Sei sulla pagina 1di 9
Separation and Purification Technology 209 (2019) 461–469 Contents lists available at ScienceDirect Separation and

Contents lists available at ScienceDirect

Separation and Puri cation Technology

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

journal homepage: www.elsevier.com/locate/seppur Modeling of fi brous fi lter media for ultra fi ne

Modeling of brous lter media for ultra ne particle ltration

Seungkoo Kang a , Handol Lee a , Seong Chan Kim a , Da-Ren Chen b , , David Y.H. Pui a ,

a Particle Technology Laboratory, Mechanical Engineering, University of Minnesota, 111 Church St. S.E., Minneapolis, MN 55455, USA b Particle Laboratory, Department of Mechanical and Nuclear Engineering, Virginia Commonwealth University, 401 West Main Street, Richmond, VA 23284, USA

T
T

ARTICLE INFO

Keywords:

Collection eciency Fibrous lter media Polydisperse bers Nanoparticle ltration

ABSTRACT

Fibrous lter media are widely applied to remove aerosol particles. The modeling and prediction of ltration performance of brous media is of importance to design lter media targeted for specic applications. In this work, we successfully developed a 2-D numerical modeling for brous lter media. In our modeling, the ow eld and fates of particles were calculated in model lter media having the ber size distribution, average solidity, and thickness the same as those of real media. An excellent agreement was obtained between the numerical and experimental particle collection e ciencies of two commercially available brous lter media for particles in the sizes ranging from 3 to 500 nm. We also investigated the e ect of ber size polydispersity (both in the unimodal and bimodal ber size distributions) on the particle collection eciency of brous media. It was found that the particle capture in brous media is noticeably in uenced by the polydispersity of bers in a unimodal distribution, especially for particles in the sizes ranging from 10 to 100 nm, and further e ected by the peak size and volume fraction in each mode of a bimodal ber size distribution.

1. Introduction

The removal of ambient ne PM 2.5 (de ned as the particulate matter with the size less than 2.5 μ m) and ultra ne PM (de ned as the particulate matter with the size less than 100 nm) is of great interest for residents in the close proximity to highways and in highly polluted cities. It is because ne and ultra ne PM could have adverse e ects on the public health, especially for children, elders and people with pul- monary diseases [1 3] . As reported by Berkeley Earth [1] , 4000 people in China die on the daily base because of air pollution. High reading of PM 2.5 is one of major air pollution contributors. According to the work of Zhou et al. [2] , hourly peaks of PM 2.5 concentration had exceeded 800 μg/m 3 in Beijing since 2013, which is 32 times higher than the 25 μg/m 3 level recommended by World Health Organization (WHO). Respirators are required to remove ne and ultra ne PM from breathing air. Fibrous lter media are widely used in respirators for its low cost and e ective removal of PM. Researches have been performed to investigate the collection e ciency of brous lter media for parti- cles in a wide size range [410]. The classical single- ber-e ciency (SFE) theory has been widely applied to predict the particle collection e ciency of brous lter media [11,12] . The SFE theory estimates the overall particle collection e ciency of brous lter media by representing the media as an en- semble of well-patterned model cells in which a single ber is located at

the centers of individual cells. The overall particle collection e ciency, E , or particle penetration, P, of lter media can then be calculated as:

4 αE t

Σ

πd

f

(1

α

)

P =− =

(1)

where t is the media thickness, α is the medium solidity, d f is the ber diameter, and E is the total SFE. The total SFE, considering all the particle capture mechanisms, i.e., by the di usion ( E D ), interception ( E R ), impaction ( E I ) and gravitational setting ( E G ), can be expressed as:

E =−−1 (1 E )(1EE)(1)(1E )(1E )

where E DR is the enhanced capture from the interception of di using particles. Notice that the gravity e ect on particle ltration is negligible for submicrometer-sized particles. The empirical formulas of E R , E I and E D were proposed in early studies [1114]. For the small particle l- tration via porous media, the particle Brownian di usion is the pre- dominant capture mechanism. Both interception and impaction play key roles in capturing particles with the sizes comparable to the lter ber diameter. Typical formulas for the single ber e ciency due to the aforementioned particle capture mechanisms are given in Table 1 [12,15,16] . Among all the parameters in the SFE formulas, the ber diameter, d f , is one of the key factors greatly in uencing the overall particle collection e ciency of lter media. The SFE theory however assumes a

(2)

1

E

exp

Σ

R

I

D

DR

G

Corresponding authors. E-mail address: dchen3@vcu.edu (D.-R. Chen).

Received 10 May 2018; Received in revised form 24 July 2018; Accepted 25 July 2018

Available online 26 July 2018 1383-5866/ © 2018 Elsevier B.V. All rights reserved.

S. Kang et al.

Table 1 Empirical formula for single ber e ciency due to various particle capture mechanisms.

Separation and Purification Technology 209 (2019) 461–469

Capture mechanism

SFE equation

 

References

Diusion

E

D

= 2Pe

2/3

[12]

E

D

= 1.6

(

1 α

Ku

)

1/3

Pe

2/3

; Pe

=

d f U

D

; D

 

kTC c

 

[16]

=

3

πμd p

Interception

E

R

= (

1 α

)

R

2

[12]

Ku

(1

+

R

)

 
 

1

α

R

2

d

lnα

 

3

α

2

[16]

Impaction

E

R

= 0.6

E =

(

(

)

Stk J

Ku

)

1

+ R

; R

==−

; Ku

p

d

f

2

4

+ −

α

4

; Stk =

2

ρ p d p UC c

; J =−(29.6 28α

0.62

2

)R 27.5R

2.8

[15]

I

2

Ku

2

18

μd f

   

for R < 0.4

Interception-di usion

E

DR

=

1.24

R

2/3

 

[12]

( Ku pe

) 1/2

uniform ber diameter in media, which is not the case in real brous lter media. Equivalent diameters were proposed to determine the ber size to be used in the SFE theory [4,6,17 20] . For the mean diameter, either the arithmetic mean (i.e., average), geometric mean, or volume- surface mean of a given ber size distribution has been proposed as the equivalent ber size [4,17 19] :

d A (Arithmetic mean):

d

f

n

d

G

n f1 f2 fn
n
f1 f2
fn

(Geometric mean): dd d

(3)

(4)

2

f

nd

nd

f

d VS

(Volume surface diameter):

(5)

where, d f is the real ber diameter and n is the total number of bers. The e ective ber diameter (i.e., pressure-equivalent diameter) can be derived from the given pressure drop across lter media at a constant face velocity [21] :

=

3 3 64 α (1 + 56 ) α 2 Uμt ⎛ ⎜ ⎝ ⎞
3
3
64
α
(1
+
56 )
α
2
Uμt
Δ P

d Eff

(6)

where, U is the face velocity, µ is the viscosity of gas, and Δ p is the pressure drop across lter media. The e ective ber diameter de ned by Davies [21] has been applied in previous studies [6,17,20] . Other de nitions of equivalent ber diameters have also been proposed in the literatures. Four equivalent ber sizes, described above, are often adopted in the studies of brous lter media. The overall collection e ciency as a function of particle size for a brous lter medium are in fact di erent when applying the SFE theory with aforementioned equivalent ber diameters. As an example, Fig. 1

equivalent fi ber diameters. As an example, Fig. 1 Fig. 1. Particle collection e ffi ciency
equivalent fi ber diameters. As an example, Fig. 1 Fig. 1. Particle collection e ffi ciency

Fig. 1. Particle collection e ciency for HF-0012 at the face velocity of 5.3 cm/ sec.

462

Table 2 Speci cations of H&V lter media.

Filter parameters

HF-0012

HF-0493

Thickness (mm) Basis weight (gram/m 2 ) Solidity DOP % Penetration (0.3 μm at 5.3 cm/s) Pressure drop at 5.3 cm/s (Pa)

0.74

0.36

6 9

65

0.039

0.076

79.9

88

13

8

gives various particle collection e ciency curves calculated by Eq. (1)

with single ber e ciency formula given in Table 1 and di erent

equivalent ber sizes for a berglass lter medium HF-0012 (Hol-

lingsworth & Vose, East Walpole, MA) . The specication of lters used in this study are given in Table 2 . It is shown that the particle collection e ciency curve of a lter media strongly depends on equivalent ber diameters and one equivalent ber diameter to calculate the particle collection e ciency of brous lter media in a wide particle size range might not be even feasible [18] . In addition, the particle collection e ciency curve of a lter medium might be varied when di erent ex- pressions for the same capture mechanism are applied in the SFE theory. 2-D and 3-D numerical modeling has been utilized to investigate the ltration performance of brous lter media. Unfortunately, a majority of numerical studies modeled lter media with uniform bers (either in patterned or random packing in media) and compared their results with those calculated by the SFE theory [22 28] . The comparison of the particle collection e ciency with the experimental data was merely made in three published studies [2931]. In the rst two studies [29,30] , the comparison was limited to the cases of two/three particle sizes, with only one size in the submicrometer range. Sambaer et al. [31] developed a ltration model based on image-based 3-D structures of nano ber media. The authors produced polyurethane nano ber media by the electrospinning process, in which relatively monodisperse nano bers in the diameter of 120 nm were manufactured. The ow eld was assumed in their modeling because of large void space and negligible nano ber e ect on the ow eld in media. Also note that the particle ltration by nano ber media is dominated by the impaction and interception, not by the particle di usion. Dierent from previous studies, we selected two commercially available lter media (i.e., HF-0012 and HF-0493 from Hollingsworth & Vose) to validate our modeling. In this study, a 2-D numerical modeling with the consideration of polydisperse bers was developed to calculate the ow eld and particle collection in selected media. The model va- lidation was accomplished by comparing our numerical result with the measured particle collection e ciency in a wide submicrometer size range (i.e., from 3 to 500 nm). Upon the model validation, the e ects of ber polydispersity in unimodal and bimodal ber distributions on the particle collection e ciency of lter media were investigated.

S. Kang et al.

2. Numerical modeling

2.1. Model for brous lter media

Selected brous lter media, i.e., HF-0012 and HF-0493, were modeled in the 2-D rectangular medium space with the width the same as the thickness of real lter media. The height of the medium space was determined by considering the total number of bers to be placed and the void space of studied lter media. In general, more than 500 bers were placed in the media space to assure that our numerical re- sult are independent of the medium space size [23] . Fibers were added to the medium space according to the measured log-normal ber size distributions and average solidity of studied media. Diameters of bers were randomly generated in the measured ber size distributions. The generated bers were randomly placed in the medium space until the average media solidity was achieved. Note that the di erence on the ltration e ciency of a given lter media calculated using the domains based on the experimental and tted log-normal ber size distributions was negligible. The choice of using the tted lognormal ber distribu- tion in the code is for easy programming and control of distribution parameters. The algorithm to arrange bers in the pre-dened media space is similar to that proposed by Hosseini and Tafreshi [23] . In the algo- rithm, bers were consecutively generated and randomly placed in the medium space while avoiding the ber overlapping. To have high- quality meshes for the ow eld calculation, a minimal distance of 0.1 µm between adjacent bers was kept when placing bers in the medium space. When a ber crossing either the top or bottom boundary of medium space, the outside-space portion of the ber was then placed on the opposite boundary of medium space because of the periodic boundary conditions assumed at the top and bottom boundaries of medium space. Note that the 0.1 µm gap is several tenths of average ber diameters of studied media. The total number of grids was in the range of 1,000,000 1,500,000 = and the range of grid size was 3 × 10 8 1 × 10 6 , depending on the ber distribution in lter do- mains and the lter types. Generally, the size of grids was ne in the region near bers and coarse in the region away from bers.

2.2. Calculation of ow eld

A typical computational domain used in this modeling and boundary conditions is shown in Fig. 2 . The computational domain typically extends 150 μm upstream and 30 μm downstream the medium space. The ow was assumed to move from the left entrance to the right exit. A uniform velocity pro le (either 10 or 15 cm/s) and pressure were assumed at the ow entrance and exit of the computational do- main, respectively. The periodic boundary conditions were applied at the top and bottom boundaries of the domain. Non-slip condition was applied to all the ber surfaces.

Separation and Purification Technology 209 (2019) 461–469

Since the ow Reynolds number based on the ber diameter is far less than 1.0 (Re 1), the Stokes ow was assumed in the computa- tional domain and solved in ANSYS Fluent with the 2-D governing equations of continuity and momentum:

(7)

·u=0

∇−∇pμ u=0

(8)

Note that the special attention had been paid to the grid distribution in the neighborhood of ber surface for the accurate calculation of ow eld and particle capture near a ber. Hosseini and Tafreshi [23] found that the pressure drop over a ber is not aected when more than 50 grid points are used in the ber perimeter. In our modeling, the number of grid points on a ber perimeter was linearly increased with the ber diameter because bers in a wide size range were distributed in the medium space. The fraction of bers having grids less than 50 on a ber perimeter were approximately 5 8% in our study, and the calculation result was not aected by the small fraction of bers.

2

2.3. Particle tracking

An in-house JAVA code was developed for the particle trajectory calculation. The Langevin equation with the consideration of di usive random motion of particles was employed to calculate the particle Brownian motion in the computational domain [32] :

f

;

m

mv =−mβ vU + X

(9)

where m, β, and v are the mass, frictional constant, and velocity of a particle, respectively. U is the ow velocity at the particle position. X is a random force. The property of random force is depicted by a Gaussian distribution function with the mean and mean square as:

̇

(

)

;

β =

f: frictional factor

'

'

<X (t) >= 0, < X (tX (t ) 6mβkTδ (tt )

(10)

where k represents the Boltzmann constant and T is the absolute tem- perature. The method of Ermak and Buckholz [32] was applied to de- scribe the Brownian motion of particles. In this method, following the work of Chandrasekhar [33] and Tien and Ramarao [34] , a bivariate probability distribution was used together with its moments to de ne uniquely position and velocity at every time step t. The Cunningham slip correction factor Cc needs to be accounted for the submicron par- ticle as the particle size is close to the mean free path, λ , of gas mo-

)) , where Kn is the

lecules [35] ,

Knudsen number (Kn = 2 λ /particle diameter). To calculate the particle collection e ciency in model lter media as a function of particle size, particles with a specic size (i.e., 3, 5, 10, 20, 30, 50, 100, 200, 300, 500 nm) were introduced at the entrance of the computational domain. A total of 1500 particles was introduced for each running case. The locations to introduce particles at the domain entrance were assigned by uniformly dividing the height of the ow

C

c

=+

1

(

Kn 1.257

+

0.4exp

Kn

(

1.1

of the fl ow C c =+ 1 ( Kn 1.257 + 0.4exp − Kn (

Fig. 2. A typical example of the computational domain for model HF-0012 media.

463

S. Kang et al.

Separation and Purification Technology 209 (2019) 461–469

Separation and Purification Technology 209 (2019) 461–469 Fig. 3. Experimental setup for the measurement of particle

Fig. 3. Experimental setup for the measurement of particle collection e ciency of studied lter media.

entrance by the total number of particles to be introduced. The move- ment of individual particles was tracked to nd out their fate, i.e., whether they are captured by bers or they penetrate through the lter media. The ltration e ciency of the media was obtained from the counts of their fates. The above calculation was repeated at least ten times for each case. The average of ten calculated collection e ciencies was used in the comparison with the experimental eciency.

3. Experimental setup and method

3.1. Experimental setup for the particle collection e ciency measurement

Fig. 3 shows the schematic diagram of an experimental setup for the measurement of particle collection e ciency of studied lter media. The setup consists of a Collison atomizer (TSI 3079, Shoreview, MN) for the generation of submicrometer-sized particles, a di erential mobility classi er (DMC, TSI 3081, Shoreview, MN) for the particle size classi- cation, and an ultra ne condensation particle counter (UCPC, TSI 3776, Shoreview, MN) to measure the upstream and downstream con- centrations of test lter media. KCL solutions of two di erent con- centrations (i.e., 0.01% and 1% by volume) were used in the atomizer to generate particles in the sizes ranging from 20 nm to 500 nm. 0.01% and 1% KCL solutions were used for test particles in 20 50 nm and 100 500 nm, respectively. The solvent in atomizer-produced droplets were dried in the di usion drier with silica gels as the desiccant. Electrical charges on dried particles was conditioned to a Boltzmann charge distribution in the Po-210 charger. A TSI DMC (TSI 3081) classi ed KCl particles in the mobility sizes of 20, 30, 50, 100, 200, 300, and 500 nm. A 2nd Po-210 charger was utilized to minimize electrical charges on DMC-classi ed particles. The penetration of particles at each tested size was calculated as the ratio of the concentrations in the downstream and upstream of test lter media:

E ( d )

x

= C

down

( d )

x

C

up

( d )

x

(11)

where, C up (d x ) is the particle concentration at the upstream of lter media, C down (d x ) is the downstream particle concentration, and E (d x ) is the penetration e ciency for a test particle size. The test face velocity was selected either at 10 or 15 cm/s. For each particle size, the pene- tration e ciency of each lter media sample was measured at least 3 times and averaged. Three di erent medium samples were randomly selected and tested for each selected lter media.

3.2. Characterization of studied lter media

The specication of two brous lter media are given in Table 2 . The media HF-0012 is thicker in the thickness and lower in the solidity than the media HF-0493. Although the DOP penetration of HF-0012 is lower than that of HF-0493 at the same face velocity, its ow resistance is higher than that of HF-0493. Fiber diameters of both HF-0012 and HF-0493 were characterized by the analysis of scanning electron microscope (SEM) images. Fig. 4 shows the SEM images of HF-0012 and HF-0493 at the magni cation of ×2000 and ×500, respectively. A total of 10 or more images were taken for each lter media. Individual ber diameters were measured from each image and the total number of measured bers exceeded more than 300. As shown in Fig. 5 , the ber size distribution measured from SEM images for a studied media could be tted by a log-normal distribution [4,18] . The mean equivalent ber diameters were calcu- lated using the measured ber size distributions for each lter media. Table 3 gives three mean equivalent ber diameters obtained from SEM images and one pressure-equivalent ber diameter calculated by Eq. (6) . Equivalent ber diameters of HF-0012 are in general less than those of HF-0493.

464

S. Kang et al.

Separation and Purification Technology 209 (2019) 461–469

Separation and Purification Technology 209 (2019) 461–469 Fig. 4. Typical SEM images of studied H&V fi

Fig. 4. Typical SEM images of studied H&V brous lter media: (a) HF-0012 and (b) HF-0493.

fi brous fi lter media: (a) HF-0012 and (b) HF-0493. Fig. 5. Cumulative fi ber size

Fig. 5. Cumulative ber size distributions for studied lter media (a) HF-0012 and (b) HF-0493.

Table 3 Equivalent ber diameters of studied lter media (calculated based on the measured ber size distribution and pressure drop).

Equivalent ber diameter (μ m)

HF-0012

HF-0493

Arithmetic mean Geometric mean Volume-surface average Eective ber diameter (Davies)

3.3

5.7

2.5

4.5

5.1

5.6

5.3

8.3

4. Result and discussion

4.1. Typical ow eld and particle trajectory in modeled lter media

Fig. 6 describes the streamlines in two computational domains (a) and (b) given in Table 4, respectively. One case (a) is with polydisperse bers and the other (b) is with monodisperse size. As shown in Fig. 6 ,

Table 4 Parameters of hypothetical lter media with unimodal ber size distributions.

Filter

Thickness

Solidity ( ) Arithmetic

Coe cient of

Number of bers ()

domain

(mm)

mean ber diameter (μm)

variation ()

no.

1

1

0.1

15

0.47

320

2

1

0.1

15

0.27

396

3

1

0.1

15

0.07

456

the ow elds in brous lter media are signi cantly aected by the ber polydispersity. Fig. 7 shows the trajectories of ten particles of 200 nm in size in the computational domain with modeled HF-0012 media (at the face ve- locity of 10 cm/s). Three particles were captured by bers and seven particles penetrate through the media. Because of periodic boundary

penetrate through the media. Because of periodic boundary Fig. 6. Examples of streamlines in the computational

Fig. 6. Examples of streamlines in the computational domains at the face velocity of 10 cm/s for two lter domains, (a) Filter domain 1 and (b) Filter domain 3, given in Table 4.

465

S. Kang et al.

S. Kang et al. Fig. 7. An example of 200 nm particle trajectory in the computational

Fig. 7. An example of 200 nm particle trajectory in the computational domain of HF-0012 media at the face velocity of 10 cm/s. 10 particle trajectories are shown.

condition, the particle crossing the top boundary was re-entered the computational domain from the bottom boundary.

4.2. Eect of modeled lter media

Because of the random ber size selection and placement in the medium space in each creation of model lter media, the actual ber size distribution and total number of bers in the modeled lter media are varied although the solidity and measured size distribution of real lter media are followed. For example, the number of bers in three HF-0012 lter medium spaces, shown in Fig. 8 , are 782, 874, and 929 with the media solidity of 3.929%, 3.899%, and 3.900%, respectively. The numerical particle collection eciencies of three model lter media HF-0012 as a function of particle size and at the face velocity of 10 cm/sec are shown in Fig. 9 . It is found that the variation of particle

Separation and Purification Technology 209 (2019) 461–469

Fig. 9. The comparison of particle collection effi ciency of three selected model HF-0012 fi
Fig. 9. The comparison of particle collection effi ciency of three selected model
HF-0012 fi lter media.

collection e ciency calculated via three model HF-0012 lter media are negligible. In our numerical study, the di erence of maximal ber numbers in all the cases was 7% resulting in the less-than-4% variation of cal- culated particle collection e ciency. To minimize the variation, 15 model media for each studied lter media were generated. The model media with the minimal, mean, and maximal numbers of bers were selected for our calculation. The nal particle collection e ciency for a given lter media and particle size was the average of the e ciencies

and particle size was the average of the e ffi ciencies Fig. 8. HF-0012 domains having

Fig. 8. HF-0012 domains having the di erent number of bers and normalized cumulative ber distributions corresponding to each domain.

466

S. Kang et al.

Separation and Purification Technology 209 (2019) 461–469

Separation and Purification Technology 209 (2019) 461–469 Fig. 10. Comparison of particle collection e ffi ciency,

Fig. 10. Comparison of particle collection e ciency, obtained from the experiment, numerical modeling, and SFE theory for (a) HF-0012 at 10 cm/s; (b) HF-0493 at 10 cm/s; (c) HF-0012 at 15 cm/s; and (d) HF-0493 at 15 cm/s. The eciency data calculated by the SFE equations given in the works of Hinds [12] .

by the SFE equations given in the works of Hinds [12] . Fig. 11. Comparison of

Fig. 11. Comparison of particle collection e ciency of three hypothetical brous lter media having various degree of polydispersity at the face velocity of (a) 10 cm/s and (b) 15 cm/s.

Table 5 Parameters of hypothetical lter media with bimodal ber size distributions.

Filter domain no.

Thickness (mm)

Solidity

Arithmetic mean ber diameter (μm)

Volume fraction of bers (%)

Coe cient of variation, CV

Number of bers

1

1

0.15

5, 20

0.005, 0.145

0.05, 0.05

253, 461

2

1

0.15

5, 20

0.005, 0.145

0.3, 0.3

235, 428

3

1

0.15

5, 20

0.005, 0.145

0.5, 0.5

200, 370

4

1

0.15

5, 15

0.005, 0.145

0.3, 0.3

240, 766

5

1

0.15

5, 30

0.005, 0.145

0.3, 0.3

235, 195

6

1

0.15

5, 30

0.015, 0.135

0.3, 0.3

710, 174

obtained from three selected model media.

4.3. Comparison of numerical and experimental particle collection

e ciency

Fig. 10 shows the comparison of collection e ciency, obtained from

467

the experiment, numerical modeling, and Eq. (1), as a function of particle size at both face velocities of 10 and 15 cm/s for studied HF- 0012 and HF-0493 lter media. The empirical formula given in Hinds [12] were used in the SFE equations. In general, the particle collection e ciency curves obtained from our numerical modeling in all the cases agree well with the experimental data for both studied lter media. The

S. Kang et al.

S. Kang et al. Fig. 12. Comparison of particle collection e ffi ciency of four hypothetical
S. Kang et al. Fig. 12. Comparison of particle collection e ffi ciency of four hypothetical

Fig. 12. Comparison of particle collection e ciency of four hypothetical lter media with bimodal ber size distributions at the face velocity of 10 cm/s.

maximal di erence between the numerical and experimental data is less than 7%. For both studied lter media, the particle collection e ciency cal- culated from the SFE theory with the geometric mean diameter sig- ni cantly deviates from the measured data. The SFE theory with the arithmetic mean diameter overestimates the collection e ciency of HF- 0012. On the other hand, it underestimates the collection e ciency of HF-0493 for particles with the sizes between 50 and 500 nm. The SFE theory with the volume-surface average diameter deviates from the experimental data for HF-0012, but it is in good agreement with the experimental data in the case of HF-0493. The SFE theory with the e ective ber diameter underestimates the collection e ciency for both HF-0012 and HF-0493. The similar trend could also be observed when using the SFE theory with the empirical equations given in the work of Lee and Liu [16] .

4.4. Eect of polydispersity

The e ect of ber polydispersity on the particle collection e ciency for model lter media having the same lter properties (i.e., lter thickness, solidity and arithmetic mean ber size) but di erent ber polydispersity was investigated in this part of study. Table 4 gives the parameters used for the creation of three model lter media (in Fig. 11 a). A model media with the higher coe cient of variation (CV) indicates that bers in the medium space has the higher degree of polydispersity. Because the size of model lter media is xed, the number of bers decreased as the degree of ber polydispersity in- creased. Fig. 11 shows the collection e ciency as a function of particle size at the face velocities of 10 and 15 cm/sec. The ber polydispersity e ect on the particle collection e ciency is negligible for particles with the sizes smaller than 10 nm and larger than 100 nm. The high di u- sivity of particles with less than 10 nm in sizes results in the capture of particles regardless of media microstructure di erence. For particles with the sizes larger than 100 nm, their sizes are close to the most pe- netrating particle size (MPPS) for studied media, in which particles are not e ectively captured by di usion, interception or impaction [12,17] . The ber numbers and polydispersity may not have signicant e ect on the particle penetration of lter media in the size range near by the MPPS. The variation of particle collection e ciency among three model lter media is clearly observed for particles in the sizes between 10 and 100 nm. More speci cally, the dierence of 5 14% in the particle col- lection e ciency were observed when comparing the cases with the model media 1 and 3 for particles in the sizes 10 100 nm. Thus, the e ect of ber polydispersity should be taken into the consideration in the modeling of ltration performance of brous lter media.

468

Separation and Purification Technology 209 (2019) 461–469

4.5. E ect of bimodal ber distributions

The particle collection e ciency of brous lter media with bi-

modal ber size distributions (i.e., ne and coarse ber modes) was

investigated in this part of study. Table 5 lists the parameters of six

hypothetical lter media. All lter media have the same thickness and

solidity. Fig. 12 shows the comparison of particle collection e ciency

of six media as a function of particle size at the face velocity of 10 cm/

sec. The media 1 3 have the di erent CVs of bers in both bimodal

modes. The media 1 has more bers to capture particles although all the

three media have the same macroscopic lter properties. As observed in

Fig. 12 , the collection e ciency of the media 1 is higher than those of

the media 2 and 3, i.e., the particle collection e ciency of lter media

is a ected by the degrees of ber size polydispersity. The media 2, 4, and 5 have di erent coarse ber modes. The media 4 has more bers in the coarse ber mode because its ber size peak is less than that of the media 2 and 5 (for a xed solidity). As a result, the particle collection e ciency of media 4 is higher than those of the media 2 and 5. Filter media 6 has the largest ber diameter in the coarse ber mode among six media and the volume fractions of ne ber mode are higher than other media. Media 6 thus has the number of bers in the ne mode higher than those of media 1 and 2. Because the increase of ber numbers in the ne mode, the particle collection e ciency of media 6 is comparable to that of media 1.

5. Conclusions

A new 2-D numerical modeling with the consideration of ber polydispersity has been proposed to predict the ltration performance of brous lter media. Dierent from early numerical studies, the va- lidation of our modeling was through the comparison of measured particle collection e ciency of real lter media, i.e., HF-0012 and HF- 0493 in a wide particle size range, i.e., 3 500 nm. In the modeling, real brous lter media was modeled in the representative medium space lled with polydisperse bers, resulting in the same lter thickness and solidity, and ber size distribution as those measured. The ow eld and particle trajectory in model lter media were calculated. The nu- merical particle collection e ciency of model lter media was derived from tracing the fates of more than 1,500 particles of a given size. The collection e ciency of studied lter media were calculated for particles in the sizes ranging from 3 to 500 nm and at two face velocities of 10 and 15 cm/sec. Experiments were also carried out to characterize the ber size distribution and the particle collection eciency of selected lter media at two face velocities. The excellent agreement between the numerical and experimental collection eciencies for particles in the sizes ranging from 3 to 500 nm evidences the success of proposed modeling in the prediction of particle collection eciency of real - brous lter media. We further investigated the variation of particle collection e ciency of brous lter media due to the variation of ber size polydispersity (both in unimodal and bimodal ber distributions). In the case with unimodal ber size distribution, it is observed that the variation of particle collection e ciency is noticeable for particles in the sizes from 10 nm to 100 nm under the xed peak ber size, solidity and lter thickness. It is also found that the change of particle collection e - ciencies of lter media becomes more signicant as the degree of ber size polydispersity increases. The ber polydispersity also inuences the collection e ciency of brous lter media with bimodal ber size distributions. The particle collection e ciency of bimodal lter media was a ected by the peak ber size, volume fraction and ber poly- dispersity in both ne and coarse modes.

Acknowledgements

The authors thank the support of members of the Center for Filtration Research: 3M Corporation, A.O. Smith Company, Applied

S. Kang et al.

Materials, Inc., BASF Corporation, Boeing Company, Corning Co., China Yancheng Environmental Protection Science and Technology City, Cummins Filtration Inc., Donaldson Company, Inc., Entegris, Inc., Ford Motor Company, Guangxi Wat Yuan Filtration System Co., Ltd, MSP Corporation; Samsung Electronics Co., Ltd., Xinxiang Shengda Filtration Technology Co.,Ltd., TSI Inc., W. L. Gore & Associates, Inc., Shigematsu Works Co., Ltd., and the a liate member National Institute for Occupational Safety and Health (NIOSH).

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.seppur.2018.07.068 .

References

[1] Berkeley Earth Publishes, KILLER AIR: Study on Air Pollution in China, 2015,

[2]

469

Separation and Purification Technology 209 (2019) 461–469

[23]

[29]

P. Tronville, R. Rivers, B. Zhou, Particle capture by air lter media having truncated

log-normal ber diameter distributions and random spacing of bers, in:

Proceedings of the 10th World Filtration Congress, Leipzig, Germany, 2008, pp.

132 136.