CFD and Laboratory Studies of Dust Cleaning Ecacy of an Ecient Four Stage Non-Clogging Impingement Filter for Flooded-Bed Dust Scrubbers

Fibrous-type lters are used to capture dust particles in mining and other occupations where personnel are exposed for prolonged periods. Dust cleansing systems including ooded-bed dust scrubbers use these mesh-type multi-layered lters. These lters trap dust particles eciently on their surface and inside their mesh. However, their continued operation leads to dust build-up and subsequent clogging. This results in increased resistance of the lter and lowered airow rate through the scrubber. This could potentially enhance the exposure of the miners. A non-clogging self-cleaning impingement screen type dust lter was designed by the authors for use in mining and industrial dust cleansing applications. The lter guides dirt-laden air through rapidly turning paths which forces it to shed heavier particles. The particles impact one of the impermeable solid metallic lter surfaces and are removed from the airstream. A full cone water spray installed upstream prevents any surface build-up of dust. This paper summaries the computer models generated to show the lter operations and laboratory experiments including optical particle counting to establish the cleaning eciency.


Introduction
Many unit operations in mining generate respirable dust and their concentration varies signi cantly across all those processes (Petavratzi, et al., 2005). This includes mechanical excavation, handling and transportation, and crushing. Respirable coal and rock dust particles could be inhaled and travel into the lungs. These particles get deposited on the lung surface and impede nominal breathing. Prolonged exposure to respirable dust has been conclusively shown to cause irreversible ailments in miners (Joy, et al., 2012;Laney & Weissman, 2014). Coal dust accumulation beyond critical concentration has also resulted in explosion events in underground mines leading to fatalities and loss of operations (Mine Safety and Health Administration, 2001;Mine Safety and Health Administration, 2014). Mine operators, therefore, use a wide variety of remedial measures to combat dust. Primarily, fresh air is sent underground to dilute the dust concentration to lower levels via ventilation systems. Water sprays effectively direct the dust-laden air away from the miners. Underground mines also use physical barriers to protect them from elevated dust levels. Underground operations including continuous miner sections use a combination of these methods. Continuous miners have an onboard fan-powered ooded-bed dust scrubber that treats the coal dust-laden air close to the active face (Colinet, et al., 2013). The scrubber has an inlet that captures the dirty air close to the miner drum. A multi-layered brous type dust lter is the primary cleaning element that traps dust particles on its undulating surface (Gupta, et al., 2021). A water spray is installed upstream of the lter and mounted at 45º to the air ow direction to keep the lter surface drenched and prevent any build-up. Cleansed air is discharged after traveling through a demister which removes water droplets from the air. Wet scrubbers were also developed in the 1970s and could capture solid and liquid particles e ciently (Eckert & R.F. Stringle, 1974).
Research on 20-layered brous-type dust lters showed that the cleaning e ciency of the lter improved with an increase in air ow and water ow through the lter (Kumar, 2018;Velasquez, 2019). Fibrous lters, however, get clogged during operations leading to dust build-up (Thomas, et al., 2001). This increases the lter resistance and lowers the air ow (Kanaoka & Hiragi, 1990). A study by Song et al. using monodispersed polystyrene latex particles showed that initial particle collection occurs inside the lter bed. Filter resistance increases with clogging and ltration tends to occur on the front edge (Song, et al., 2006). Yue et al. studied the ltration process through random three-dimensional brous lters using computational uid dynamics modeling and discrete element methods. They showed that particles were initially captured by the bers. The ltration process led to blockage by deposited particles which increased the resistance (Yue, et al., 2016). Agranovski and Shapiro studied the clogging process due to dust particles in wet and dry lters. They observed that the resistance of the dry ltration process was about half of the wet ltration resistance. Non-wettable lters were also found to get rapidly blocked (Agranovski & Shapiro, 2001). Computer models have shown that the compressibility of aerosol particles has a signi cant impact on lter clogging with trapped compressible particles leading to early lter replacement (Stöcklmayer & Hö inger, 1998). Experimental studies on lter media used in industrial air cleaning by Callé et al. showed that the ltration e ciency of the media increased during the collection phase and diminished in the cleaning phase. The residual pressure drop was also found to increase overloading and cleaning cycles (Callé., et al., 2002). Research by Japuntich et al. showed that lter loading by different size particles contributes to pressure drop on all stages of loading (Japuntich, et al., 1994). Laminger et al. developed mathematical models to demonstrate an increase in residual pressure drops over cycles of ltration and regeneration operations. The model assumed the particle deposition area on the lter as a signi cant factor that governed the ltration mechanism (Laminger, et al., 2016).
Fotovati et al. generated numerical models for pleated lters, which are also used most in industrial applications, and showed that pressure drop is minimum for an optimum pleat count. Triangular pleats were also found to result in a lower pressure drop (Fotovati, et al., 2011). The air ow capture e ciency of the clogged lter, in all the cases discussed here, is diminished which might lead to an elevated exposure of the miners to dust. Therefore, research to develop non-clogging e cient dust lters is critical to enhancing the health of miners and the safety of operations.
Filters that do not use brous type irregular mesh type lters could be e cient alternatives and the dust particles are made to follow a pre-de ned trajectory and guided towards impermeable surfaces where they could be trapped. This follows the mechanism of particle impactor systems, designed by Marple in the 1960s, frequently used for particle size classi cation and analysis (Marple & Willeke, 1967). These devices force air between parallel plates with slits. This change in direction forces aerosol particles to be cast out and hit the plates where they are trapped. The spacing within successive plates also diminishes which assists in capturing smaller particles on the plates downstream. Impingement-type lters are also very e cient in capturing aerosol particles (Dahneke, 1971).
There has not been signi cant research in this area leading to industrial applications in mining or similar industries that deal with dust exposure. A self-cleaning bar-type screening unit working on a vibrating mechanism was patented in 1969. This was designed to screen silica particles measuring 325 mesh (0.004 in.) size (Pearsali, 1969). Higginbotham was granted a patent in 2007 for his invention of a nonclogging screen lter, though it was designed to prevent clogging of sewage systems and dealt with large particles (Higginbotham, 2007). Cyclone-type scrubbers were also studied and found to be e cient in removing particles from the airstream (Krames & Buttner, 1994). The University of Kentucky designed Vortecone to capture overspray paint particles of size up to 300.0 µm on automobile painting lines (Tanigawa, et al., 2008). A reduced scale model of the Vortecone was tested and its excellent e ciency in capturing coal dust particles from the airstream was established . Parametric studies of geometrical features were carried out to redesign the lter later to lower the resistance and to improve its dust-laden air capture e ciency (Taylor, et al., 2019).
However, the vortex-type complex design of the Vortecone would require a thorough redesign of systems. A ooded-bed dust scrubber could use the proposed thin lter as an alternative to the brous lter.
Preliminary research on a reduced-scale prototype of an impingement lter showed its e cacy . The authors have designed and investigated the performance of a four-screen lter with perforated staggered openings to remove particles from the airstream. This paper describes the design and performance testing of the non-clogging lter using computational uid dynamics modeling and laboratory testing using optical particle counting.

Filter Description
A four-stage lter with 1.5 mm thick at impingement screen surfaces was designed. All screens had long rectangular openings 6.0 mm in width and 78.0 mm in length. The rst and third screen surfaces are identical in design. The second and the fourth surfaces are identical and have their rectangular openings displaced in the screen plane by 6.0 mm. Filter surfaces were separated by spacers measuring 3.0, 2.0, and 3.0 mm in length, respectively. The total thickness of the lter, therefore, is 14.0 mm. This con guration makes the lter perfectly blind to straight ows. Dust-laden air is forced to follow constricted turns between all the lter surfaces. Heavier dust particles are drawn out of the airstream due to their inability to change directions suddenly. These particles impact the lter surfaces due to much higher momenta and are removed from the air stream. The lter will be installed at 45º to the general air ow direction. The physical prototype of the lter is shown in Fig. 1. Air ow patterns are shown in the following paragraphs.

Computational Fluid Dynamics Modeling
Computation uid dynamics (CFD) modeling analysis was done to optimize the performance parameters as the geometrical dimensions were altered iteratively during the design stage. CFD modeling is a ow volume discretization technique wherein the Navier-Stokes equations of uid dynamics are solved using numerical approximation. ANSYS Fluent software, version 18.2 was used to generate the CFD models. The structure of the ow domain was drawn in AutoCAD and imported into ANSYS Fluent. The integrity of the uid ow region was ascertained in the design modeler module. Inlet and outlet segments were demarcated and assigned unique names. A vertical symmetry plane was drawn to bifurcate the ow volume. This created a much smaller analysis region, saved on computing resources, and allowed the generation of ne meshes to resolve the air ow patterns around the lter. Figure 2 shows the geometry used for analysis.
Boundary conditions for this setup included ux and wall functions. The inlet surface was assigned a normal velocity value of 3.80 m/s. Corresponding turbulence intensity and turbulence viscosity values were computed. The outlet surface was imparted a static pressure of 0.0 Pa. Surfaces of the symmetry planes were assigned free-slip symmetry plane conditions while all other surfaces were imparted wall functions.
The SIMPLE scheme was used for pressure-velocity coupling. The least-squares cell-based gradient calculation method was followed.

Meshing and grid independence studies
The multi-zone method meshing technique was followed and the ow domain was split into three connected volumes. The volume around the lter had tetrahedral elements while the ones towards the inlet and the outlet had polyhedral elements. This enabled a high resolution of ow close to the lter and saved on element count. A dense mesh was a prerequisite since micron-size dust particles had to be tracked in the transient-state simulations later to determine the particle removal e ciency. Three meshes with increasing packing density were generated. The meshes had about 10.98, 15.72, and 21.84 million grid elements. The total pressure drop was computed for an incoming airspeed of 3.80 m/s. Microsoft Excel solver was used to compute absolute and relative errors. The mesh parameters are shown in Table 1. Extrapolated errors in velocity and ne grid convergence indices were computed using Richardson's method and are shown in Table 2.
Relative, extrapolated, and ne grid convergence index decreases with an increase in grid packing density indicating good grid convergence. The second mesh with 15.72 million elements was used for further simulations. CFD models were generated on windows based high-performance computer.

Steady and transient-state simulations
Steady-state simulations were run until the solution converged according to the pre-assigned conditions. Models were generated and run until convergence in ow parameters was achieved. Figure 3 shows the contours of velocity magnitude on a plane in the ow domain for an air ow rate of 0.71 m 3 /s. Air was observed to accelerate between the lter surfaces which will assist in forcing particles towards one of the surfaces leading to their capture. Steady-state simulations were run for different air ows and corresponding total pressure drops were computed. These were compared to the pressure drop obtained in the laboratory experiments and were found to be in excellent agreement as discussed later.
Transient-state simulations were developed to determine the pro le of particles' sizes that impact the lter surface under known air ows. An array of spherical aerosol particles of known diameters adjusted using Cunningham's slip correction factor was generated at the domain inlet for 0.10 s (Cunningham, 1910). An initial velocity was assigned to the particles based on the average air ow rate through the lter. Particles were tracked for a period of 1.00 s. The lter surfaces were programmed to trap the particles.
Particles underwent an inelastic collision with all other surfaces with a coe cient of restitution value of 0.95. Particles were counted on the inlet and the lter surface. These particle counts were used to compute the particle removal e ciency of the lter from the air. These results are included along with the results obtained from laboratory testing.

Experimental Setup And Procedure
Filter resistance and cleaning e ciency are the two critical parameters that de ne its performance. A prototype made of aluminum was constructed to determine these parameters and to validate the computer models. Spacers measuring 3.0, 2.0, and 3.0 mm were used to separate the successive at lter surfaces. The lter measured 14.0 mm in thickness.
A test set up as shown in Fig. 4 was constructed. The experimental set-up consisted of a system of ducts with a centrifugal fan attached to it. The fan was operated using a variable frequency drive (VFD) that enabled setting up any air ows precisely. The duct measured 0.30 m x 0.45 m in cross-section and was 2.40 m in length to allow a well-developed ow pro le. This was followed by a Dwyer pressure measurement station used to obtain total and static pressures in the duct. This also had honeycomb-like structures that served as the ow straighteners. Vane and rail arrangements were built around the corners to encourage a good air ow pro le and to minimize the shock losses. The lter was installed downstream of the bend in a duct with an identical cross-section area and length. A 1.20 m long duct was built with transparent polycarbonate sheets to install the lter at 45º to the air ow direction like a oodedbed dust scrubber set up. All lines of contact were sealed to prevent any leakages. A full cone water spray was installed upstream of the lter to inject water normal to the lter plane as shown in Fig. 5. A ow control valve and an inline digital owmeter were installed in series to control the water ow rate precisely.

Steady-state testing in the dry state
Tests were rst run to determine the lter resistance. VFD frequency was set at 10.0 Hz and increased in steps of 5.0 Hz. Total and static pressure at different air ows were measured. Velocity pressure was calculated and was used to determine the air ow through the lter. These steps were repeated three times to obtain a good representative average. A curve of best t through these points was drawn as shown in Fig. 6 that represents the ow-pressure drop curve showing a quadratic dependence of pressure drop on volumetric ow rate for the lter. Filter resistance using this curve was obtained to be 1.675 kNs 2 /m 8 . An adjusted R 2 value of 0.9998, a standard error value of 0.0034, and a t-value of 496.82 indicate an excellent t. Once the fan system curve was constructed, the duct discharge was connected to the dust exhausting system of the laboratory to minimize exposure of the researchers to coal dust particles while running cleaning e ciency experiments. Velocity in the duct was computed again with the water spray running using the pressure measurement station to determine the air ow through the lter.

Isokinetic sampling for aerosols
Two identical TSI optical particle sizers (OPS 3330) were used for air ow sampling. Air was sampled isokinetically upstream and downstream of the lter. This is a technique in which aerosol particles are extracted from the airstream without altering the air ow speed in the vicinity of the sampling nozzle (Wilcox, 1956). If not done properly, this could lead to under-sampling or over-sampling of the aerosol particles leading to inaccurate results. Sampling nozzles were designed, and 3D printed for air ows of 0.47, 0.71, and 0.94 m 3 /s in the duct.
Keystone Mineral Black 325 A and limestone particles were used to study the cleaning e ciency of the dust lter. This coal dust specimen has a known particle size distribution and has been used to investigate scrubber performance earlier.
This coal dust specimen has a known particle size distribution and has been used to investigate scrubber performance earlier. An Arduino-controlled stepper motor-assisted dust injection system was built to inject dust upstream of the lter. TSI optical particle sizer (OPS) 3330 device was used to count and size the aerosol particles. Densities of coal (1,220 kg/m 3 ) and limestone dust (2,200 kg/m 3 ) were programmed in the OPS. The complex refractive index of coal (1.76 -i 0.60) and limestone (1.50 -i 0.005) were programmed to run experiments with suitable particles. The dead time correction parameter was enabled to minimize errors due to coincidence due to many particles in the sample. Dust particles were injected into the pressurized duct using compressed air. A sensitive digital weighing balance was used to precisely measure 3.2 gm of coal dust and 3.5 gm of limestone dust for each run of the tests. Dust particles were introduced into the duct for 5.0 minutes. The controlled injection of dust particles into the duct ensured that none of the OPSs was overwhelmed with dust particles. This effectively eliminated the coincidence error which would have resulted in an incorrect particle count and size. Iso-kinetic sampling nozzles were designed with dimensions computed using air ow velocity magnitudes, 3D printed and attached to the tube. The duct was traversed precisely to locate points where the average air ow speed equaled the speed reported by the pressure measurement station.

Cleaning e ciency results
Particle concentration by count data obtained from the two OPSs installed upstream and downstream of the lter were converted into gravimetric concentrations using the aerosol instrument manager software. Figure 7 shows one such plot of the gravimetric concentration of coal dust particles upstream and downstream of the lter for an air ow of 0.94 m 3 /s. The difference in gravimetric concentration upstream and downstream of the lter is the cleaning e ciency under the operating conditions. These were computed for particle sizes 2.00 µm and above for the three air ows. Figure 8 shows the cleaning e ciency of the impingement lter for coal dust particles. The plot shows that the e ciency of the removal of dust particles from the airstream increases with an increase in air ow through the lter. Bigger and heavier particles are removed signi cantly more compared to the smaller particles that tend to follow the airstreams and might escape the lter. Figure 9 shows the cleaning e ciency using limestone particles. Water ow was kept steady at 7.57 l/min (2.0 gpm) for these tests. The impingement type lter captured removed bigger particles. Figure 10 compares the particle removal e ciency from the air obtained from transient-state CFD models and laboratory experiments. The computer models agree well with the laboratory testing results.

Conclusions
Micron size dust particles produced in mining and other industries are detrimental to the health of the workers and the safety of the operations. These industries use brous-type dust lters to capture dust particles. These are e cient over a wide range of particle diameters. However, these are maintenance intensive and get clogged due to particles getting caught on the surface and inside the brous layers.
This enhances the lter resistance, increases the pressure drop, and leads to a much lower volume of air being cleaned. This severely affects the operational e ciency of the machines using these lters.
A non-clogging lter was designed based on the aerosol particle impaction principle. The lter has 6.0 mm wide and 78.0 mm long openings. These large openings allow the impingement lter to clean the dirt-laden air without the need for any maintenance arising out of clogging, unlike the conventional brous lters that tend to get clogged rapidly. Particles are removed from the dust-laden air by forcing them to negotiate sharp bends. At higher air ows, high momentum possessed by the particles casts them out of the airstream and forces them to impact one of the screen surfaces. The inability of the heavier particles to change directions suddenly leads to their removal from the air stream. The non-clogging nature of the lter ensures that there will not be any cyclic jump in pressure drop due to lter clogging as observed in all brous lters. Therefore, this lter will continue to operate at a known point for prolonged periods. This design also allows engineers to design an impingement type lter for a known air ow rate or to maximize the capture of particles of a speci c size. This lter also does not require a high water ow rate, unlike the brous lter in which the brous surface is required to be kept ooded to capture the particles e ciently. So, the ow rate was kept constant at a relatively low value of 7.57 l/min (2.0 gpm).
Since this lter works on the particle impaction principle, its cleaning e ciency will improve further at higher air ows. Denser aerosol particles will also be removed more e ciently due to higher mass and momentum. The dimension of the openings, their separation, and screen spacings impact the pressure drop and cleaning e ciency at known air ows. Since the impingement lter will not get clogged, need minimal maintenance, and will have higher mechanical availability. These lters could be designed for known air ows and can be a good replacement to the conventional brous lters in mining operations.

Declarations
Funding Velasquez, O., 2019. In uence of water injection rate on the Vortecone, an impingement screen, and a conventional lter screen cleaning e ciency, Lexington, KY: s.n.