The high-flow NRD sampler consists of four principal design elements as shown in Figure 1a: inlet; impactor stage; diffusion stage; and final filter. The dimensions of the high-flow NRD are 150 mm in length by 42 mm in width, and its weight is 165 g. The inlet and housing for the impaction and diffusion stages were machined from an aluminum alloy. The holder for the final filter is a standard 37-mm filter cassette holder. When air is drawn through the sampler, particles larger than 300 nm are removed due to impaction on a greased collection plate. The remaining airborne particles are passed into a polyurethane foam, which collects particles via diffusion with collection efficiency following the nano-particulate matter (NPM) criterion (Cena, Anthony et al. 2011). This criterion specifies gradually decreasing collection efficiencies with increasing particle size, and has a d₅₀ of 40 nm. The final filter collects the remaining airborne particles with a target collection efficiency of 100%.

Collection efficiencies by size of the impactor stage, diffusion stage, and final filter were obtained individually using the experimental setup shown in Figure 2. All collection efficiency tests were conducted with a flow rate of 10 L/min. A 0.9% salt solution (NaCl, 2F7123, Baxter Healthcare Co., Deerfield, IL, USA) aerosolized by a vibrating mesh nebulizer (Aeroneb Solo System, Aerogen, Ireland) was used to measure collection efficiencies in the size range from 20 nm to 700 nm. Dry and clean air regulated with a mass flow controller (MFC; MPC20, Porter Instrument, Hatfield, PA, USA) was delivered to a vibrating mesh nebulizer. Generated aerosol was passed through a silica diffusion dryer to ensure solid crystal salt particles and an ⁸⁵Kr charge neutralizer (3054, TSI, Shoreview, MN, USA) to render the charge distribution on the particles to a Boltzmann distribution. A 200-L coagulation chamber was used to increase particle size and to obtain a stable size distribution.

Particle number concentration by electrical mobility diameter was measured using scanning mobility particle sizer (SMPS; 3938, TSI, Shoreview, MN). The SMPS consisted of an electrostatic classifier (Model 3082), an advanced aerosol neutralizer (Model 3088), a long differential mobility analyzer (DMA, Model 3081A), and a condensation particle counter (CPC, Model 3788). The SMPS was set to a sampling flow rate of 0.3 L/min and a sheath flow rate of 3 L/min. Salt aerosol passed through an empty housing (bypass) and through a tested element (impactor, foam, or filter) at least three times each. For each design element and SMPS particle size bin, collection efficiency (η_c) was calculated as follows: Equation 2.1 ηc=1−CelementCbypass,

where C_element is an average number concentration exiting the impactor, foam, or filter; and C_bypass is an average number concentration exiting the empty housing. Collection efficiency of particles larger than 500 nm were measured using aerosolized glass microspheres (Spheriglass, 5000 A, Lot No: 080536476K, Potters Industries Inc., NJ, USA) by an Aerodynamic Particle Sizer (APS; 3321, TSI, Shoreview, MN). A fluidized bed generator (3400A, TSI Inc., Shoreview, MN, USA) was used to aerosolize the glass microspheres which were subsequently mixed with clean air (250 L/min) in a mixing chamber (0.64 m × 0.64 m × 0.66 m) and sampled in a separate sampling area (0.53 m × 0.64 m × 0.66 m). The particle size channels from the APS were converted from aerodynamic diameter to mobility diameter following Peters et al. (1993) using a shape factor of 1 and density of 2500 kg/m³.

The collection efficiency of the diffusion stage was adjusted for the presence of the impactor operated with Nozzle Section 1. The adjustment was made because of excess pressure demands on the SMPS system when both impactor and foam were in series. The foam efficiency adjusted for the presence of the impactor was defined as ηf/i: Equation 2.2 ηf/i=ηfoam×(1−ηimpactor).

The experimentally adjusted collection efficiency curves for the diffusion stage were compared to the NPM curve by applying the coefficient of determination, Rf/i2, calculated as: Equation 2.3 Rf/i2=1−Σ(ηf/i−NPM)2Σ(ηf/i−ηf/i,avg)2,

where ηf/i,avg is the average foam efficiency adjusted for the presence of the impactor.

The experimental pressure drop across the impaction stage, diffusion stage, and final filter was measured using a pressure gauge (Wallace and Tiernan SN: 9481B, Lawrenceville, GA) at the flow rate of 10 L/min. Each design element’s pressure drop was measured individually and then measured for a complete sampler.

The minimum sampling time was based the metal mass that would be collected on the foam substrate after sampling at concentrations of 1/10^th of the Occupational Exposure Limit (OEL) for a particular metal element. Metal content of the foam was measured using Inductively Coupled Plasma (ICP)-Mass Spectrometry (MS) (iCAP RQ ICP-MS, Thermo Fisher Scientific, MA, USA) analysis after microwave digestion of the foam using a Microwave Reaction System (MARS 6, CEM Corporation, Matthews, NC) following the NIOSH method 7302 (Ashley 2016). The limits of quantification (LOQs) of cadmium (Cd), chromium (Cr), copper (Cu), iron (Fe), manganese (Mn), nickel (Ni), titanium (Ti), and zinc (Zn) were calculated as the mean blank concentration plus 10 times the standard deviation of the blanks (Shrivastava and Gupta 2011). The LOQ represents the lowest amount of mass required to accurately distinguish sampled metals from background content of the media. An estimated minimum sampling time (t_min) needed when sampling at the flow rate Q = 10 L/min and at a concentration C equal to 1/10^th of the Recommended Exposure Limit (REL) for titanium dioxide and 1/10^th of the Threshold Limit Values (TLVs) for all other metals was calculated using the following equation: Equation 2.4 tmin=LOQC*Q.

Collection efficiencies by particle mobility diameter for three impactor nozzle sections are shown in Figure 3. The collection efficiency curves of all nozzle sections exhibited a similar behavior with a sharp increase from 0.01 (± 0.03) at 100 nm to 0.96 (± 0.03) for particles at 542 nm. Nozzle Section 1 met the target d₅₀ at 305 nm whereas Nozzle Section 2 had d₅₀ at 328 nm and Nozzle Section 3 had d₅₀ at 406 nm (Figure 3 and Table 1). Nozzle Section 1 was selected to be used in the final design of the high-flow NRD sampler due to its d₅₀ being the closest to the target value of 300 nm. Collection efficiency for larger diameters (up to 2000 nm) was measured using the APS only for Nozzle Section 1 and was close to 100% (Figure 3). These data show that particle bounce from the impaction plate is minimal. The collection efficiency of the APS was higher than that measured by the SMPS for particles from 310 nm to 400 nm. This discrepancy is thought to stem from differences in the density and shape factor of the salt and glass bead aerosol.

Larger jet diameters such as for Nozzle Section 3 (W_nozzle = 0.626 ± 0.008 mm) had higher precision, wheras smaller nozzle diameter as for Nozzle Section 1 (W_nozzle = 0.566 ± 0.144 mm) were difficult to make with high precision (Table 1). The fact that the sharpness of the collection efficiency curve was similar for Nozzle Section 1 and Nozzle Section 2 suggests that this change in precision had little influence on performance. Marple and Willeke (1976) recommended that the S/W_nozzle ratio should be greater than 1 for constant efficiency curves which was assured for all designed nozzle sections, ranging from 1.4 for Nozzle Section 2 to 1.73 for Nozzle Section 1. T/W_nozzle ratio greater than 1 was also obtained for all nozzle sections which should ensure uniform flow through the nozzles. Re number over 2000 was calculated for Nozzle Sections 1 and 2. The jet velocity, V was the highest for Nozzle Section 1 at 5447 cm/s and was greatly reduced to 2641 cm/s for the larger jet diameter of Nozzle Section 3.

The high-flow NRD impactor was designed with slightly different physical characteristics compared to the low-flow NRD sampler but both types of impactors met Marple and Willeke (1976) recommendations. For example, the S/W_nozzle ratio for the impactor of the 2.5 L/min NRD sampler was 1.9 whereas this ratio for Nozzle Section 1 used in high-flow NRD sampler was 1.73, and the Re number of the 2.5 L/min NRD impactor was 2212 whereas the Re of high-flow NRD impactor was 2057 (Cena, Anthony et al. 2011). The high-flow NRD sampler design also included a tapered inlet whereas the 2.5 L/min NRD sampler did not (Cena, Anthony et al. 2011). In comparison, PENS sampler’s S/W_nozzle ratio of the impactor with their target d₅₀ was 13.8 but the Re was below the recommend range (500-3000) at 385 (Tsai, Liu et al. 2012).

The impactor-adjusted collection efficiencies by size for the diffusion stages with foam depths of 5, 7, and 10 cm are shown in Figure 4. The R² values calculated using Equation 2.3 for each of the foam depths were 0.92 for 10 cm, 0.96 for 7 cm, and 0.90 for 5 cm. The corresponding d₅₀ for the foam depths of 10 cm, 7 cm, and 5 cm were 50 nm, 43 nm, and 32 nm, respectively. Collection efficiency of the diffusion stage adjusted by the impactor with Nozzle Section 1 was the closest match to the NPM criterion. As expected, the collection efficiency of the foam decreased with increasing particle size due to diffusion. For particles from 21 nm to 40 nm in size, the collection efficiency of the foam was 10% greater than the NPM criterion for foam depths of 7 cm and 10 cm. For particles around 100 nm in size the collection efficiency was approximately 0.20 and decreased to 0.05 for particles around 300 nm. Due to the impactor adjustment, the collection efficiency was the lowest for particles larger than 300 nm. The foam depth of 7 cm in the diffusion stage had the highest R² value and provided d₅₀ closest to the 40 nm target, hence it was selected to be used in the final design of the high-flow NRD sampler.

Achieving lower sampling times to meet metal LOQs for the NRD diffusion stage was the primary motivation for the development of the high-flow NRD sampler. The minimum sampling times required to reach LOQs for metals tested in this work at 1/10^th exposure limit concentrations for the high-flow sampler were significantly reduced (almost in half for many metals and almost 4-fold for Ti) compared to the 2.5 L/min sampler (Table 2). These results were expected. The high-flow NRD sampler has four times higher flow rate in comparison to the 2.5 L/min sampler and can collect more sample in the same sampling time.

However, the LOQs for the high-flow NRD sampler were substantially larger than for the 2.5 L/min sampler due to the larger volume of foam required to achieve the NPM target collection efficiency (Table 2). This situation is unlike typical air sampling in which the same filter is used with different airflows. In filter sampling, the LOQ is constant because the filter does not change and sampling time is inversely related to flow rate alone. Future work should examine ways to reduce the LOQ of the media used in the diffusion stage as such a reduction would also shorten sampling times. The foam used in this work may be prewashed to remove impurities. Alternatively, different substrates could be used as diffusion stages.

The collection efficiency by size was near the target of 100% for the final filter supported by the modified pad (Figure 5). A slight dip in collection efficiency was observed for particles between 31 nm and 224 nm with the lowest collection efficiency at 0.992 ± 0.001 for 88 nm particles. The MCE filter had a substantial pressure drop of 3.0 kPa (12 in H₂O). The support pad pressure drop was 2.7 kPa (11 in H₂O) and it was decreased to 1.4 kPa (5.5 in H₂O) with the modification described in Methods section, while the pressure drop for a system of MCE filter and the pad decreased from 5.5 kPa (22 in H₂O) to 4.4 kPa (17.5 in H₂O) after this modification.

The final design for high-flow NRD sampler consists of Nozzle Section 1 with d₅₀ of 305 nm, 7-cm long polyurethane foam with d₅₀ of 43 nm, and 5-μm pore size MCE filter with near 100% collection efficiency. The pressure drops for each component and the assembled sampler are shown in Table 3. The impactor stage had a pressure drop of 4.2 kPa (16.9 in H₂O). The diffusion stage had the lowest pressure drop of all NRD stages at 1.7 kPa (6.8 in H₂O). The pressure drop across the final filter with the support pad was 4.4 kPa (17.5 in H₂O). With all components, the complete sampler had a pressure drop of 8.5 kPa (34 in H₂O), which is substantially less than the sum of the pressure drop for individual components. This difference is attributed to the pressure drop of the housing, which was approximately 1 kPa. The pressure drop of the complete sampler cannot be handled by commercial personal pumps (e.g., Gilian 12 from Sensidyne, St Petersburg, FL, USA). Thus, the high-flow NRD sampler can only be used with such personal pumps when operated without the final filter. Other options to decrease NRD sampler pressure drop are to use a larger filter (e.g. 47 mm in diameter) or to use a filter with larger pore size (e.g. 8 μm), which may reduce the pressure drop but may also lower collection efficiency.

A limitation of this work was that we did not test the impactor under non-ideal conditions. Specifically we did not evaluate clogging of the impactor nozzles with highly fractal or large diameter particles. We also did not determine the effect of particle loading on the impactor cutoff curve. We did not characterize the particle loss in the dome-shaped inlet, which could be the subject of future work. Clogging and particle loading may be an issue under field conditions due to the small diameter nozzles and the absence of a size-selective inlet. Such an inlet would remove larger particles from the airstream, thereby protecting the nozzles from clogging and the plates from over-loading.