Numerical Simulation and Experimental Validation for the Filtration Performance of Fibrous Air Filter Media with LB Method

The performance of fibrous air filter media is one of the key factors to influence the indoor air quality. In order to develop low-resistance and high-efficiency media, it is necessary to obtain a realistic media model. Based on the truncated log-normal model, the D2G9 scheme of Lattice Boltzmann (LB) method was applied to predict both the resistance and the filtration efficiency of the flow through microscale porous media. The influences of the boundary conditions were investigated. The slip and no-slip boundary conditions on fiber surfaces, as well as the non-equilibrium extrapolation scheme and the periodic scheme, were included in this study. By validating the simulated results with both the experimental and empirical values, it was found that the simulated resistance with the slip boundary on fiber surface and the periodic scheme on upper/lower walls was closer to the real value. The model established in this paper and the LB method presented here provide support for further research on optimization of the fibrous air filter media.


Abbreviations
distribution function at position r and time t along the direction α ( , ) eq f t  r equilibrium distribution function at position r and time t along the direction α F B Brownian force per unit particle mass, (m s -2 ) F D drag force by fluid per unit particle mass, (m s -2 ) k B Boltzmann constant, (J K -1 ) Kn Knudsen number L filter media thickness, (m) N the wall normal coordinate pointing into the fluid N mesh number of a characteristic length NY number of lattices along y direction p in static pressure at inlet of media, (Pa) p out static pressure at outlet of media, (Pa) Q flow rate through the filter media, (m 3 •s -1 ) R gas constant, (J mol -1 K -1 ) R f average diameter of fibers, (m) r b a parameter to measure slippage degree (0 ≤ r b ≤ 1) t time, (s) T absolute temperature, K u p particle velocity, (m s -1 ) u s slip velocity, (m s -1 ) U inlet velocity, (m s -1 ) Greek Letters δ t discrete time step, (s) τ relaxation time, (s) τ p relaxation time of particle (= ρ p •d p •d p •C s /18μ, s) μ dynamic viscosity of the gas, (kg m -1 s -1 ) υ kinematic fluid viscosity, (m 2 s -1 ) δ x discrete lattice step, (m) ρ 0 average density of the fluid, (kg m -3 ) ρ p particle density, (kg m -3 ) α solid volume fraction ξ Gaussian random variable Δt time step for the movement of particles, (s) Δp pressure drop through fibrous filter media, (Pa) Δp no-slip pressure drop with no-slip boundary, (Pa) Δp slip pressure drop with slip boundary condition, (Pa)

INTRODUCTION
Air filtration is a major concern in many sectors of industry such as hospital, clean operating rooms and automotive cabin (Kim et al., 2016).Fibrous air filter is currently one of the most economical and effective ways, which has been widely applied.The performance of filter media is crucial to the performance of filter, since it is related to the energy consumption and efficiency of HVAC system.It is therefore essential to be able to predict the performance of the fibrous air filter media with appropriate models.
There are many analytical, numerical and empirical models that have been developed for fibrous media made up of circular fibers (Tronville and Rivers, 2005;Hosseini and Tafreshi, 2011;Joubert et al., 2011;Fotovati et al., 2012;Qian et al., 2013;Zhou et al., 2013;Xu et al., 2014).Some researchers have found that fluids flowing past fiber surfaces experience a slip which could reduce the flow resistance in a microscale channel.The slip phenomenon in porous media has been vastly studied in recent literatures (Zhao and Povitsky, 2009;Hosseini and Tafreshi, 2010;Chai et al., 2011;Hosseini and Tafreshi, 2011;Kirsh and Shabatin, 2015).
Recently, the Lattice Boltzmann equation (LBE), a discrete version of the continuous Boltzmann equation which can capture non-equilibrium gas flows, has been considered as a promising numerical approach for microscale gas flows (Tang et al., 2005;Kim et al., 2008;Przektop and Gradon, 2008;Agarwal et al., 2009;Chen et al., 2009;Wang et al., 2009;Neumann and Rohrmann, 2012;Cho et al., 2013;Bang and Yoon, 2014).The gas flow in the microscale channels has a big difference from the conventional situation.Kn number is a parameter used to determine the rarefaction of gas flow and whether there exists slip phenomenon on the surface.Based on Kn number, the flow could be classified into the regimes as follows, i.e., the continuum regime (Kn ≤ 0.001), the slip flow regime (0.001 < Kn ≤ 0.1), the transitional regime (0.1 <Kn ≤ 10), and the free molecular regime (Kn > 10) (Zheng et al., 2012).When the flow enters the slip flow regime, the slip phenomenon occurs.In other words, the slip velocity should physically increase with the increase of Kn.The slip effects imposed at a solid-fluid interface were well captured by the current LB method.Two mesoscopic kinetic boundary conditions are widely used to realize the slip boundary conditions, i.e., the combination of bounce-back and specular reflection (BSR) scheme (Przekop and Gradoń, 2008;Guo et al., 2002;Succi, 2002;Tang et al., 2004;Sbragaglia and Succi, 2005) and the discrete Maxwellian (DM) scheme (Ansumali and Karlin, 2002;Niu et al., 2004;Tang et al., 2005;Zheng et al., 2012).Guo et al. (2002) also pointed out that BSR and DM schemes are useful for different applications.Zheng et al. (2012) analyzed three kinds of microboundary schemes, i.e., BSR, combination of Maxwell and bounce-back (MB) and combination of Maxwell and specular reflection (MR) schemes, to show how to choose the combination coefficients in these mathematical expressions.Yamamoto et al. (2009) and Yamamoto and Yamauchi, (2013) investigated the flow field and particle capture process in diesel particulate filter by using LBM.They found that after particle deposited on filter, the flow field was different from the initial condition.Przekop and Gradoń (2011) studied the non-steady-state process of particle capture in the fibrous media, where the dentrite shape formed on fiber surface by capture particles was investigated.By applying LBM, Rebaï et al. (2011) compared the random fibrous media models with single fiber diameter and dual fiber diameters.They found that under fixed flow rate and pressure drop, the filtration performance of fibrous media with larger porosity and larger weight was better than that with less fibers.Wang et al. (2012) investigated the initial pressure drop and efficiency of multiple circular fibers, as well as the particle loading process.Later they studied the parallel and staggered fibrous media models (Wang et al., 2013).They found that the performance of staggered model is better than that of parallel model.Huang et al. (2016) investigated the performance of the noncircular fibers, including pressure drop and collection efficiency in the diffusion dominant regime with LBM.They found that the pressure drop is related to the aspect ratio and the orientation angle.
However, the investigations of slip phenomenon on fiber surface and its influence on filtration performance of filter media are still rare.In a fibrous filter, aerodynamic slip takes place when the fiber diameter is comparable to the mean free path of the gas molecules.In this case, collisions between the gas molecules and the fibers become so infrequent that the gas can no longer be treated as a continuum phase with respect to the fibers (Hosseini and Tafreshi, 2011).Moreover, the Cunningham slip factor should be considered in the model for particles with diameter near 1 µm.
Therefore, in this study the truncated log-normal distributed model (Zhou et al., 2009) will be investigated by using LB method when slip boundary condition is considered.Both the initial pressure drop and the filtration efficiency will be investigated, which is aimed to provide a feasible theoretical and numerical model for further study of fibrous air filter media with low-resistance and high-efficiency.

Computational Model
The computational domain and the log-normal distributed random fiber model (Zhou et al., 2009) were adopted, which represents the image of a cutting plane perpendicular to the filter media.Modeling selects only a small fraction of the total width, but includes the entire thickness, taking explicit account of the microstructure and the variability of fibers in both diameter and location.Fiber diameters were sampled by measuring the width of those fibers intersected by straight lines drawn at random across the image.The media was modeled with realistic descriptions of real structures of filters, where circles are randomly spaced by C++ programs.It provides an accurate model basis for simulating the microscopic flow conditions in media samples.
The schematic diagram of media F6 model is illustrated in Fig. 1.The media thickness L = BC, and the media width W = CF.The width was set to be 75 µm.The distance from the inlet to the media AB = 120 µm which was set to ensure that the velocity field at the windward side of the media (Line BG in Fig. 1) was not affected by the influence of inlet flow.Likewise, the distance from the outlet to the media CD = 120 µm which was set to prevent the inverse flow for the full development of the air flow in the media.The sensitivity analysis of media thickness and width has been performed and the optimal values are adopted (Zhou et al., 2009).In Table 1, the geometric information is presented for the media model.

Lattice Boltzmann Method
LB method is a discrete process in macro, but a continuous process in micro.Therefore, LB method is called mesoscopic simulation method.It provides the possibility and reality to link macro and micro, and it also sets an effective way for simulating the porous media flow as well.In this paper, the D2G9 scheme of LB method on a two-dimensional square lattice was adopted, which can simulate incompressible fluid flow accurately and efficiently (He et al., 2009).The evolution of fluid particles can be decomposed into migration and collision processes.The code was compiled by C++ programs.
The evolution of the distribution function f α (r, t) also obeys the Lattice Boltzmann equation: where δ t = 1 is the discrete time step and eq f  is the equilibrium distribution function.
In kinetic theory, the kinematic fluid viscosity υ can be defined as (Tian et al., 2009): The kinematic fluid viscosity υ was given as (He et al., 2009): where the lattice speed of sound For the two-dimensional and nine-speed LB models, c s = RT .
In Eq. ( 2), the dimensionless relaxation time τ can be derived from Eq. ( 4), which was widely used in simulating microscale air flows with LB method (Zheng et al., 2012): where Δ = 1/N = δ x /L, N is the mesh number of a characteristic length; δ x is the discrete lattice step (He et al., 2009).In Eq. ( 4), the average Kn number in Table 2 was taken, so that LB models can be applied to simulate microscale air flows (Tian et al., 2009).
The equilibrium distribution function ( , ) eq f t


x was redefined as the form of Guo's incompressible model (He et al., 2009): where constant ρ 0 was the average density of the fluid; s α was defined as follows: e u e u u u (6) In the numerical simulation, d 0 , d 1 and d 2 can be chosen as 5/12, 1/3 and 1/12, respectively (Li et al., 2015).
The weight coefficient w α was defined as: In the process of evolution, it is required to maintain the conservation of mass and momentum, namely: Macroscopic pressure of the fluid could be obtained from Eq. ( 10).
With the Chapman-Enskog expansion, the macroscopic equation of D2G9 model can be used to derive the full N-S equations.
For describing the movement of airborne particles, the CA probability model was applied.Both the drag force and Brownian diffusion on particles were considered.Eqs. ( 11) and ( 12) were used to obtain the variation of particle velocity and position, respectively.
where u p is the particle velocity; τ p is the relaxation time of  11) and ( 12), Eqs. ( 13) and ( 14) were obtained.
where the superscript n denotes the current time; n + 1 denotes the next step.

Boundary Condition Slip Boundary Condition
In order to estimate the possibility of slip phenomenon taking place on the given fibers' surface, the range of the fiber diameter was needed to be known.According to the measured fiber diameter data, the values corresponding to each Kn range are illustrated in Table 2.
It can be seen that the Kn of all fibers range from 0.001 to 0.1, where N-S equation with slip boundary conditions may be applied to flows in slip flow regime or marginally transitional regime.For filter media F6, all fibers can be considered in the slip flow regime.Table 2 shows that no fiber is in free molecular flow regime.
In this paper, the slip velocities on all the fiber surfaces were implemented by the boundary condition combining the bounce-back with the specular reflection (BSR scheme), which was proposed by Succi (2002).Here, taking the wall boundaries for example, the BSR scheme can be described as follows (Tang et al., 2004).
At bottom wall of fibers (arc I-L-J in Fig. 2): At top wall of fibers (arc I-K-J in Fig. 2): where NY is the number of lattices in y direction, and the proportion coefficient r b is a parameter to measure slippage Fig. 2. The diagram of a fiber inside fibrous media.
degree (0 ≤ r b ≤ 1).In other words, r b and 1 -r b represents the probability for a fluid particle to be bounced back and slipped forward, respectively.When r b = 1, it corresponds to pure bounce-back scheme, which means that a fluid particle changes its moving direction to the opposite one when it contacts the solid surface (Przekop and Gradoń, 2008).
Whereas r b = 0 corresponds to pure specular reflection scheme used for free slip boundary conditions.Tang adopted r b = 0.7, which was used to best capture the slip-velocity on the solid-gas wall, i.e., more fluid particles will be reflected in the backward direction than the forward direction.Tang also pointed out that the value of r b was probably Kn number dependent and r b = 0.7 worked well for the present Kn number range (0.055-0.16) as the predicted flow results were shown to be closer to the available data (Tang et al., 2004).For the Kn ranges of filter media F6, the value of r b was needed to be recalculated.
In the slip flow regime (0.001 < Kn ≤ 0.1), the commonly used second-order slip boundary condition takes the form (He et al., 2009): where u s is the slip velocity; A 1 and A 2 are the first-order and second-order slip coefficients; n is the wall normal coordinate pointing into the fluid; the subscript w represents the quantity at the wall.Hadjiconstantinou rescaled the values of A 1 and A 2 considering Knudsen layer effects and obtained A 1 = 1.11 and A 2 = 0.61, respectively (Hadjiconstantinou, 2003).
Parameter r b of BSR scheme is chosen as (He et al., 2009):

Bounce-Back Boundary Condition
In order to reflect the slippage effects, the no-slip boundary conditions were considered by an improved bounce-back scheme with second-order accuracy.For standard bounceback scheme, the distribution function of boundary nodes can be obtained by the relative fluid nodes after bounce-back: To satisfy the no-slip condition, the distribution functions f α can take the following expressions as Eq. ( 20).Take the bottom wall for example: The velocity boundary condition was applied to the inlet, whereas the fully developed format was used for the outlet boundary.Guo et al. (2002) proposed the non-equilibrium extrapolation boundary in 2002, which has a second order accuracy in time and space.In Fig. 3, C/O/A is the solid boundary, and the node E/B/D is the fluid node located in flow field.

Non-Equilibrium Extrapolation Boundary
The distribution function collision at point O can be approximated as follows.

 
When the collision process is considered, the distribution function collision at point O after collision can be expressed as: In cases A and B in Table 4, both the upper and the lower boundaries adopted non-equilibrium extrapolation format.

Periodic Boundary
The periodic boundary means that when the fluid particles leave from one side of the flow field, they will enter into the flow field from the other side at the next step.Therefore, this boundary treatment can strictly ensure the conservation of mass and momentum of the entire system.
The treatment of the periodic boundary can be given as follows.
where the expression f 1,5,8 (0, j) = f 1,5,8 (N x , j) represents that the equilibrium distribution functions f 1 , f 5 and f 8 of the virtual fluid node (0, j) were equal to that of the fluid node (N x , j).Cases C and D in Table 4 both corresponded to the situation when the upper and lower boundaries adopted periodic boundary format.

Theoretical Expressions for Resistance
For fibrous filter media, the resistance ∆p is related to the following variables: the flow rate Q, the cross-sectional area A, the media thickness L, the dynamic viscosity μ, the density ρ, the mean free path of air molecules λ, the solid volume fraction α, and the average diameter of fibers R f .

Expression of Resistance in Simulation
To investigate the performance of two filter media models, horizontal sampling line was set with two terminal sampling points placed at the midpoint of AH and DE, respectively.The simulation resistance value Δp can be expressed as follows: where Δp is the pressure drop across a filter media with a thickness L. p in and p out are the static pressures at inlet and outlet of media, respectively.
To compare with the experimental and analytical data, the dimensionless resistance value Δp 1 is required to be converted to the actual resistance Δp p .Eq. ( 30) can be used, where subscript p denotes the dimensionless parameters and subscript 1 denotes the actual parameters.
Test Rig for Experiment Experiments were performed on filter media test rig in the Aerosol Technology and Filter Testing Laboratory, Politecnico di Torino, Italy (Zhou et al., 2010).
Indoor air from the laboratory was delivered through the supply fan and exhaust fan.The flow rate of the test rig can be varied between 6 m 3 s -1 and 60 m 3 s -1 by the inverters.The incoming air passed through the absolute air filter (HEPA).The clean air was mixed with the aerosol in the mixing box (aerosol-air mixing chamber).Aerosol was generated with a Laskin nozzle which was put inside a container.Then aerosol was injected towards the mixing box through a plastic pipeline with diameter 65 mm.The mixing box was a cubic with the side length 495 mm.
The holder chamber for the filter media was also s cubic.Its side length was 300 mm.The section for measurement of the flow rate with the manometer was placed at the downstream of this holder chamber.
Two aerosol sampling pipelines were set upstream and downstream of the holder chamber for the media.The diameters of these two sampling probes were chosen according to the isokinetic sampling theory.
The pressure drop was measured with a differential manometer.Lax-X optical particle counter manufactured by Particle Measuring Systems Co.Ltd was used to measure the upstream and downstream particle concentrations, which were applied to obtain the efficiency value of the media.The schematic diagram of the test rig is shown in Fig. 4.

Validation of the Mesh Dependence
Before extracting the information about the flow field of filter media, the validation of the mesh dependence was studied to ensure the accuracy of the simulations.For the convenience of comparison, we take the mesh density on the single fiber with diameter D = 2 µm as an example.Table 3 lists four kinds of mesh densities on single fiber.
The comparison of the resistance values with various mesh densities are shown in Fig. 5.It is shown in Fig. 5 that the resistance of media F6 differs a lot when the number of mesh occupied by a single fiber with diameter 2 µm varies from 1 to 4. When the mesh density is bigger than 4, the resistance is very close to each other, i.e., the resistance can reach stable when the meshing of media F6 is 670 × 75.Therefore, in order to ensure the accuracy of the simulation, the final selected mesh densities for media F6 was 670 × 75.

Velocity Magnitude Contour
From the velocity magnitude contours shown in Fig. 6, obvious flow channels with higher velocity appear in the middle of the media models.In the randomly arranged fibers, the velocity will increase a lot when the space between fibers turns larger, i.e., the warm color areas.Likewise, the velocity is relatively small when fibers are densely occupied, i.e., the cool color areas.For the case where inter-space between fibers is large, air flow channels will be formed through these spaces and the average velocity through the channel is comparatively large.This was also discovered by Chen and Papathanasiou (2008).

Validation of Resistance Comparison of Resistance with Non-Equilibrium Extrapolation Schemes
Simulation was performed to obtain the resistance of filter media F6 with various inlet velocities.The resistance Δp from numerical simulation was validated against experimental and analytical data.Analytical results from Davies (1973), Kuwabara (1959) and Happel (1959) were included.Table 4 illustrates cases with different boundary conditions during simulation.
In Fig. 7, both case A and case B corresponded to situations where the upper and lower boundaries adopted non-equilibrium extrapolation format.Case A took no-slip boundary condition on fiber surface and case B took slip boundary condition.For case A, it is shown that the resistance values of two filter media almost increased linearly with the inlet velocity, which agrees with the Darcy's law.For case B, the variation of resistance values with the inlet velocity also showed the same growth trend.
The resistance values from expressions by Happel and Davies were both lower than the experimental values, whereas the values from Kuwabara's expressions were closest to it.The resistance values Δp for case B were much lower than that for case A. In other words, aerodynamic slip makes a significant decrease in the resistance for filter media.
Thus, it is essential to pay attention to the slippage effect during the study on the flow field of fibrous media.
The above simulation results were obtained when the upper and lower boundaries adopt non-equilibrium extrapolation format.It can be seen that the resistance values Δp for case B still have a large difference from the experimental results.Therefore, the influence of the boundary conditions on the F6 model was further investigated.

Comparison of Resistance with Periodic Schemes
Since the upper and lower boundaries can be regarded as infinitely far as to have negligible influence on the flow field, periodic boundary format was also investigated.This treatment is in line with the actual situation, which is shown in Fig. 8.It cannot only reduce the impact of the upper and lower wall on the flow field, but also ensure the conservation of mass and energy for entire system (He et al., 2009).
In Fig. 9, case C took no-slip boundary condition on fiber surface and case D took slip boundary condition.The resistance values Δp for case D were smaller than that for case C. Compared with the results in Fig. 7, the resistance values of simulation were much smaller, which show less influence by upper and lower boundaries.The resistance values for case C were higher than the experimental data.However, for case D, the resistance values of media F6 were closer to the experimental data with the increase of the inlet velocity.Under various inlet velocities, the simulation results are all between the two dash lines, i.e., within the range of the deviation degree.The resistance values obtained have a good agreement with the experimental data.
It is shown that only the treatment of the upper and lower boundaries can exert a large influence on the simulation results.For laminar flow with low Reynolds number, the initial conditions of the flow field have little effect on the final results, while the boundary conditions may have more impact on it, together with numerical stability and computational efficiency.
The dimensionless resistance reduction Π is defined as (Yu et al., 2015): The numerical relationship between Π and the inlet velocity reflects the degree of resistance decrease, as shown in Fig. 10.
Compared with the simulation results for case C and case D, the dimensionless resistance reduction Π of media F6 was obtained.Generally, with the increase of the inlet velocity U, Π of media F6 decreased first and then increased.Within the inlet velocity range, the maximal reduction Π of media F6 was 33.7% when U was the smallest.Moreover, Π of media F6 decreased to the minimum value 25.0% when U = 0.046 m s -1 .In short, within a certain range of flow rate, the application of slip boundary can bring a    decline rate of resistance for media F6 to be 25.0%.
From Fig. 10 it is shown that slippage effect turns significant when the inlet velocity is low enough.For the case of smallest inlet velocity, the boundary layer around fibers is comparatively thicker than that of large inlet velocity.Therefore, the implementation of slip boundary condition on fiber surface will have more significant effect on the flow field inside fibrous media, which results in the larger relative difference of resistance values between slip and no-slip boundary conditions.As the inlet velocity increase, the thickness of the boundary layer near the wall will be reduced.This means that when the no-slip wall boundary is implemented on the fiber surface, the influence of the boundary layer on the pressure drop of the model will be   decreased with the reduced thinner boundary layer.Therefore, the difference between no-slip and slip boundary conditions in this case becomes smaller.
However, as the inlet velocity further increases, the wake region forms at the downstream of the cylinder and the fluctuation of the wake flow becomes fierce.The possibility of vortex shedding becomes larger.As the turbulence intensity increases, the drag force on the fiber when no-slip boundary is applied will be increased.On the contrary, the phenomenon for the fluctuation of the wake flow will be reduced if the slip boundary is utilized.This will reduce the pressure drop of the model.In this case, the difference between no-slip and slip boundary conditions in this case becomes larger again.

Validation of Efficiency Experimental Value of Efficiency
Table 5 illustrates the fractional efficiency of media F6 with the filtration velocity 0.0617 m s -1 .

Simulated Value of Efficiency
After the stabilized flow field was obtained, 500 particles with diameter 0.1 µm, 1 µm and 5 µm were injected at the entrance of the filter media, respectively.In order to reduce the influence of the initial particle positions, independent random numbers were generated so that particles were distributed randomly in the region 1 ≤ x ≤ 120 µm and 1 ≤ y ≤ 75 µm.The density of the simulated particles was 912 kg m -3 , which corresponds with that of DEHS aerosol.The processes for particle movement and capture were monitored.Once particles contact the fiber surface, they were considered to be captured and then assumed to disappear.
Figs. 11-13 present the state diagram of particles released at the entrance, particles near the fibers and the whole condition inside the media, respectively.
According to Table 5, the experimental efficiency for particles with diameter between 0.75 µm and 1.00 µm reached 23.4% ± 2.8%, while that for particles with diameter between 1.50 µm and 2.00 µm reached 38.1% ± 4.1%.Fig. 14 illustrates the comparison of efficiency between simulation and experiment.The mean diameter in each size range interval shown in Table 5 is used for plotting Fig. 14.It is equal to the geometric average of the lower and upper border diameter.It should be noted that in the experiment data, the filtration efficiency for particles with diameter 0.1 µm was less than that for 0.2 µm.When the measurement error    is considered, the summation of the average efficiency and deviation is used to plot the data corresponding 0.1 µm in Fig. 14.The simulated filtration efficiency for particles with diameter 1 µm was 38.0% under the filtration velocity 0.0617 m s -1 .The difference between simulation and experiment for this particle diameter is nor large.However, it is found that the simulated particle capture efficiency for particles with diameter 5 µm is smaller than the experimental data.The possible reason is that in real fibrous filter media, binder is usually used for fixing the place of fibers.The existence of binder may increase the actual particle capture efficiency.This is why the experimental data is larger than the simulated value with the current model.Therefore, the binder is suggested to be added in the future model of the media.

CONCLUSIONS
By applying LB method for the truncated log-normal distribution model of fibrous media F6, the influence of boundary conditions on the resistance of the media was investigated.Compared with the no-slip boundary condition, the simulated resistance with the implementation of slip boundary condition on fiber surface agrees well with the experimental data.For low Re number flow inside the fibrous filter media, the resistance of the upper and lower boundaries with the periodic scheme is much closer to the experimental value.This reflects the real condition of the fibrous media.In short, the influence of boundary condition is larger than that of the initial condition on the simulated results.
Moreover, Kuwabara's expression is the best to describe the resistance of the fibrous filter media, compared with these expressions by Happel and Davies.It should be noted that the Re number in the simulation on the simplified 2D fiber model is the same as that in the experiment.This dimensionless parameter is one important factor for the similarity between simulation and experiment.
By validating the resistance and the filtration efficiency on the truncated log-normal model, it is found that the LB method can be used to predict the performance of the fibrous filter media.Meanwhile, the model to include the effect of binder in fibrous media is needed in future study.
is the relaxation time with the expression C c = 1 + λ/d × [2.34 + 1.05 × exp(-0.39× d/λ)]; ρ p is the particle density; d p is the particle diameter; F B is the random Brownian force; F D is the drag force by the fluid; k B is the Boltzmann constant; μ is the dynamic viscosity of the gas; T is the thermodynamic temperature; ξ is the Gaussian random variable with average value 0 and variance 1; Δt is the time step for the movement of particles.By integration of Eqs. ( 1

Fig. 8 .
Fig. 8.The cross section of filter media and the investigated region.

Fig. 13 .
Fig. 13.Schematic diagram of particles in the media.

Table 1 .
The geometric information of the media model.

Table 2 .
The Kn ranges of filter media F6.
0 , d 1 and d 2 were model parameters, which were given by:

Table 3 .
Different mesh densities on single fiber with diameter D = 2 µm.

Table 4 .
Cases with different boundary conditions.

Table 5 .
Filtration efficiency of media F6 with the rated filtration velocity 0.0617 m s -1 .