The AAC classifies airborne particles by relaxation time (s) and selects particles with a narrow range of aerodynamic diameters (Tavakoli and Olfert 2013; Tavakoli et al. 2014). The experimental setup for fiber separation using the AAC is shown in Figure 1. The glass fiber aerosol was generated using a vortex mixer shaking method that was described in detail previously (Ku et al. 2013, 2017). Briefly, the glass fiber powder was placed in a Pyrex tube that was fixed to a pop-off cup on a vortex mixer (Vortex-Genie 2, Scientific Industries Inc.). During vortex mixing, high-efficiency particulate air (HEPA) filtered air (1.0 L/min) was input through a port at the top of the tube for fiber suspension. The generated glass fiber aerosols were passed through a second port and conductive tubing that led to the inlet of the AAC. The operating conditions of the AAC are listed in Table 1, which includes selected size, cylinder rotation speed, sheath flow rate, and sample flow rate. The aerodynamic size distribution of separated glass fibers from the AAC was measured using an Aerodynamic Particle Sizer (APS, Model 3321, TSI Inc., Shoreview, MN). Three measurements were made for each selected size at the AAC outlet. The separated glass fiber aerosols were collected for further analysis using mixed cellulose ester filter (MCE; SKC Inc., Eighty Four, PA) and polycarbonate membrane (PC; 0.4-μm pore size, 25 mm, Sterlitech, Kent, WA, USA) filters mounted in a cowl sampler (225–321 A, SKC Inc.). The length of glass fiber aerosols separated by the AAC and collected on the MCE filter were inspected with a phase contrast microscope (PCM) for a rough estimate of fiber length. First, the MCE filter was cleared with an acetone vaporizing unit (QuickFix, RJ Lee Instrument Inc., Trafford, PA), and subsequently the length of the fibers was measured using a PCM with 400x magnification and Motic software (Motic Incorporation Ltd., Hong Kong). In addition, the classified glass fibers collected on PC filters were analyzed by field emission scanning electron microscopy (FESEM, model S-4800–2, Hitachi High Technologies America Inc.). Samples were prepared for FESEM analysis by placing PC filters in centrifugal tubes with 5 mL of isopropyl alcohol and extracting fibers from the PC filters using a combination of sonication and vortex mixing. Extracted glass fibers were deposited on another PC filter (0.4-μm pore size) using a borosilicate filtration apparatus (MilliporeSigma, Burlington, MA) and a vacuum pump. The PC filter was placed on a SEM specimen aluminum mount with a conductive carbon double-sided adhesive tape (SPI Supplies, West Chester, PA). The length and width of the classified glass fiber aerosols were examined by the FESEM.

The multi-cyclone sampling array was previously employed for the size-segregation of crystalline silica for toxicological evaluations (Mischler et al. 2013, 2016). In this study, the same sampling array was utilized for the separation of glass fibers using the experimental setup shown in Figure 2. The system included two different cyclones and cowl sampler loaded with a PC filter in a series. The cyclones were a Higgins-Dewell (HD) type (BGI4L, Mesa Labs, Butler, NJ; cut off diameter (d₅₀): 4 μm at 2.2 L/min), and sharp-cut type (SCC) with a 0.74-mm cut size (SCC 0.695, Mesa Labs, Butler, NJ; d₅₀: 0.74 μm at 2.2 L/min) and a 0.38-μm cut size (SCC 0.695, Mesa Labs, d₅₀: 0.38 μm at 4.4 L/min). To match the final flow rate (4.4 L/min), two HD and SCC cyclones with 2.2 L/min were used in parallel. The glass fiber aerosol generation method was the same as that described in the section above. After sampling with the multi-cyclone array, the separated glass fiber aerosols were collected from: (a) two grit pots of HD cyclones (Stage 1); (b) two grit pots of SCC cyclones with 2.2 l/min (Stage 2); and (c) the grit pot of the SCC cyclone with 4.4 L/min (Stage 3). Material from each grit pot was collected separately by washing them with isopropyl alcohol and collecting the sample in conical centrifuge tubes (Fisher Scientific, Pittsburgh, PA). The PC filter from the cowl sampler (Stage 4) was directly placed on a SEM specimen aluminum mount with conductive carbon tape.

Each sample was coated with a thin layer of gold/palladium utilizing a sputter coater (SPI Supplies, West Chester, PA). A sequence of fields was selected at random locations and an image of each field acquired. The length and width of approximately 300 fibers for each sample were manually measured with ImageJ software (Schneider et al. 2012).

Aerodynamic diameter was calculated to compare particle size distributions between separated glass fiber aerosols and APS measurement. The aerodynamic diameter may be a good criterion to determine respirable fraction of the fibers. Fiber aerodynamic diameter depends on the fiber physical dimensions [diameter (d_f ) and length (L)] and on the orientation of the fiber in the measuring flow field (Cox 1970, Kulkarni et al. 2011). The aerodynamic diameter of the glass fiber aerosol was calculated with the fiber diameter and length measured by the FESEM using the following equations: (1)dae,‖=df{9ρf4ρ0[ln(2β)−0.807]}1/2 (2)dae,⊥=df{9ρf8ρ0[ln(2β)+0.193]}1/2, where d_ae, ∥ is aerodynamic diameter when the fiber is parallel to relative gas motion, d_ae,⊥ is aerodynamic diameter when the fiber is aligned perpendicular to relative gas motion, β=Ldf is the aspect ratio, and ρ_f and ρ₀ are the fiber (2250 Kg/m³) and unit densities, respectively. For random orientation, the aerodynamic diameter was calculated by: (3)dae=(dae,‖+2dae,⊥)3.

Phase contrast microscope images (400x magnification) of separated (selected AAC aerodynamic diameters of 1, 2, and 3 mm) and not separated glass fiber aerosols are shown in Figure 3. Most of the thick and long glass fibers in the nonseparated sample (Figure 3(d) were removed by the AAC as shown in Figure 3(a–c). Figure 4 shows (a) the normalized number-weighted distribution of glass fiber aerosols classified with the AAC in different selected particle sizes as a function of calculated aerodynamic diameter and (b) the cumulative number fraction of glass fibers. A significant difference between distributions was not found by the Mann-Whitney U test. Figure 5 shows (a) the normalized length-weighted distribution of glass fiber aerosols separated with the AAC as a function of fiber length and (b) the cumulative length fraction of glass fibers. The shortest group (AAC size selection 0.5 μm) showed significantly differences with longer groups (AAC size selection >1.25 μm) by the Mann-Whitney U test. Table 2 shows the aerodynamic diameter selected with the AAC, the number of the glass fibers analyzed with the FESEM, the geometric mean (GM) of the calculated aerodynamic diameters of the glass fibers using Equations (1)–(3), the geometric standard deviation (GSD, σ=d84%d50%) of the aerodynamic diameter, the GM of glass fiber length, the σ of glass fiber length, the GM of glass fiber width, and the σ of glass fiber width. Table 3 shows average count median aerodynamic diameter (CMAD, average of three particle distributions), average σ, average number concentration, and average mass concentration measured with the APS for each selected size with the AAC. The average mass concentration was calculated using particle number concentration and density of the glass fiber aerosols. Selected aerodynamic diameter with AAC, the geometric mean diameter from FESEM analysis, and the CMAD from the APS are not significantly different each other by paired t-test (p > 0.05). The GM aerodynamic diameter of the separated glass fiber aerosols increased with an increased aerodynamic diameter selected with AAC. A positive and weak correlation was found between length and width of the glass fibers separated with AAC (Pearson correlation coefficient, r =.285, p > 0.05) indicating that longer fibers are thicker fibers.

Figure 6 shows (a) the normalized number-weighted distribution of glass fiber aerosols classified with the multi-cyclone sampling array as a function of calculated aerodynamic diameter and (b) the cumulative number fraction of glass fibers. Stages 1 and 4 showed significantly difference in their number-weighted distributions as a function of calculated aerodynamic diameter. Figure 7 shows (a) the normalized length-weighted distribution of glass fiber aerosols classified with the multi-cyclone sampling array as a function of fiber length and (b) the cumulative length fraction of glass fibers. Stages 1 and 4 showed significant difference in their number-weighted distributions as a function of fiber length. Table 4 shows the number of glass fibers analyzed with the FESEM, GM of the calculated aerodynamic diameters of the glass fibers, σ of the aerodynamic diameter, GM of glass fiber length, σ of glass fiber length, GM of glass fiber width, and σ of glass fiber width. A positive and moderate correlation was found between length and width of the glass fibers separated with multi-cyclone sampling array (r = 0.583, p > 0.05).

There have been several methods developed to separate EMPs by length or other characteristic of concern. The classification of the glass fiber aerosols using dielectrophoretic mobility (FLC) showed reasonably monodisperse distributions (Baron et al. 1994; Deye et al. 1999; Ku et al. 2013). The GSD of the classified glass fiber aerosols with FLC ranged from 1.19 to 1.35 in four different classifications of glass fibers by length (different applied voltages, 1–4 kV). Ku et al. (2014) investigated the use of various nylon net filters (10, 20, and 60-μm mesh sizes) to efficiently separate fibers based on their length and found that single screens were not particularly effective in separating long fibers; however, an alternative configuration, especially a centrally blocked screen configuration, yielded samples substantially free of the fibers. The GSDs ranged from 1.89 to 2.99 in different fiber length distributions obtained with nylon net screens. Padmore et al. (2017) developed a glass fiber separation method using a combination of crushing (high and low pressure for short and long fibers, respectively), sonication, and sedimentation. The study showed that the fiber length distributions were confirmed to be log-normal, where the mean physical length was 7.0 μm and 39.3 mm for short and long fibers, respectively. The GSDs of fiber length distributions from the present study are slightly larger than the Ku et al. (2017) study; GSDs from classification with the AAC ranged from 1.49 to 1.69 (Table 2), GSDs from the multi-cyclone sampling array ranged from 1.60–1.82 (Table 4), and GSDs from the Ku et al. study ranged from 1.19 to 1.35 when the glass fibers were separated using the FLC. The shortest length groups separated with the AAC (aerodynamic size 0.5 μm) and multi-cyclone sampling array (Stage 4) showed significant difference (p > 0.05).

An additional experiment was conducted to remove short fibers (<10 μm) from a separated fiber group (AAC selection 3.0 mm) using an electrostatic precipitator (ESP, custom made), DC power supply (model 3015B and EMCO high voltage converter (model 4100 N)). An aerosol neutralizer (Model 3087, TSI Inc.) and the ESP were connected to the inlet and outlet of the AAC, respectively. APS was connected to the outlet of the ESP to monitor the particle size distribution. The applied voltage ranged from 200 to 1,000 volts to allow removal of short fibers from the AAC aerosols. However, it was observed that the ESP removed both small and long fibers at the same time.

The number and mass concentration estimation of glass fibers classified with the AAC is based on the APS measurement (Table 3). The collection time can be estimated with those mass concentrations. For example, the mass concentration of the separated glass fibers with the AAC 3.0-μm size selection was 702 μg/m³ and it would take approximately 17 hr to collect 702 μg. However, the inlet air of the APS was diluted in half with a HEPA filter (Figure 1). Thus, the mass concentration would be two times larger, which reduces the sampling time by half (8.5 hr) to collect 702 μg. The sampling flow and sheath flow rates of the AAC were 0.3 and 11.36 L/min (Table 1), respectively, and the mass concentration was diluted at the ratio of sampling flow to sheath flow rate. The sampling flow rate can be increased up to 1.5 L/min and sheath flow can be reduced, which might lower the resolution on the distribution.

The fiber dimensions (length and diameter) of interest for toxicological evaluation has been summarized. Early biological evidence showed that fiber length between 10 and 50 μm were related with the major asbestosis hazard and short fibers (shorter than 5 μm) were more effectively cleared from the lungs (Walton 1982). The NIOSH fiber counting “A” rule, which apples to fibers longer than 5 μm and aspect ratio greater than 3 originated from these findings. A report from the expert panel on health effects of asbestos and synthetic vitreous fibers: the influence of fiber length separated health effects on the short fibers (<5 μm) by types of health effects, i.e., cancer and noncancer effects (Eastern Research Group 2003). The report concluded that short fibers are not related to cancer in humans but may be pathogenic for pulmonary fibrosis. Later, Lippmann (2014) determined that the critical fiber diameters and lengths affecting disease pathogenesis and the critical fiber dimension were dependent on diseases: asbestosis (surface area if fibers with length > 2 μm, diameter > 0.15 μm), mesothelioma (number of fibers with length > 5 μm, diameter <0.1 μm), and lung cancer (number of fibers with length > 10 μm, diameter > 0.15 μm). Two different groups in different lengths for toxicological evaluations, longer than 5 mm and shorter than 5 μm, may be necessary but counting longer than 5 μm fiber was selected as the lower size limit for counting (a margin of safety) (Walton 1982; Langer et al. 1978). The respirable size fraction of fiber glass that was collected with the respirable size-selective sampler (Casella, Buffalo, NY) and horizontal elutriator (MRE type 113 A) was previously investigated using an APS without information of fiber diameter and length (Iles 1990). The present study showed that collection of EMPs in two different length groups might be achievable with the AAC for shorter than 10 μm and longer than 10 μm or shorter than 5 μm and longer than 5 μm. For example, more than 90% of separated glass fiber aerosols with AAC 0.5-μm size selection were shorter than 10 μm and about 20% of glass fiber aerosol with AAC 3.0-μm size selection were shorter than 10 μm within the limited fiber length measurements in the samples (approximately 300 fibers for each group). The smallest selected aerodynamic diameter with the AAC was 0.5 μm in the present study, but smaller aerodynamic diameter can be selected with the AAC to make a group of shorter fibers less than 10 μm (or 5 μm). Particle selection of the AAC ranges from 0.025 to 5 um. If median fiber length is shorter than glass fiber aerosols generated in the present study (about 18 μm), it is likely to collect a greater number of samples in a shorter period of time. Generally, size distribution of EMPs are log-normal indicating that shorter fibers are dominant (Chatfield 2018). Shorter fibers can be produced by more grinding of reference materials. However, it should be noted that material preparation and manipulation of the reference materials, including ball milling, might change crystallinity, and reduced crystallinity might reduce biological activity (Langer et al. 1978; Spurny et al. 1979). A combination method for classification of the EMPs is also achievable using multi-cyclone sampling array followed by the AAC; EMPs might be separated using the multi-cyclone (operating in high flow rates) sampling array to eliminate long EMPs, and separated EMPs may be separated again with the AAC to reduce a burden in the separation process. The present study utilized glass fiber aerosols as a surrogate of asbestos, but the classification of the regulated asbestos and its non-asbestiform analogs may be necessary. Prior to the toxicology studies, the classified EMPs materials should be fully characterized.