velocity [17] F. White, Viscous Fluid Flow, McGraw-Hill Series in Mechanical Engineering, third
ed.McGraw-Hill Higher Education, 2006 (ISBN 9780071244930).
through the venturi, the net mass flow rate into the bag and the peak [18] T.H. Shih, W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new k-epsilon eddy viscosity model
pulse pressure in the bag for three designs: a small throat venturi for high Reynolds number turbulent flows: model development and validation,
(60 mm), a large throat venturi (78 mm) and no venturi. A large venturi Technical Report NASA-TM-106721, NASA, 1994.
[19] W. Rodi, Experience with two-layer models combining the k-epsilon model with a
throat diameter allowed for substantial backflow exiting the venturi,
one-equation model near the wall, 29th Aerospace Sciences Meeting. AIAA-91-
particularly in combination with the misaligned jet. Reducing the ven- 0216, American Institute of Aeronautics and Astronautics, 1991http://dx.doi.org/
turi diameter restricted the flow from escaping the bag, thereby increas- 10.2514/6.1991-216.
[20] S. Sarkar, B. Lakshmanan, Application of a Reynolds stress turbulence model to the
ing the peak pulse pressure by 7–9%. Removal of the venturi allowed for
compressible shear layer, AIAA J. 29 (5) (1991) 743–749, http://dx.doi.org/10.
a significant increase in backflow, a reduced mass flow rate into the bag 2514/3.10649.
and a 30–34% reduction in the peak pulse pressure. In conclusion, the [21] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering appli-
working principle of the venturi is to restrict the flow from escaping cations, AIAA J. 32 (8) (1994) 1598–1605, http://dx.doi.org/10.2514/3.12149.
[22] I. Schildermans, J. Baeyens, K. Smolders, Pulse jet cleaning of rigid filters: a literature
the bag and thereby maximise the mass flow and peak pressure in the review and introduction to process modelling, Filtr. Sep. 41 (5) (2004) 26–33,
bag. http://dx.doi.org/10.1016/S0015-1882(04)00234-4.
In summary, the present work highlights the importance of under- [23] E. Rothwell, Pulse-driven injectors for fabric dust filters III: comparative perfor-
mance of model and commercial assemblies, Filtr. Sep. 27 (5) (1990) 345–349,
standing the complex physics of pulse-jet cleaning and the challenges http://dx.doi.org/10.1016/0015-1882(90)80366-S.
introduced by low-pressure cleaning. The present 3D CFD model has [24] J. Sievert, F. Löffler, Fabric cleaning in pulse-jet filters, Chem. Eng. Process. Process
Intensif. 26 (2) (1989) 179–183, http://dx.doi.org/10.1016/0255-2701(89)90010-X.
[25] T.T. Lim, On the role of Kelvin-Helmholtz-like instability in the formation of turbu-
lent vortex rings, Fluid Dyn. Res. 21 (1) (1997) 47, http://dx.doi.org/10.1016/
Table 4 S0169-5983(96)00059-7.
Number of cells, maximum convective Courant number and computation time on a 16- [26] T. Murugan, S. De, C. Dora, D. Das, Numerical simulation and PIV study of compress-
core Sandy Bridge HPC node. *NB: refinement at nozzle/jet 7 is disabled for the Extra Fine ible vortex ring evolution, Shock Waves 22 (1) (2012) 69–83, http://dx.doi.org/10.
mesh to limit the cell count. 1007/s00193-011-0344-9.
[27] A. Ooi, T. Rochwerger, A. Risborg, C. Harkin, J. Soria, M. Li, et al., Investigation of the
Mesh Coarse Medium Fine Extra Fine flow structures in supersonic free and impinging jet flows, J. Fluids Eng. Trans. ASME
Number of cells 591,484 650,453 920,635 1,441,787* 135 (3) (2013)http://dx.doi.org/10.1115/1.4023190.
Courant number 1.7 3.5 7 14 [28] J. Panda, R.G. Seasholtz, Measurement of shock structure and shock–vortex interac-
tion in underexpanded jets using Rayleigh scattering, Phys. Fluids 11 (12) (1999)
Computation time [h] 23 25 33 56
3761–3777, http://dx.doi.org/10.1063/1.870247.
298 B.O. Andersen et al. / Powder Technology 291 (2016) 284–298

[29] R. Dennis, J. Wilder, D. Harmon, Predicting pressure loss for pulse jet filters, J. Air [31] R. Klingel, F. Löffler, Influence of cleaning intensity on pressure drop and residual
Pollut. Control Assoc. 31 (9) (1981) 987–992, http://dx.doi.org/10.1080/00022470. dust areal density in a pulse jet fabric filter, Filtech conference 1983, pp. 306–314.
1981.10465316. [32] P. Chakraborty, R.J. Balachandar, R.J. Adrian, On the relationships between local vor-
[30] W. Humphries, J. Madden, Fabric filtration for coal-fired boilers: dust dislodgement tex identification schemes, J. Fluid Mech. 535 (2005) 189–214, http://dx.doi.org/10.
in pulse jet filters, Filtr. Sep. 2 (21) (1983) 40–44. 1017/S0022112005004726.