Separation and Purification Technology 126 (2014) 52–61

Contents lists available at ScienceDirect

Separation and Purification Technology journal homepage: www.elsevier.com/locate/seppur

Combination of Single-Photon Emission and X-Ray Computed Tomography to visualize aerosol deposition in pleated filter P.-C. Gervais a,b,⇑, S. Poussier c,d, N. Bardin-Monnier a,b, G. Karcher c,d, D. Thomas a,b a

Université de Lorraine, LRGP, UMR 7274, Nancy, F-54001, France CNRS, LRGP, UMR 7274, Nancy, F-54001, France c Université de Lorraine, NanCycloTEP, Vandœuvre-lès-Nancy, F-54500, France d CHU Nancy, Pôle Imagerie, Service de Médecine Nucléaire, Vandœuvre-lès-Nancy, F-54500, France b

a r t i c l e

i n f o

Article history: Received 6 September 2013 Received in revised form 2 December 2013 Accepted 10 February 2014 Available online 24 February 2014 Keywords: Pleated filter Aerosol filtration Initial deposit Clogging rate Nuclear imaging

a b s t r a c t We use a combination of Single-Photon Emission Computed Tomography and X-Ray Computed Tomography to visualize the behavior of a radioactively marked aerosol in pleated filters under different operating conditions. We first validate this atypical method as a mean to comprehensively observe the filtration process in a filter at the macroscopic scale. This non-intrusive technique highlights the influence of the filtration velocity on the areas where the particles first tend to settle out in blank HEPA filter. We demonstrate that the pleating geometry and the local media permeability act on the flow and account for the preferential location of the deposit. Moreover, e show that the increase in the filtration rate leads to a more homogeneous distribution of the tracer on the entire height of the pleats and hence a more uniform arrangement of the flow. The rigid separators, placed on the media to increase the effective filtration surface, act as obstacles around which the flow is splitted, thus reducing the available area of filtration. Surface observations of loaded filters show that an inhomogeneous growth of the cake induces the formation of preferential channels for the solid aerosol to flow in. Caution should then be taken when carrying out the tomographic analyses because of the competition between the local air resistance and the local efficiency that can prevent from determining the areas where the radioactive aerosol accumulates. Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction When it comes to air purification, media filters are so far the most efficient filters within the different filtration devices. Among them, fibrous pleated filters are widely used in areas related to air treatment such as air admission, bacteriological and nuclear containment, clean rooms, etc. In addition to the fact that they are easy to use and maintain, High Efficiency Particulate Air (HEPA) pleated filters provide excellent purification efficiency. Their lifetime is however conditioned by the pressure drop due to the clogging. Pre-filters are usually installed upstream of HEPA filters. In normal conditions of use, these latters are not exposed at high concentration of particle. But, in accidental circumstances, typically in event of fire, the particle concentration can become very high. The prefilter clogging can lead to its rupture. The HEPA filter is then in a position of last barrier to maintain the efficiency. The velocity can be affected and the pressure drop can drastically increase. ⇑ Corresponding author. Present address: The Institute for Radiological Protection and Nuclear Safety, Saclay, France. Tel.: +33 648101372. E-mail address: [email protected] (P.-C. Gervais). http://dx.doi.org/10.1016/j.seppur.2014.02.011 1383-5866/Ó 2014 Elsevier B.V. All rights reserved.

For example, the experiments of [1], using soot combustion aerosols, were performed until 20 times the initial pressure drop of the filter. The topic is of great interest in occupational and environmental hygiene research as well as nuclear safety. Models have been developed in an attempt to predict the pressure drop and the efficiency of pleated filter media. They can be classified as follows:  The specific models [2–5]. They have originally been developed to optimize the design of a specific pleated filter element, for a given application. They come from numerical analysis from which the equations describing the motion of the fluid are determined through finite element methods. Although the results obtained with these models are satisfactory, they are limited to a specific geometry of pleat. Moreover, they are generally difficult to use and/or to adapt for a user who is not familiar with numerical analysis.  The generalist models [6,7]. They do not require numerical analysis and are easier to use than the specific models. They come from phenomenological approaches and are intended to cover

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

a wide range of pleat geometries. Nevertheless, they require an experimental fitting in order to fix a parameter due to the pressure drop of the singularities. In the current state of knowledge, the use of models is hardly convincing due to the wide range of operating conditions (filtration rate), the aerosol characteristics (nature, size distribution) as well as the geometry of the pleated filter itself. Besides, in this specific case, clogging results in the variation of the local velocity in the pleats. This leads to an inhomogeneous deposition and thus a partial filling of the pleats. Del Fabbro et al. [8] notably observed the formation of arches in between pleats, deposit at the bottom of the pleats and fully clogged pleats. Clogging is then a complex phenomenon and calls for the development of new models relevant for pleated filters. One can decompose the clogging into three distinct phases, the first two being equivalent to those observed for flat filters. First of all a period of in-depth filtration, during which the particles are trapped within the media, has to be considered. Then, a filtration at the surface of the media, with the formation of a cake of particles, follows. The final step, specific to pleated filters, consists in a reduction of the filtration surface that results in a significant rise in the pressure drop. Developing a thorough model requires full comprehension of all the physical processes involved: flow in porous media, particle deposition and particle/media interactions. Unfortunately, it remains very time-consuming to observe them experimentally due to the large number of parameters to take into account. As a consequence numerical approaches, consisting of building virtual structures together with solving transport equations, can be considered as a powerful tool even if they require a validation step. They allow an overall characterization of each process within a reasonable time compared to experiments [9–12]. Observing the preferential deposit in a specific filter can give some clue about the physical phenomena involved. Furthermore, the implementation of an original technique in order to characterize aerosol deposit on a particular type of filter, could be used to experimentally prove the efficiency of the numerical tool. The purpose of this work is to contribute to improve the knowledge of the flow in HEPA pleated media as well as the clogging phenomenon in the particular case of a micron-sized solid aerosol filtration thanks to the combination of Single-Photon Emission Computed Tomography (SPECT) and X-Ray Computed Tomography.

2. Influence of operating conditions on the pressure drop of filter media This review focuses on solid aerosols. Numerous experiments conducted on HEPA flat filters showed the influence of solid particles size on the evolution of the pressure drop [13]. During the indepth filtration period, the geometry of the deposit depends on the collection mechanisms and consequently on the particle size [14]. Submicron-sized particles, mainly subjected to interception and Brownian diffusion mechanisms, preferentially deposit on top of each other and form dendrites [15–17]. This type of highly porous deposit, composed of particles with high specific surface area, offers a large apparent volume and therefore a high resistance to the flow. Accordingly, for the same collected mass, submicronsized particles cause a higher pressure drop than the micron-sized particles. Those latters are collected as compact clusters of low specific surface area. Thomas et al. [18] conducted experiments to measure the pressure drop of fiberglass flat filter during the filtration process. A 0.31 lm uranine aerosol was used in a range of velocity from 0.01 to 0.5 m/s. The results indicate that the pressure drop due to the cake of particles does not depend on the filtration velocity.

53

Similar findings have been observed in a previous work on submicron-sized particles [19]. For larger particles, contradictory results have been reported and no clear trend emerged [20,21]. This shows the complexity of the filtration velocity influence on the pressure drop associated to the collection of micron-sized particles. In the case of pleated filters, the first experimental study on the influence of operating conditions on the aerosol deposit has been carried out by Del Fabbro et al. [6]. Through surface observations of filters after clogging, the authors noticed that the filtration velocity influences the heterogeneity of particles deposit in the pleats. When increasing velocity, from 0.01 to 0.10 m/s, the particles lay at the bottom of the pleat due to inertial effects. For low velocities, the particles are arranged progressively causing a faster filling. Regarding to the particle diameter, ranging from 1 to 8 lm, for the same collected mass, the pressure drop for smaller particles increases due to their higher specific surface. Nevertheless surface observations require the cutting and unfolding of the filter, which obviously induces a change in the structure of the deposit. Moreover, no observation has been made at the beginning of the filtration process, and consequently the role of in-depth filtration on the heterogeneity of the initial deposition has not been evidenced yet. Recently, [22] achieved the only experimental study of threedimensional visualization identified to date. This work has been performed by X-ray microtomography and the aim was to validate a numerical simulation method to predict the clogging point of pleated filter elements [23]. The author finds encouraging results by comparing the values of cake heights from numerical and experimental results. These conclusions from three-dimensional visualizations are still limited to a single case study. Unfortunately they do not allow characterizing the influence of the operating conditions on the heterogeneity of particle deposit. 3. Experimental work 3.1. Experiments principle Experiments aim for the in situ visualization of the filtration process with no alteration of the three-dimensional structure of the particle deposit. We based our approach on medical imaging techniques and the determining of the position of a 99mTc marked aerosol by combining SPECT with X-Ray Computed Tomography. Two kinds of experiments have been carried out:  First, we wanted to highlight the influence of the filtration velocity on the preferential area of initial deposit in the blank HEPA filter. To achieve this, we have generated, filtrated and located a monodisperse radioactive aerosol, which traces the flow.  In a second step, we focused on the dynamics of the deposition to determine the impact of the clogging level on the flow. Filters were preloaded by means of a solid aerosol of alumina and then the radioactive aerosol was generated, filtered and located using SPECT-CT. 3.2. Materials 3.2.1. Mini-pleated filters The studied HEPA filters are H14 classified by the European norm EN1822/2009 that defines classes for filters by their retention at most penetrating particle size (MPPS). H14 class corresponds to an efficiency over 99.995%. A mini-pleated filter (Fig. 1) consists of a 15 cm  15 cm square stainless steel frame; the fibrous medium is made of fiberglass to which organic binders have been added at low concentration to ensure the mechanical rigidity. To avoid the media/media contact in order to optimize

54

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

Fig. 1. Picture of mini-pleated filter designed by Camfil in experimental form.

the filtration surface, rigid separators are used. They are made of hot-melt adhesive and ensure a uniform pleat spacing. Six separators are placed each 2.2 centimeters perpendicular to the pleats direction. Tightness between the medium and the frame is provided by a polyurethane sealant. The pleating characteristics are the following ones: the height of pleat h = 27.5 mm and the distance between adjacent pleats p = 2.2 mm. The tested filters are not electrically charged. The characteristics of the fibrous media are given in Table 1.

3.2.2. Experimental test bench The experimental set-up (Fig. 2) is a ventilation line made with cylindrical pipe (diameter 40 mm) in stainless steel. It is composed of a monodisperse aerosol generator (MAG 2010 from Palas GmbH, Germany) able to produce Di-Ethyl-Hexyl-Sebacat (DEHS) based aerosol. This apparatus operates according to Sinclair-LaMer principle [25]. It consists of a nebulizer, supplied with nitrogen in a 0.5 g/L solution of sodium chloride in order to generate the condensation cores, an evaporator for vaporizing DEHS depending on

the temperature, a heater and a condenser, in which the DEHS condenses heterogeneously around the condensation cores. The volume flow is 3.5 L/min. Upstream of the tested pleated filter, an optical particle counter (WelasÒ digital 2100HP from Palas GmbH, Germany) is used for real-time aerosol characterization in the particle size range of 0.15–10 lm. This sensor is based on the scattered light analysis of the individual particle. The system is combined with a digital individual signal processing which allows coincidence correction. Depending on the airflow, brass nozzle tips with a diameter between 2 and 5 mm, allow fitting the sampling line in order to respect isokinetic sampling. PDControl 1.0 Palas Software is used for temporal registration, coincidence and particle size analysis. Pre-filtered airflow in the pipe is maintained by a side channel blower provided by Elmo Rietschle (Gardner Denver, France). An external frequency converter is used for controlling the electric motor speed to 5000 rpm. The corresponding flow range of interest extends from 20 to 80 m3 =h, which corresponds to a wide area around the filter nominal flow rate (43 m3 =h). The airflow is directly verified through a diaphragm flow meter (M3 series from Eletta Flow AB, Sweden) whose the reliability has been previously shown by comparison with a volumetric flow meter. The design of the filter holder was optimized using the CFD code from FLUENT Inc. to provide a flat velocity profile for the entire range of flow rates studied. Pressure tappings on both sides of the filter holder allow connecting a differential pressure sensor to monitor the filter clogging degree. The temperature and relative humidity of the airflow are directly checked downstream of the filter with the appropriate sensors. A safety filter is located before the mechanical fan to avoid contamination of the exhaust air.

3.2.3. Radioactive aerosol generation In order to generate radioactive condensation cores, the 99m TcO 4 based solution is added to the sodium chloride solution

Table 1 Main characteristics of fibrous media [24]. Count median diameter (lm)

Geometric standard deviation (–)

Thickness (lm)

Weight ðg=m2 Þ

Packing density (–)

Filtration surface ðm2 Þ

0.6

2.2

521  31

92  2

0.071  0.006

0.42

Fig. 2. Schematic diagram of the experimental setup.

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

55

2

St ¼

Cu  qp  dp  U 18  qf  L

ð2Þ

where Cu is the Cunningham correction factor for air, dp is the particle diameter, qp and qf are the particle density and the air density, respectively. Fig. 4 also displays the Stokes number evolution as a function of the particle size, calculated in the worst case, i.e. for the largest fluid velocity. We note that St  0:1, meaning that the particles are not subjected to inertia. As a consequence, the particles react almost instantly to the movements of the fluid. The radioactive monodisperse aerosol can be considered as a tracer of the flow.

Fig. 3. Frequency and cumulative volume particle size distributions of the radioactive monodisperse aerosol.

in the MAG nebulizer. It is mainly composed of sodium chloride (9 mg/mL), sodium nitrate (0.5 mg/mL) and sodium pertechnetate (0.097 lg/mL). MAG is set in order to produce submicronic monodisperse aerosol, which traces the flow. The particle concentration is 106 P=cm3 . The resulting radioactive monodisperse aerosol is then characterized by the optical particle counter. Fig. 3 displays the frequency and cumulative volume particle size distributions of the radioactive monodisperse aerosol. The particle size distribution ranges from 0.45 to 0.55 lm in volume. It is an important point because it influences the way the particles travel through the flow [26]. Submicronic particles are subjected to Brownian diffusion while larger than 1 micron particles undergo inertial motion. Intermediate size particles tend to follow the streamlines according to convective transport. The ratio between the convective and the diffusive transport rate can be determined by the Péclet number:

3.2.4. Solid aerosol generation For the second step of the experiments, a solid aerosol is generated. The generation is carried out by rotating brush generator (Palas RBG 1000), as shown in Fig. 2. Inside the pipe, the injection takes place 110 cm upstream of the media. During the generation, a conical brass plug allows condemning the online sample without noticeable modification of the flow field. The principle of the generator is based on the contact of solid particles, alumina in the present case ðqAl2 O3 ¼ 3970 kg=m3 Þ, with a rotating brush. Before the experiments, moisture is removed from the powder in a drying-oven at 333 K during 24 h. The powder is placed in a tank with a piston, the aerosol mass flow rate is 135.7 g/h. Fig. 5 shows the frequency and the cumulative volume particle size distributions, through the aerosol counter. The mass median diameter of alumina particles is centered at 1.0 lm with a geometric standard deviation rg ¼ 3:2. 3.3. Non-invasive imaging methods

where L is the characteristic length of the flow, U and D are the face velocity and the diffusion coefficient, respectively. Fig. 4 depicts the Péclet number evolution as a function of the particle size. It is calculated in the worst case, i.e. for the lowest fluid velocity. Because of the symmetry, the inner half-side of the filter holder (6.5 cm) is chosen as the characteristic length. We note that Pe  1, meaning that the particles are not subjected to diffusion. The Stokes number can also be calculated in order to quantify the influence of inertia:

3.3.1. Single-Photon Emission Computed Tomography SPECT is a nuclear imaging technique used in medical imaging. It allows estimating the 3D distribution of the radioactive tracer [27]. In the case of a functional diagnosis, the distribution of the radiopharmaceutical (radioactive tracer associated with a biological tracer) is measured in vivo. In our case, only the tracer is used to follow the monodisperse aerosol. This non-invasive method allows observations throughout the filter without damaging the fibrous media. The interpretations can be qualitative (detection of lesions in lung scintigraphy) or quantitative (measurement of tracer concentration). The qualitative approach allows us to observe the initial aerosol deposit in pleated filters as well as the overall uniformity of the deposit. Quantitative interpretations of the average intensity will help characterizing the influence of operating conditions on the preferential deposit areas.

Fig. 4. Péclet and Stokes number values as a function of the particle size of the radioactive aerosol.

Fig. 5. Frequency and cumulative volume particle size distributions of the alumina powder.

Pe ¼

LU D

ð1Þ

56

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

The choice of radioactive tracer is based on its physical properties (type and energy of radiation, period). Those that emit gamma rays are favored for their benefit/risk ratio: detection quality/radiation received by the patient. Currently, the radioisotope that best satisfies these criteria is technetium-99m (99mTc), whose is used in 90% of medical examinations. 99mTc is an isomer of the isotope of technetium, which isomeric transition from 99mTc to 99Tc emits gamma radiation at 141 keV. With a relatively short period (T = 6 h), 99mTc has an activity of 1:95  1017 Bq=g. 99mTc is produced from molybdenum-99 (99Mo), which decays with 87% probability to 99mTc. 99Mo is recovered as a product of the nuclear fission reaction of uranium-235. 3.3.2. X-ray Computed Tomography X-ray Computed Tomography, also known as scanner, is a medical imaging technique that can measure the distribution of the human tissues density in a given cutting plane. The information come from the attenuation of an X-rays beam due to the crossing of an absorbent body [28]. A ring-shape linear multi-detector rotating around the center of interest is located opposite to the X-ray source. The resulting cutting plane is described by a function which is proportional to the energy density for a given linear attenuation coefficient. 3.3.3. Routine protocols SPECT acquisitions were performed using a conventional double-head c-camera (Symbia T2 SPECT/CT Siemens Medical Healthcare). The images are reconstructed from the acquisition of projections (planar scintigraphy). Thirty-two projections of 30 s each were acquired on a 180° circular orbit centered on the filter (from 0° to 180° orientations, 200-mm radius of rotation), 128  128 matrix, 126–154 keV energy window. The total acquisition time was 20 min. The images were thereafter reconstructed using an ordered-subset expectation maximization iterative process (‘‘Flash3D’’, Siemens Medical Healthcare) iterative process (4 iterations and 8 subsets). The voxel size of the reconstructed images was 4.8 mm and the intrinsic spatial resolution of the detectors was 3.8 mm in the central slice when determined by the full width at half maximum on a point source of 99mTc placed at 10 cm of the detector without scatter. Attenuation and scatter corrections were performed. The images were slightly post smoothed with a 3D spatial Gaussian filter (full-width at half maximum 5 mm).

Fig. 6. x axis dissociation in VOIs.

CT scans were obtained before scintigraphy using scanners (Siemens Medical Healthcare). The CT scans were performed according to the routine protocols parameters (30 mA, 130 kV). Image reconstruction resulted in images with a slice width of 1 mm using a 1mm reconstruction increment with standard B20 reconstruction kernel. The windowing of the CT was adjusted appropriately and kept constant for all the slices. Isotropic multi-planar reconstruction of CT resulting in high spatial and contrast resolution and low imaging noise was used in this study to perform quantification. In order to match SPECT’s matrix to the CT’s matrix, an automatic registration of the SPECT and the CT images was performed using a commercially available 3D volume registration and fusion tool (Inveon Research Workplace, Siemens Medical Solutions), which allows sub-voxel 3D rigid-body transformation with 6 degrees of freedom. The measurement of transversal and axial extensions was performed with the SPECT images displayed in the ‘‘Volcano’’ color lookup table and the CT images in a normal gray scale and a gamma of 1, to maintain linearity. 3.3.4. Filter dissociation in Volumes Of Interest The spatial dissociation of the filter in Volumes Of Interest (VOIs) is performed on the CT images. It is then applied to SPECT images to measure the activity in the filter for post-processing. The dissociation is performed in the three spatial directions as follows: – 28 VOIs of 1 mm  121 mm  121 mm, built along the x axis (see Fig. 6). This projection, perpendicular to the flow, allows determining the preferential deposit area along the pleat depth. – 173 VOIs of 0.7 mm  121 mm  28 mm, built along the y axis (see Fig. 7). The plans are directed tangentially to the flow and longitudinally to the pleats. This representation allows determining the preferential deposit area along the width of the filter. – 43 VOIs of 2.8 mm  121 mm  28 mm, built along the z axis (see Fig. 8). The preferential deposit area along the height of the filter is identified in this way. 3.3.5. Measurement accuracy Considering the need in material ressources (filters, tracer, ccamera time) as well as the human ressources (medical radiation technologist), a reduced number of experiences was performed. The ability to locate a deposit has been beforehand tested and

Fig. 7. y axis dissociation in VOIs.

57

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

Fig. 8. z axis dissociation in VOIs.

validated by the NanCycloTEP nuclear imaging platform. The generation of the radioactive monodisperse aerosol with the provided 99m Tc-based doses led to acquire activity with an intensity three to four times higher than the background noise. For each VOI, the average activity which is called Mean Voxel Intensity (MVI), and the resulting standard deviation are calculated. The standard deviation is not relevant because it takes into account the output channels, where the activity is zero. The MVI is directly proportional to the amount of 99mTc and therefore to the deposit area of the traced particles. The measurement strongly depends on the initial injected activity as well as the radioactive decay between experiment and measurement. In order to compare the acquisitions with each other, the results are presented relative to the total activity present in the filter.

(DPt ¼ 2, 3, 4, 8 and 18DP 0 ). Once the target value is reached, the ventilation rate is gradually decreased until the fan stops. After connecting the MAG generator, a second permeation test is performed during the gradual increase until the nominal flow rate. The next step is to generate the marked aerosol. The total injected activity is between 3 and 16 mCi. All the manipulations are performed at a fan frequency of 31.8 Hz, corresponding to the nominal flow rate at DP 0 . The filter is then removed from the filter holder and weighed. The liquid aerosol mass (approximately 1.5 g) is negligible compared to that of the solid aerosol. Table 2 summarizes all the measurements. If the frequency of the fan is kept constant, the flow is not regulated. Then the increase in the pressure drop due to clogging phenomenon causes a non negligible decrease in the filtration rate, especially for high clogging rates. Thereby, rescaled values of DPt =DP0 take the filtration velocity decrease into account. Fig. 9 presents the clogging curves (rescaled and not). Given the limited number of experimental points, it is difficult to mark out the 3 steps of filtration. However, filter face observations after clogging give an idea of the process. Below a value of 4DP 0 , no particle is visible on the filter face. For a value of 9.2DP 0 , the particles begin to be visible but no pleat is obstructed. The pleat obstruction is observed at 24.7DP 0 , when the filter is totally clogged.

4. Results 4.1. Influence of the filtration velocity on the initial deposition in the blank filter Fig. 10 shows the MVI, relative to the total injected activity, depending on the position in the pleat depth for three filtration rates. A homogeneous concentration of the aerosol as well as an incompressible flow are assumed. If the particles trace the flow, the initial deposit can be assimilated to the velocity profile. On a fluid dynamics side, if the media permeability is homogeneous,

3.4. Experimental protocol 3.4.1. Visualization of the initial aerosol deposition in the blank filter A new filter has to be used for each experiment. Airtightness is checked by a preliminary permeation test over a range of flow rates between 20 and 60 m3 =h. The filtration time of the marked aerosol ranges from 105 to 240 min depending on the 99mTc production capacity as well as the available time of the c-camera. The total injected activity in the generator is generally between 2 and 12 mCi. Tomographic acquisitions occur at variable time after the preparation of the dose. Manipulations are performed at three different fan frequencies, which correspond to three initial filtration velocities: 1.6, 2.5 and 4.7 cm/s.

3.4.2. Visualization of the dynamics of the deposition The experimental setup as well as the studied filters are identical to those used in the previous study. First, the filter is weighed on a precision balance (accuracy: 103 g) and placed in the filter holder. Solid aerosol is then filtered until the target pressure drops

Fig. 9. Relative pressure drop as a function of the aerosol collected mass.

Table 2 Clogging experimental measurements summary. Exp.

DP 0 (Pa)

DP t (Pa)

2DP 0 3DP 0 4DP 0 8DP 0 18DP 0

195  13 199  13 200  13 221  13 200  13

394  16 642  20 835  22 1700  35 3720  66

Clogging time (min)

Measured mass (g)

Flow rate decrease (%)

Rescaled DP t =DP 0

12 13 36 62 53

16.05 27.23 40.18 69.07 108.18

0.98 3.05 3.14 7.15 24.6

2.2  0.2 3.3  0.2 4.2  0.3 9.2  0.7 24.7  2.0

58

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

Fig. 10. Mean intensity per voxel, relative to the total injected activity, measured along the pleats direction. Effects of the filtration velocity on the location of the initial aerosol deposit in the blank filter.

the Stokes flow along the pleat could lead to a homogeneous deposit. However, we can qualitatively observe a preferential initial deposit in the second third of the pleat (between 10 and 20 mm). The shape of the distributions exhibits that the amount of aerosol deposit is much lower as it is close to the folded areas of the media. On the filter media side, the pleating process could damage the media and induce a lower local permeability in the folded areas. In disagreement with the observations of Del Fabbro et al. [6], the filtration rate does not seem to have any influence on the initial deposition at the bottom of the pleats. For the three velocities, the maximum intensity is located at a pleat depth between 12.5 and 18.5 mm. The increase in the filtration rate leads to a more homogeneous distribution of the tracer, and thus, of the flow. This could be explained by the appearance of small sized eddies in the flow along the pleats. To characterize the flow along the pleats direction, Rebaí et al. [5] suggested the channel Reynolds number. It is calculated according to the average longitudinal velocity component, as well as the channel half-width, at the inlet of the pleat entrance channel. As a first approximation, the face velocities (44, 70 and 130 cm/s) correspond to the average longitudinal velocity component at the inlet of the pleat entrance channel. The channel halfwidth is the half-distance between adjacent pleats (2.2/2 mm). The channel Reynolds number is between 31 and 93 which is characteristic of the intermediate regime. These orders of magnitude of Reynolds number are consistent with the supposition of small sized eddies appearance in the flow. Figs. 11 and 12 respectively show the MVI relative to the total activity depending on the position in the width and in the height of the filter for the three filtration rates. For the nominal (2.5 cm/ s) and the lower (1.6 cm/s) velocities, the distribution of the radioactive aerosol is homogeneous. In the highest velocity case, the initial deposit is located on the periphery of the media. This dispersion results from the way the fluid flows upstream of the filter. The Reynolds number of pipe, which characterizes the flow upstream of the filter, is calculated according to the face velocity and to the aerodynamic diameter (the filter holder edge is 13 cm). For the highest velocity case, the Reynolds number of pipe is larger than 104 , which is characteristic of a turbulent flow. We can therefore suppose that eddies appear in the flow, developing recirculations upstream of the filter. The distribution as a function of the position along the filter height exhibits the same behavior. However, the peripheral deposit also occurs for 2.5 cm/s. This representation is built from average activity values in planes perpendicular

Fig. 11. Mean intensity per voxel, relative to the total injected activity, measured along the filter width. Effects of the filtration velocity on the location of the initial aerosol deposit in the blank filter.

Fig. 12. Mean intensity per voxel, relative to the total injected activity, measured along the filter height. Effects of the filtration velocity on the location of the initial aerosol deposit in the blank filter.

to the pleats. The turbulent nature of the flow, upstream of the media, is more significant than in the case of visualization in the width of the filter. It appears that the pleats orientation influences the turbulent intensity. We also note the presence of minima in the distribution of the tracer. For more visibility, Fig. 13 focuses on the nominal filtration velocity. Five periods of 2.2 cm each are repeated regularly. This means that the flow is splitted at regular intervals. This phenomenon can be explained by the 6 rigid separators. Their main function is to prevent the pleats from touching each other in order to increase the effective filtration surface. The gradient of activity on each side of the maxima reveals on the contrary that they form an obstacle for the flow, reducing the available area of filtration. 4.2. Influence of the clogging rate on the deposition at nominal airflow Fig. 14 shows the MVI, relative to the total activity present in the filter, depending on the position in the pleat depth. For the sake of clarity, only the results concerning the 2, 4 and 8DP 0 experiments are displayed. While filters face observations show that increasing the degree of clogging causes a deposit of particles onto

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

59

Fig. 13. Mean intensity per voxel, relative to the total injected activity, measured along the filter height. Effects of the rigid separators on the location of the initial aerosol deposit in the blank filter.

Fig. 15. Mean intensity per voxel, relative to the total injected activity, measured along the filter width. Effects of the clogging level on the location of the aerosol deposit inside the filter.

Fig. 14. Mean intensity per voxel, relative to the total injected activity, measured along the pleats direction. Effects of the clogging level on the location of the aerosol deposit inside the filter.

Fig. 16. Mean intensity per voxel, relative to the total injected activity, measured along the filter height. Effects of the clogging level on the location of the aerosol deposit inside the filter.

the front surface of the filter, the results from the tomographic analysis showed that the level of clogging should not impact the location of the radioactive aerosol. We qualitatively observed a preferential deposit in the second third of the pleat (between 10 and 20 mm). The shape of the three distributions is consistent with the experiments of initial deposit at the nominal flow rate. Note that the MVI is an averaged value on a filter slice, taking all the pleats into account. Each pleat may have a different local level of clogging, as we see on front face observations. This suggests that preferential channels for the radioactive aerosol appear due to the inhomogeneity in the growth of the solid particles cake. Figs. 15 and 16 respectively show the MVI relative to the total activity depending on the position in the width and in the height of the filter. The level of clogging does not seem to follow any trend regarding to the location of the radioactive aerosol. The experimental measurements can be interpreted in different way as suggested by the following discussion on the technique validity.

4.3. Discussion on the technique validity According to the filter front faces observations, it is assumed that the clogging is heterogeneous from one pleat to another.

During the filtration of the radioactive aerosol, the deposit area will depend on the local efficiency as well as on the airflow rate which is related to the local air resistance. As a first approximation, the media may be schematically decomposed into two zones, the clogged area (in dark gray) as well as an area poorly or not clogged (in light gray). If the local particle concentration C is constant, three simple scenarios can be proposed. They are schematically illustrated and can be explained as follows: – The air resistances are of the same order for both areas ðRclogged area Rpoorly=not clogged area Þ and the local efficiency is homogeneous because of the high efficiency of the filter ðEclogged area Epoorly=not clogged area Þ. In this case (see Fig. 17), the local flow rate is homogeneous ðq1 ¼ q2Þ and the radioactive aerosol distribution should be homogeneously distributed on the media surface. – The air resistances are of the same order for both areas ðRclogged area Rpoorly=not clogged area Þ but the local efficiency is higher in clogged areas ðEclogged area > Epoorly=not clogged area Þ. In this case (see Fig. 18), the local flow rate is homogeneous ðq1 ¼ q2Þ but the radioactive aerosol distribution should be preferentially distributed on the surface of the clogged areas.

60

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

As a consequence, a competition is pointed out between the air resistance influence on the flow rate and the efficiency. This phenomenon does not allow us to determine the path where the mass of radioactive aerosol is the most important. For our clogging rate test, due to the very high efficiency of the filters, it is likely that we are in the last case (Fig. 19). Nevertheless it is complex to identify trends regarding the influence of the degree of clogging of the aerosol deposit area, which is a limitation of this experimental technique.

5. Conclusion

Fig. 17. Competition between the local air resistance and the local efficiency: case (a).

Fig. 18. Competition between the local air resistance and the local efficiency: case (b).

An original method was used and validated to observe the filtration process in a filter at the macroscopic scale. Knowing that the particles follow the streamlines, the observations of initial deposition on pleated media allowed us to characterize the flow as a function of the operating conditions. All the representations of our results concern three filtration velocities (1.6, 2.5 and 4.7 cm/ s) distributed on both sides of the nominal filtration rate. The results indicate that the flow depends on the local permeability of the medium. The larger the local airflow resistance, the smaller the tracer quantity. Moreover, the increase in the filtration rate subsequently leads to a more homogeneous distribution of the tracer on the entire height of the pleats and hence a more uniform arrangement of the flow. The rigid separators, placed on the media to increase the effective filtration surface, in fact, act as obstacles around which the flow is splitted, reducing the available area of filtration. Although the results are encouraging, we must remain cautious. The distributions as a function of the position in filter height and weight show deposit located on the periphery of the media. This dispersion results from the way in which the fluid flows upstream of the filter which strongly depends on the bench design. A second part of this work dealt with the visualization of the dynamics of aerosol deposit using SPECT on previously clogged filters. The surface observations of the media were then used to visualize the progression of the clogging process from the early stages until the obstruction of some pleats, characteristic of the reduction step surface. The results from the tomographic analysis show that the degree of clogging would not affect the location of the radioactive aerosol. The change of aerosol and the method of analysis may partly explain the phenomenon. We addressed the issue that a competition exists between the local flow rate, depending on airflow resistance, and the local efficiency to determine the area where the radioactive aerosol is the most important. In order to complete the understanding of the operating conditions on the flow characteristics, a point of interest will consist in determining the particle diameter influence on the initial deposition on the blank filter. Regarding a potential validation of a numerical tool, the interpretation of the influence of the clogging rate on the deposit is a limitation of our experimental technique. An attractive perspective would be to visualize and mapping the aerosol deposit without using a tracer element. Recently, X-ray tomography techniques give promising results on samples of few pleats [29].

Fig. 19. Competition between the local air resistance and the local efficiency: case (c).

Acknowledgements – The air resistance is higher in the clogged area ðRclogged area > Rpoorly=not clogged area Þ and the local efficiency is homogeneous ðEclogged area Epoorly=not clogged area Þ. In this case (see Fig. 19), the local flow rate is greater through the least obstructed areas ðq1 > q2Þ and the radioactive aerosol distribution should be preferentially distributed on the surface of the partially clogged area.

Authors are grateful with the French Environment and Energy Management Agency (ADEME) for the financial support of the ATENA Project (Agreement Number: 11-81-C0084). Radioactive tracer and medical imaging techniques were funded by the NanCycloTEP economic interest group. This work would not have been possible without the Nuclear Medicine staff of the Nancy University Hospital.

P.-C. Gervais et al. / Separation and Purification Technology 126 (2014) 52–61

Authors would like to thank the mechanical service department of the Reactions and Chemical Engineering Laboratory for the helpful technical support in designing the experimental bench. References [1] V. Mocho, F.-X. Ouf, Clogging of industrial pleated high efficiency particulate air (HEPA) filters in the event of fire, Nucl. Eng. Des. (2011). [2] R. Raber, Pressure drop optimization and dust capacity estimation for a deeppleated industrial air filter using small sample data, in: WFC 3, Wilmington, Delaware, 1982, pp. 52–59. [3] H. Yu, C. Goulding, Optimized ultra high efficiency filter for high-efficiency industrial combustion turbines, in: International Gas Turbine and Aeroengine Congress and Exposition, ASME, Cologne, Allemagne, 1992. [4] D.R. Chen, D.Y.H. Pui, B.Y.H. Liu, Optimization of pleated filter designs using a finite-element numerical model, Aerosol Sci. Technol. 23 (1995) 579–590. [5] M. Rebaï, M. Prat, M. Meireles, P. Schmitz, R. Baclet, A semi-analytical model for gas flow in pleated filters, Chem. Eng. Sci. 65 (2010) 2835–2846. [6] L. Del Fabbro, J.-C. Laborde, P. Merlin, L. Ricciardi, Air flows and pressure drop modelling for different pleated industrial filters, Filtr. Sep. 39 (2002) 34–40. [7] S. Callé-Chazelet, D. Thomas, J. Rémy, S. Artous, L. Bouilloux, Performances de filtration des filtres plissés, in: Récents progrès en génie des procédés, Paris, France, 96, SFGP, 2007. [8] L. Del Fabbro, P. Brun, J.-C. Laborde, J. Lacan, A. Renoux, L. Ricciardi, Study of the clogging of industrial pleated filters by solid particles, J. Aerosol Sci. 31 (2000) 210–211. [9] M. Rebaï, M. Prat, M. Meireles, P. Schmitz, R. Baclet, Clogging modeling in pleated filters for gas filtration, Chem. Eng. Res. Des. 88 (2009) 476–486. [10] S. Fotovati, S.A. Hosseini, H.V. Tafreshi, B. Pourdeyhimi, Modeling instantaneous pressure drop of pleated thin filter media during dust loading, Chem. Eng. Sci. 66 (2011) 4036–4046. [11] S. Fotovati, H.V. Tafreshi, B. Pourdeyhimi, A macroscale model for simulating pressure drop and collection efficiency of pleated filters over time, Sep. Purif. Technol. 98 (2012) 344–355. [12] P.-C. Gervais, N. Bardin-Monnier, D. Thomas, Permeability modeling of fibrous media with bimodal fiber size distribution, Chem. Eng. Sci. 73 (2012) 239–248. [13] V.J. Novick, P.R. Monson, P.E. Ellison, The effect of solid particle mass loading on the pressure drop of HEPA filters, J. Aerosol Sci. 23 (1992) 657–665.

61

[14] J.I.T. Stenhouse, D.A. Japuntich, B. Liu, The behaviour of fibrous filters in the initial stage of filter loading, J. Aerosol Sci. 23 (1992) S761–S764. [15] C.N. Davies, Air Filtration, Academic Press, London – New York, 1973. [16] A.C. Payatakes, C. Tien, Particle deposition in fibrous media with dendrite-like pattern. A preliminary model, J. Aerosol Sci. 7 (1976) 85–100. [17] A.C. Payatakes, Model of transient aerosol particle deposition in fibrous media with dendritic pattern, AIChE J. 23 (1977) 192–202. [18] D. Thomas, P. Penicot, P. Contal, D. Leclerc, J. Vendel, Clogging of fibrous filters by solid aerosol particles. Experimental and modelling study, Chem. Eng. Sci. 56 (2001) 3549–3561. [19] P. Pénicot, D. Thomas, P. Contal, D. Leclerc, J. Vendel, Clogging of HEPA fibrous filters by solid and liquid aerosol particles : an experimental study, Filtr. Sep. 36 (1999) 59–64. [20] P. Merlin, S. Artous, J.-C. Laborde, V. Mocho, Influence of the filtration velocity on the gazeous clogging of planar filter medium by solid aerosols, in: CFA 16, Paris, France, 2000 (in French). [21] G. Kasper, S. Schollmeier, J. Meyer, Structure and density of deposits formed on filter fibers by inertial particle deposition and bounce, J. Aerosol Sci. 41 (2010) 1167–1182. [22] P. Hettkamp, Simulation and measurement of dust loading of pleated air filter, in: 24th AFS Conf., Louiseville, USA, 2011. [23] P. Hettkamp, G. Kasper, J. Meyer, Influence of geometric and kinetic parameters on the performance of pleated filters, in: WFC 11, Graz, Austria, 2012. [24] A. Joubert, J.-C. Laborde, L. Bouilloux, S. Chazelet, D. Thomas, Modelling the pressure drop across HEPA filters during cake filtration in the presence of humidity, Chem. Eng. J. 166 (2011) 616–623. [25] Z.D. Ristovski, L. Morawska, N.D. Bofinger, Investigation of a modified SinclairLaMer aerosol generator in the submicrometer range, J. Aerosol Sci. 29 (1998) 799–809. [26] J.G. Santiago, S.T. Wereley, C.D. Meinhart, D.J. Beebe, R.J. Adrian, A particle image velocimetry system for microfluidics, Exp. Fluids 25 (1998) 316–319. [27] I. Buvat, J. Darcourt, La Tomographie médicale : imagerie morphologique et imagerie fonctionnelle, Hermès Sci. (2002) 101–121 (in French). [28] J.-L. Amans, G. Ferretti, La Tomographie médicale : imagerie morphologique et imagerie fonctionnelle, Hermès Sci. (2002) 33–56 (in French). [29] P. Hettkamp, J. Meyer, G. Kasper, Surface dust-loading of pleated filters: comparison of simulation with experiment, in: FilTech 2013, Wiesbaden, Germany, 2013.