Analyzing the Evolution of Membrane Fouling via a ... - ACS Publications

Subscriber access provided by the Henry Madden Library | California State University, Fresno

Article

Analyzing the Evolution of Membrane Fouling via a Novel Method Based on 3D Optical Coherence Tomography Imaging Weiyi Li, Xin Liu, Yi-Ning Wang, Tzyy Haur Chong, Chuyang Y. Tang, and Anthony Gordon Fane Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.6b00418 • Publication Date (Web): 08 Jun 2016 Downloaded from http://pubs.acs.org on June 12, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 33

Environmental Science & Technology

1

2

3

Analyzing the Evolution of Membrane Fouling via a

4

Novel Method Based on 3D Optical Coherence

5

Tomography Imaging

6 7

Weiyi Lia, *, Xin Liua, b, Yi-Ning Wanga,

8

Tzyy Haur Chonga, b, Chuyang Y. Tangc, Anthony G. Fanea, b

9 10

a. Singapore Membrane Technology Centre, Nanyang Technological University, Singapore

11

b. School of Civil and Environmental Engineering, Nanyang Technological University, Singapore

12

c. Department of Civil Engineering, the University of Hong Kong, Hong Kong SAR, China

13 14 15

*Corresponding author at:

16

Singapore Membrane Technology Centre

17

Nanyang Technological University

18

1 Cleantech Loop, Clean Tech One

19

Singapore, 637141

20

Tel: +65 6592 7726

21

Email: [email protected] (W. Li)

22 1 ACS Paragon Plus Environment


Page 2 of 33

23

Abstract

24

The development of novel tools for studying the fouling behavior during membrane

25

processes is critical.

26

quantitatively interpret the formation of a cake layer during a membrane process; the

27

quantitative analysis was based on a novel image processing method that was able to

28

precisely resolve the 3D structure of the cake layer on a micron scale.

29

experiments were carried out with foulants having different physicochemical

30

characteristics (silica nanoparticles and bentonite particles). The cake layers formed at a

31

series of times were digitalized using the OCT-based characterization. The specific

32

deposit (cake volume/membrane surface area) and surface coverage were evaluated as a

33

function of time, which for the first time provided direct experimental evidence for the

34

transition of various fouling mechanisms. Axial stripes were observed in the grayscale

35

plots showing the deposit distribution in the scanned area; this interesting observation

36

was in agreement with the instability analysis that correlated the polarized particle groups

37

with the small disturbances in the boundary layer. This work confirms that the OCT-

38

based characterization is able to provide deep insights into membrane fouling processes

39

and offers a powerful tool for exploring membrane processes with enhanced

40

performance.

This work explored optical coherence tomography (OCT) to

41

42

43

2 ACS Paragon Plus Environment

Fouling

Page 3 of 33

44


TOC Art

45

46

47



48 49

1. INTRODUCTION Membrane technology is playing an important role in dealing with environmental

50

issues.1-5

51

membrane fouling, which usually results in the growth of a foulant layer on the

52

membrane surface;2, 6-8 extra energy is therefore consumed to overcome the hydraulic

53

resistance increased by the fouling. Advancing techniques for mitigating membrane

54

fouling entails the knowledge that precisely describes the evolution of membrane fouling,

55

that is, the mechanisms accounting for the variation in fouling behavior during a

56

membrane process and the effect of parameters that control them, such as crossflow

57

velocity, flux and pressure.

The efficiency of a membrane process could be significantly limited by

58

The conventional approach for fouling characterization is to measure the variation in

59

permeation flux during a constant pressure membrane process9-11 or to measure the

60

transmembrane pressure (TMP) as a function of time while maintaining the permeation

61

flux at a certain value;12, 13 the degree of fouling is then described using a flux-decline or

62

TMP-increment curve. Although it is a relatively simple way to measuring the variation

63

in permeation flux or TMP, the information obtained in this way for interpreting the

64

fouling is limited. It is even possible to misinterpret the data since the process could be

65

significantly affected by some coupled effects, such as concentration polarization (CP).12,

66

14, 15

67

other advanced microscopy techniques (e.g., atomic force microscopy17, 18) are usually

68

employed to autopsy the cake layer formed at a certain time. Nevertheless, it is difficult

69

for autopsy characterization to capture the dynamic effects resulting from the complex

70

interplay between the foulant particles and fluid flow.

When external fouling is dominant, scanning electron microscopy (SEM)16, 17 or


Page 4 of 33

Page 5 of 33


71

A variety of novel techniques have been explored for in situ characterization of fouling

72

during membrane processes and most of these techniques are based on either non-optical

73

or optical signal detection.19 Typical examples of fouling monitoring techniques via non-

74

optical signal detection are ultrasonic time domain reflectometry (UTDR)16,

75

electrical impedance spectroscopy (EIS),21-24 which detect the growth of a foulant layer

76

using sound waves and alternating electrical currents, respectively. Optical microscopes

77

have been extensively employed to observe fouling processes and various modifications

78

have been made to enhance the sensitivity and resolution. For example, Li et al.25

79

proposed the direct observation through the membrane (DOTM) technique that focuses

80

the objective lens on the fluid-membrane interface during the filtration process, though its

81

application is limited to membranes that are transparent when wet. As a conventional

82

optical technique, microscopy is able to instantaneously generate images that are readily

83

used to analyze the surface coverage as a function of time.26, 27 However, microscopy-

84

based fouling characterization has difficulty in generating the depth profiles of the cake

85

layer, except for some special cases (e.g., lateral view is achievable when imaging the

86

foulant deposition on the outer surface of hollow fibers28).

20

and

87

Of particular interest in analyzing the evolution of a foulant layer is the variation in

88

local thickness, which can be measured in situ using advanced optical techniques. Gao et

89

al. demonstrated the potential of visualizing the velocity field29 and the growth of a

90

fouling layer30 via optical coherence tomography (OCT), which has the ability to obtain

91

depth profiles using low coherence interference (LCI).31, 32 Compared with confocal laser

92

scanning microscopy (CLSM),17,

93

sectioning by precisely controlling the depth of focus, OCT is able to scan the sample at a

33

which also has the ability to perform optical



Page 6 of 33

94

relatively high rate while achieving a microscale resolution.34

95

advantages, OCT is getting more attention in the field of membrane technology and

96

several studies have recently investigated the growth of a biofilm in a membrane cell via

97

OCT.35-37

Owing to these

98

Although quantitative analysis based on OCT imaging has been developed to

99

investigate membrane fouling in recent studies,30, 37 the precision of resolving the cake

100

morphology was limited by the adopted image processing algorithms and therefore the

101

analysis was restricted to a rough estimation of the deposit amount. The objective of this

102

work was to quantitatively investigate the evolution of membrane fouling by particulate

103

foulants using a novel OCT-based characterization method that was able to precisely

104

detect the distribution of the foulants in close proximity to the membrane surface. Both

105

silica nanoparticles and bentonite particles were employed as the model foulants for the

106

ultrafiltration (UF) tests; the cake-layer formation was recorded by continuously

107

performing 3D OCT scans to evaluate the specific deposit (volume/membrane surface

108

area) and surface coverage. The proposed data-analysis method allowed studying the

109

distribution of the foulants on a micron scale and capturing the dynamic effects that

110

underscored the important role of boundary-layer instability during the cake-layer

111

buildup.

112

113

2. EXPERIMENTAL PROCEDURES

114

2.1 3D OCT imaging


Page 7 of 33


115

Compared with conventional optical imaging, OCT is able to reconstruct depth profiles

116

of a semi-transparent sample in terms of the reflectivity variation in the sample

117

structure.34 A Fourier-domain OCT system (GANYMEDE-SP5, Thorlabs, USA) was

118

employed in the current study to achieve relatively high scan rates and its basic principle

119

is schematically displayed in Figure 1. The probe beam is emitted from a broadband light

120

source and split into the sample and reference arms. The light beam backscattered from

121

the sample is then recombined with the one reflected from a fixed reference mirror; the

122

resulting light beam is then resolved into its frequency components using a diffraction

123

grating.

124

generate the intensity spectrum that is Fourier-transformed to yield a complex signal.

The spectrometer measures the intensity of each frequency component to

125

In terms of scattering theory,38 the magnitude of the complex signal is actually the

126

interferogram whose envelope (i.e., the outline of the high frequency fluctuations)

127

characterizes the reflectivity variation in the depth direction.

128

intensity as a function of optical distance was used to reconstruct the depth profile at a

129

certain point in the sample (A-scan). The depth resolution (i.e., the coherence length)

130

yielded by the adopted light source (central wavelength 905 nm with a bandwidth of

131

~200 nm) was ~2 µm in water, whereas the lateral resolution determined by the scan lens

132

(LSM03-BB) had a relatively large value of ~4 µm. As depicted in the right panel of

133

Figure 1, 2D and 3D imaging were achieved by performing a series of A-scans along a

134

single line and a set of parallel lines, respectively. Moreover, the phase shift of the

135

complex signal was used to distinguish the moving particles from the stagnant cake layer,

136

i.e., the Doppler OCT imaging.34

137

2.2 Membrane fouling experiments 7 ACS Paragon Plus Environment

Therefore, the OCT


138

Membrane fouling experiments were implemented using the constant pressure mode.

139

In particular, the TMP was maintained at a constant value of ~1.3 bar, though filtration

140

tests with higher pressures can be challenged by enhancing the pressure tolerance of the

141

observation cell.

142

Corporation, USA) with an effective area of ~60 cm2 was positioned in a modified

143

membrane cell (182 mm×90 mm×20 mm) whose upper plate accommodated an optical

144

window (20 mm×20 mm×2 mm quartz plate) for the OCT scans. A schematic of this

145

modified membrane cell is shown in Figure 1 elucidating how the OCT-based

146

characterization was incorporated with a membrane fouling process.

The ultrafiltration membrane (OmegaTM UF PES, 10 kDa, Pall

147

The foulants employed in the current study were silica nanoparticles (Sigma-Aldrich,

148

product No. 637246, USA) and bentonite microparticles (Sigma-Aldrich, product No.

149

285234, USA); their size range and particle morphology were measured using Field

150

Emission SEM (JSM-7600F, JEOL Ltd., Japan).

151

particles in the feed solution (deionized water) was 0.1 g/L for both foulants. 5 L feed

152

solution was freshly prepared for each fouling experiment such that the change in the

153

foulant concentration was negligible during the fouling. The feed solution was circulated

154

through the upper channel (112 mm×53.5 mm×2 mm) using a gear pump (Cole Parmer,

155

model No. 75211-35, USA); the crossflow rate was controlled at a constant value of ~140

156

mL/min yielding a bulk velocity of ~6 cm/s (the Reynolds number was ~230 in terms of

157

the hydraulic diameter of the channel).

The concentration of the foulant

158

The duration of each fouling experiment was 180 min and the permeation flux was

159

continuously measured using a digital balance (Mettler Toledo, ML4002E, USA). The

160

3D OCT scans were performed within a region (2 mm×2 mm) approximately centered at 8 ACS Paragon Plus Environment

Page 8 of 33

Page 9 of 33


161

the membrane surface. The scan range in the depth direction was ~1 mm that was

162

covered by 700 data points. Although each 3D scan (consisting of 500×500 A-scans)

163

could be done in ~10 s using the maximum A-scan rate (30 kHz), extra time (~8 s) was

164

needed to transfer the data from the video memory to the hard disk. Therefore, the 3D

165

scan rate was set at 2 scans per minute (spm) for the initial fouling and then decreased to

166

1 spm and 0.5 spm after 5 min and 90 min, respectively. The Doppler imaging was

167

performed at the end of each fouling process by recording the variations in both structural

168

intensity and phase angle shift at sampling rate of ~4 fps. All the fouling experiments

169

were repeated at least 3 times to ensure a reliable analysis.

170

2.3 Data analysis methods

171

Each 3D OCT scan generated a dataset containing a 500×500×700 matrix of OCT

172

intensities. Every entry in this matrix was referred to as a voxel when converted into a

173

grayscale index (0 ~ 1) for 3D image rendering; each of these voxels corresponded to a 4

174

µm×4 µm×1.5 µm space that could be occupied by the foulant particles, membrane, or

175

liquid. In this study the foulant voxels were defined as the voxels corresponding to the

176

regions where the foulant particles were significantly concentrated or formed a stagnant

177

layer. The data analysis was then aimed at identifying the foulant voxels that were

178

connected with the voxels representing the membrane surface. It was expected that these

179

foulant voxels had the most valuable information characterizing the fouling, which were

180

therefore referred to as the fouling voxels.

181 182

In prior studies30,

37

the pixels/voxels identification was implemented simply using

background subtraction.39 That is, all datasets were digitally compared with the one



183

sampled at time 0 and it was assumed that the pixels/voxels showing a significant

184

intensity change should correspond to the foulant layer. However, this approach may fail

185

to deal with the datasets describing the initial fouling that is a critical period of the cake

186

formation. The major concern is that the membrane surface could be shifting due to

187

several factors (e.g., membrane compaction and flow fluctuations), whereas the

188

background subtraction assumes that the background (i.e., the membrane and fluid) never

189

changes. The errors caused by the unstable membrane surface cannot be ignored when

190

only a small amount of foulants is conveyed onto the membrane surface during the initial

191

fouling.

192

In order to overcome the difficulty in analyzing the initial fouling, a novel strategy was

193

proposed in this study. The key idea was to identify the membrane surface voxels from

194

the dataset sampled at time 0 and to track the surface shifting in the subsequent datasets.

195

Knowing the precise coordinates of the membrane surface, one can readily determine the

196

foulant voxels in a way independent of the initial dataset; that is, an intensity-based

197

threshold39 for the foulant voxels can be chosen merely in terms of the background noise

198

in the liquid regions since the membrane voxels (i.e., the voxels beneath the membrane

199

surface) have been ruled out. In this work the coordinates of the membrane surface in the

200

OCT intensity matrix were estimated by thresholding the corresponding matrix of

201

intensity gradients and labeling the connected voxels in the resulting binary matrix.39

202

The surface shifting was determined by comparing the intensity matrix with the one

203

sampled at an earlier time and the shifted coordinates were then calculated in terms of the

204

statistical similarity (i.e., shifting the coordinates to a position that minimizes the cross-

205

correlation coefficient40). A foulant voxel was identified as a fouling voxel if it was 10 ACS Paragon Plus Environment

Page 10 of 33

Page 11 of 33


206

connected with at least one foulant voxel directly touching the membrane surface. All

207

these algorithms were programmed using MATLAB.

208

209

3. RESULTS AND DISCUSSION

210

3.1 Visualizing a cake layer via 3D OCT imaging

211

The intensity of a foulant voxel is mainly determined by the relative refractive index

212

(RI) of the foulant compared with the liquid. When the particle sizes are much smaller

213

than the diameter of the light beam, it is reasonable to assume that the OCT intensity

214

should be approximately proportional to the concentration of the particles.

215

assumption is the key to understanding the OCT images obtained from the fouling

216

experiments with the silica nanoparticles whose sizes were ~10 nm as shown by the SEM

217

(in Supporting Information Figure S-1a) and much smaller than the light-beam size (~4

218

µm).

This

219

Although single silica nanoparticle could not be visualized by the OCT, significant

220

signals were detected within the regions where the nanoparticles were concentrated or

221

formed a cake. Using the algorithms described in Section 2.3, the fouling voxels were

222

identified in each OCT dataset and used to describe the fouling process in a 3D fashion.

223

The upper panel of Figure 2a shows a 3D rendering that was generated from the

224

identified fouling voxels corresponding to the sample at t=180 min. It is evident that the

225

rejected silica nanoparticles formed a cake layer with a number of tapered protuberances

226

or peaks. The lower panel of Figure 2a displays a cross section of the cake layer, which

227

corresponds to the region sliced by a blue frame in the 3D rendering. This 2D image 11 ACS Paragon Plus Environment


228

clearly indicates that the calculated surface coordinates (denoted in red) consistently

229

matches the interface between the membrane and cake layer, while the foulant cake (the

230

upper boundary is denoted in green) is precisely separated from the background.

231

Compared with the cake layer formed by the silica nanoparticles, whose roughness was

232

~37.5 µm in terms of the roughness parameter Ra,41 the cake layer formed by the

233

bentonite particles had a relatively smooth surface (Ra=~12.3 µm) as given by the 3D

234

rendering (upper panel) and cross section (lower panel) in Figure 2b, which were

235

generated by the OCT dataset sampled at t=180 min.

236

morphology could be primarily attributed to the difference in the particle shape. Unlike

237

the spherical silica nanoparticles, the bentonite particles are platelets whose larger

238

dimensions (~10 µm) could be a hundred times greater than their thicknesses (see the

239

SEM in Supporting Information Figure S-1b). It is interesting to note that some striped

240

patterns appeared on the surface of the bentonite cake and the orientation of these stripes

241

was coincident with the bulk flow direction (denoted by the blue arrow). It will be clear

242

in the later discussion that these stripes could result from particle groups moving on the

243

cake surface.

244

3.2 Quantifying the evolution of membrane fouling

The changes in the cake

245

The classical fouling models account for the external fouling during a membrane

246

process in terms of surface blockage42 and cake filtration.43, 44 These fouling mechanisms

247

were rarely validated by direct observations in prior studies. The data analysis method

248

proposed in this study makes it possible to apply the 3D OCT imaging to quantify the


Page 12 of 33

Page 13 of 33


249

evolution of a fouling process. That is, useful information on the fouling process can be

250

obtained by analyzing the variation in the digitalized cake layer.

251

In this study an important quantity employed to characterize the progress of the fouling

252

process is the specific deposit, which is defined as the volume of the foulants deposited

253

over a unit membrane area and has the units of [mm3/mm2]. When a cake layer is

254

digitalized using the proposed method, the specific deposit can be approximately

255

evaluated by,

256

Specific Deposit ≈

Number of Fouling Voxels ⋅ Voxel Height . Number of Membrane Surface Voxels

257

In addition, the surface coverage, i.e., the fraction of membrane surface covered by

258

foulants, is also of particular interest in analyzing the initial fouling. In terms of the

259

digitalized cake layer, this quantity can be approximately evaluated by,

260

Surface Coverage ≈

Number of Foulant Voxels Directly Connected with the Surface . Number of Membrane Surface Voxels

261

A series of OCT datasets was sampled during the 180-minute fouling; the specific deposit

262

and surface coverage were then calculated for each dataset. The calculated results are

263

plotted as a function of time and compared with the measured flux decline curves in

264

Figure 3.

265

As a conventional approach to demonstrating a fouling process, the plots of permeation

266

flux versus time in Figure 3a clearly indicate that more severe flux decline resulted from

267

the fouling by the bentonite particles. However, it is difficult to infer the underlying

268

fouling mechanisms according to these flux decline curves. Although some fouling 13 ACS Paragon Plus Environment


269

models44-46 can be used to obtain a best fit to the flux decline data, which offers some

270

insight into the mechanisms dominating different fouling periods, direct experimental

271

evidence is not available to support the modeling.

272

Figures 3b and 3c illustrate the temporal variations in the specific deposit and surface

273

coverage, respectively.

These OCT characterization results clearly demonstrate a

274

transition of the fouling mechanisms; that is, both the fouling processes were dominated

275

by surface blockage during the initial fouling (~10 min), while cake growth accounted for

276

the long term fouling behavior. To the best of our knowledge, this is the first time that

277

direct experimental evidence has been provided to validate the fouling-mechanism

278

transition predicted by various mathematical models.44, 46 The flux decline curves shown

279

in Figure 3a indicate that the fouling mechanism transition should correspond to the

280

initial period in which the fouling resulted in a relatively high decline rate, but the flux

281

decline curves themselves provide little information about the transition.

282

Comparing the two foulants, a relatively rapid surface blockage was observed for the

283

bentonite particles where most of the membrane surface (~95%) was covered by the

284

bentonite particles after ~20 min, while it took ~40 min for the silica nanoparticles to

285

achieve the same coverage. It should be noted that the bentonite particles had a flake-like

286

shape whose larger dimensions were ~10 µm, whereas the silica nanoparticles were

287

spheres with a diameter of ~10 nm. Therefore, it is reasonable to infer that a larger area

288

could be covered by the bentonite particles than the silica nanoparticles when the same

289

amount of foulants was conveyed onto the membrane surface, thereby giving rise to a

290

higher blockage rate. However, what needs to be emphasized here is that, owing to the

291

resolution and sensitivity limits of the current OCT system, the specific deposit and 14 ACS Paragon Plus Environment

Page 14 of 33

Page 15 of 33


292

surface coverage measured by the OCT-based method could be less than the true values.

293

For example, an extremely thin deposition layer (e.g., a monolayer of the silica particles)

294

may result in a very small change in the signal intensity that could be covered by the

295

background noise.

296

As claimed in prior studies,47 the competition between the particle polarization and

297

depolarization mechanisms will limit the growth of the cake layer formed during a

298

constant TMP process. This prediction is directly validated by the OCT results for the

299

bentonite particles, whose specific deposit attained a maximum at t=~60 min as shown in

300

Figure 3b. However, this maximum was followed by a dramatic decrease before it

301

reached a relatively stable value. This decrease in the specific deposit could be caused by

302

the increase in the cross-flow velocity since the cake layer (~0.09 mm) markedly reduced

303

the effective height of the fluid channel (~1 mm). Another possible reason could be cake

304

compaction considering the loose packing of the bentonite platelets.

305

It is interesting to note that the corresponding permeation flux continued to decrease

306

even when the specific deposit had become effectively constant. This is possible as the

307

voids in the cake layer could be gradually infiltrated and clogged by small particles that

308

cannot be detected by the OCT. Similar trends were observed by Baker et al.48 during

309

crossflow filtration of TiO2 suspensions.

310

measured directly by cake removal and analysis. Although cake load tended to plateau,

311

the resistance continued to grow and this coincided with a measured increase in the fine

312

content of the cake. The trend for the silica nanoparticles in our study was different as

313

the specific deposit continued (toward a far asymptote) (Figure 3b) and flux slowly

314

declined (Figure 3a). These trends are consistent with our other studies using colloidal

In that study cake load (mass/area) was



315

SiO2 for UF at constant flux, where deposit height increased toward an asymptote and

316

resistance steadily increased; in this case the increase was attributed to slow cake layer

317

aggregation due to colloid instability.49

318

3.3 Analyzing the fouling behavior on a micron scale

319

While the temporal variations in the specific deposit and surface coverage characterize

320

the fouling process in a statistical sense, the variation in the foulant distribution can be

321

readily analyzed on a micron scale by converting the 3D intensity matrices into a series

322

of grayscale 2D images. The grayscale values range from 0 to 1 and each of them

323

correspond to the cake thickness at a spot (~4 µm×4 µm), which was scaled by the

324

maximum of the cake thicknesses throughout the 180-min fouling.

325

Approximately 140 datasets were sampled (at varied sampling rates) in each fouling

326

experiment. All these datasets were converted into a series of grayscale images that was

327

used to generate a video clip showing the variation in the foulant distribution in a

328

continuous fashion.

329

representative datasets ( t=1, 2, 5, 10, 20, 30, 60, 120, and 180 min) are used to elucidate

330

the fouling evolution, while the corresponding video clips can be found in Supporting

331

Information Figures S-2 and S-3.

Owing to format limitations, image sequences composed of 9

332

Figure 4 displays the image sequence for the fouling by the silica nanoparticles. It is

333

evident that a number of particle groups were randomly deposited on the membrane

334

surface during the very initial fouling (t=0 to 2 min) and the number and size of these

335

particle groups were progressively increased (t=2 to 10 min). As a result, the covered

336

regions were gradually connected with each other to yield a non-uniform cake layer (t=~ 16 ACS Paragon Plus Environment

Page 16 of 33

Page 17 of 33


337

20 min). Although the deposition tended to balance the cake thicknesses at different

338

locations, some protuberances consistently appeared on the cake surface during the long

339

term fouling (t=30 to 180 min) as indicated by the spots that are markedly darker than the

340

neighboring regions. Another interesting observation is the dark stripes appearing in

341

Figures 4h and 4i, though they are hardly discernible in the corresponding 3D rendering

342

(Figure 2a).

343

The image sequence in Figure 5 depicts the fouling by the bentonite particles.

344

Although these grayscale images indicate a similar transition from surface blockage to

345

cake growth, the stripes were much more significant and almost dominated the entire

346

fouling process. A similar striping phenomenon was reported about 30 years ago in a

347

study by Jonsson50 that investigated concentration polarization during UF. This work

348

demonstrated pictures taken by a traditional camera that showed dark stripes on the white

349

membrane surface where the concentration boundary layer was formed by iron dextran,

350

which is a red colored macromolecular complex. In comparison, imaging the striping

351

phenomenon via OCT is to a great extent independent of the colors of the particles and

352

membrane. Therefore, the stripes can be clearly observed in the grayscale images even

353

when most of the membrane surface has been covered by a thick cake layer, which could

354

significantly reduce the contrast between the stripes and background in an image taken by

355

a traditional camera.

356

In 1991, Larsen51 proposed a mathematical model that accounts for the striping

357

phenomenon in terms of the linear theory of hydrodynamic stability.

358

indicates that the stripes could be dissipative structures arising from small disturbances in

359

the boundary layer. When decomposing the disturbances into a series of normal modes 17 ACS Paragon Plus Environment

This model


360

corresponding to different spatial wavenumbers, Larsen established a stability criterion

361

indicating that only the normal mode with a specific spatial wavenumber kc can be

362

sustained for a particular hydrodynamic environment, i.e., the neutral stability. The

363

dimensionless spatial wavenumber γ≡kc·h was mathematically correlated with a modified

364

permeate Reynolds number that was defined as Re'≡(JV·h/υ)(h/δ0), where JV , h, υ, and δ0

365

are the permeate flux, channel height, solution kinematic viscosity, and the thickness of

366

the concentration boundary layer, respectively.

367

The γ-Re' correlation derived by Larsen can be used to estimate the spatial

368

wavenumber of the stripes when assuming a quasi-steady state. The thickness of the

369

concentration boundary layer can be estimated using the equation that correlates the

370

Sherwood number with the Reynolds and Schmidt numbers.52 For the bentonite particles

371

whose sizes are approximately 10 µm, the estimated thickness of the concentration

372

boundary layer is approximately 3 µm and thereby gives rise to Re'≈14. In terms of the

373

γ-Re' correlation, the estimated spacing between the stripes is ~80 µm when achieving

374

the neutral stability, which is close to that observed in Figure 5 (the spacing typically

375

Analyzing the Evolution of Membrane Fouling via a ... - ACS Publications

Recommend Documents