Waste process heat generally maintained the manufacturing area at acceptable temperatures during winter. A manually controlled gas-fired radiant heater was mounted above the processing area to provide supplemental heating and operated for a short duration on one occasion during measurements. Direct-fired gas heating system efficiencies may exceed 90% [American Conference of Governmental Industrial Hygienists (ACGIH), 2007]. However, they may also act as significant ultrafine particle sources (Peters et al., 2006, Heitbrink et al., 2007; Evans et al., 2008).

General ventilation in the production area was provided by three 1.5-kW roof-level exhaust fans but typically not in operation during colder months (when measurements described here were taken). No dedicated makeup air system was present to replace air removed by the roof exhaust fans or local exhaust systems. During warmer months, with roof fans in operation, unconditioned makeup air entered the building through open windows located to the west side of the facility. During cooler months, unconditioned makeup air from outdoors entered the facility through openings and gaps in the building structure. Air movement was not generally perceivable when standing in the manufacturing area since major local exhaust duct inlets were closer to roof rather than ground level. A battery-powered forklift truck was present but was not considered a significant particle source, whereas similar combustion-based vehicles (propane, gasoline, or diesel powered) may generate significant quantities of ultrafine particles (Kuhlbusch et al., 2004; Evans et al., 2008). In an earlier study (Heitbrink et al., 2007), battery-powered vehicles were in use without notable elevations in particle number concentrations.

Measurements of particle number concentration were performed with a CPC (CPC 3007; TSI Inc., Shoreview, MN, USA), with the working fluid regularly replenished. Response of the 3007 is linear to particle concentrations below ∼10⁵ particles cm³; however, counting coincidence errors are introduced above this value (Hämeri et al., 2002). Therefore, in environments where high particle number concentrations may be encountered, dilution is often required, and initial measurements made as part of an earlier walk-through survey of the facility obviated this need. Simple portable solutions are preferred in these applications; therefore, the dilutor consisted of a high efficiency particulate absorption (HEPA) filter cartridge (Whatman 6702-3600), with a single orifice (drill size #70, 0.028″ or 0.7112 mm diameter) drilled through the end cap. Similar devices have been used to perform high concentration measurements with the 3007 in automotive production environments (Peters et al., 2006; Heitbrink et al., 2007; Evans et al., 2008).

A dilution ratio of ∼90:1 was achieved and determined experimentally pre- and postsampling in a manner described by Peters et al. (2006). Briefly, this was calculated by successive addition and removal of the dilutor from the CPC inlet in a stable submicrometer particle concentration environment (i.e. distant from local particle sources such that large deviations in total particle counts over several minutes were not observed). The conference room (Fig. 1, Location D) was selected for this purpose. A consistent dilution ratio was obtained pre- and postsampling; however, it should be noted that the dilution ratio may be subject to variation if the filter media becomes heavily laden with particulate. Similar devices utilizing capillaries rather than orifices have also been reported. Although orifice size may be carefully controlled, with an extended length and residence time within the capillary, diffusional particle loss to the inner surfaces may contribute to a nonlinear relationship with concentration (Knibbs et al., 2007).

Real-time respirable mass estimates were obtained using a photometer (DustTrak Model 8520; TSI Inc.), with the 10-mm Dorr-Oliver cyclone providing a respirable size-selective inlet at 1.7 l min⁻¹ sampling flow.

Active surface area measurements were provided by a diffusion charging (DC)-based instrument (DC 2000 CE; EcoChem Analytics, Murrieta, CA, USA). The DC response has been compared with that of other methods (Ku and Maynard, 2005). The DC has provided active surface area measurements within automotive engine manufacturing facilities (Heitbrink et al., 2009) and for monitoring diesel engine exhaust exposure (Ramachandran et al., 2005).

A photoelectric aerosol sensor or PAS (PAS 2000 CE; EcoChem Analytics) was used to provide the photoelectric response of particles within the facility in real time. An overview of the PAS is provided by Burtscher (1992) and Siegmann et al. (1999). Briefly, the PAS responds to aerosols from incomplete combustion, for example, and, with a low work function, is particularly sensitive to particle-adsorbed polycyclic aromatic hydrocarbons or PAHs. At a higher work function, the signal is mostly determined by elemental carbon. The presence of PAHs on preliminary air samples, taken as part of an earlier walk-through survey, suggested that the PAS may be a useful tool in this facility. In a similar application to our own, the PAS was used to indicate the presence of soot (containing fullerenes) during the harvesting of product from electrical arc reactors in a small nanomaterial facility (Yeganeh et al., 2008). Both the DC and the PAS operated at 1.0 l min⁻¹. The operating principles of both instruments and their combined use are discussed by Ott and Siegmann (2006).

Air quality measurements were conducted with a Q-Trak Plus (Model 8554; TSI Inc.) and included relative humidity, temperature, carbon monoxide (CO), and carbon dioxide (CO₂). CO and CO₂ measurements may often be used to indicate the presence of combustion products in the workplace, but by association may also indicate the presence of combustion-derived ultrafines. For example, alloy pouring and exhaust from a propane-fueled vehicle resulted in elevations in both particle number and CO concentrations within a foundry (Evans et al., 2008).

Particle size distributions from 7 nm to 10 μm aerodynamic diameter were obtained with an ELPI (Dekati Ltd, Tampere, Finland) operating at a nominal 10 l min⁻¹. The ELPI is a real-time cascade impactor utilizing DC and electrometer detection. An operational overview is discussed in Baltensperger et al. (2001). An effective aerosol density of 1.0 g cm⁻³ was assumed in estimating aerosol charging efficiencies in addition to correcting for both diffusional and space charge losses. In a similar manner, the ELPI was previously used in automotive manufacturing environments to characterize the particle size profiles of identified contaminant sources (Heitbrink et al., 2007; Evans et al., 2008). Oiled sintered metal impaction substrates were utilized to minimize particle bounce and reentrainment. In this instance, the external vacuum pump supplied with the ELPI (Leybold-Sogevac SV25-10991) was replaced with a dry scroll pump (SH-110, Varian Inc. Lexington MA) to reduce weight, noise, and power requirements.

Mobility equivalent particle size distributions between 5 and 500 nm were obtained with an FPSS (DMS50; Cambustion Ltd, Cambridge, UK) operating at a nominal 6.5 l min⁻¹. The DMS50 FPSS has a built-in dilution device, which may be user actuated during use but was not required during this study.

On board data logging capabilities were utilized for the CPC, DustTrak, DC, PAS, and Q-Trak Plus, and the shortest log time intervals were selected: 1 s for the CPC and DustTrak, 10 s for the DC and PAS, and 1 s for the Q-Trak Plus. A single laptop computer with software (ELPI VI 4.0 and DMS500 UI v2-08) was used for control and data acquisition from both the ELPI and the FPSS, respectively. Raw size distribution data were collected every 1 and 2 s from the ELPI and FPSS, respectively. All instruments were time synchronized with the laptop prior to the commencement of sampling.

A custom mobile particle measurement platform (Fig. 3) was used to house and power particle monitoring equipment and ancillaries at given locations. Being vertically arranged, the platform provided a smaller footprint, greater maneuverability, and carrying capacity over previous commercially available carts (e.g. Peters et al., 2006). Mobile power was provided through deep-cycle marine batteries and an inverter, providing several hours of uninterrupted instrument operation. With a reduced platform footprint, batteries and pump were located at the base to ensure a sufficiently low center of gravity.

Spatial variation in particle concentration over short distances, and, in particular, close to sources, can be considerable in workplace environments (Brouwer et al., 2004; Peters et al., 2006; Heitbrink et al., 2007; Evans et al., 2008). In such situations, instruments placed in general proximity to each other may experience substantial discrepancies in sampled concentrations. For example, large concentration differences were encountered by investigators making comparative measurements in carbon black manufacturing facilities (Kuhlbusch et al., 2004), and close attention to these differences is required if quantitative comparisons between instruments and particle metrics are to be reliably conducted in the workplace.

To ensure that the same aerosol was presented to the instruments, a sampling manifold was installed to the underside of the top surface of the platform. This single line supplied sampled air to instruments responding primarily to submicrometer particle sizes, including the FPSS, CPC (with dilutor), DC, and PAS. The manifold consisted of a thin-walled stainless steel tube with a sharp beveled entry. In addition, stainless steel tube fittings and a flow splitter (3708; TSI Inc.) were utilized, with each instrument connected through short lengths of flexible conductive silicone tube.

Since inertial particle loss required consideration when sampling with the ELPI (particles up to 10 μm aerodynamic diameter), it was equipped with its own thin-walled inlet with sharp beveled entry and gentle 90° radius. The entry was positioned parallel to and within 50 mm from the manifold. In the worst case, at a diameter of 10 μm, particle losses were calculated at <7% (Brockmann, 2001). The DustTrak utilized a size-selective inlet (Dorr-Oliver cyclone) positioned within 150 mm of both manifold and ELPI inlets. Considered as area or static sampling, at ∼1.5 m from ground level, inlets were at a height representative of an observer’s ‘personal breathing zone’ (Leidel et al., 1977; Vincent, 1995, 2007).

Sampling locations are annotated in Fig. 1. Simultaneous measurements on 14 December 2007 were first conducted in the production control room (A) from ∼11:25 to 12:20, then adjacent to Reactor A in the fiber production area (B) from ∼12:20 to 13:10, subsequently in the processing area (C) from ∼13:10 to 16:40, and finally in the conference room (D) from ∼16:40 until sampling concluded. Movement from one location to the next took <1 min.

Since particle size distribution and composition of airborne material may influence photometer response (Gebhart, 2001), measurements conducted with the photometer are considered useful for monitoring the ‘relative changes’ in mass concentration in a workplace rather than considered ‘absolute’ values. By default, the photometer comes complete with a factory calibration using the respirable fraction of standard ISO 12103-1, A1 test dust, formerly Arizona Road Dust (TSI, 2006). By several successive redispersions of milligram quantities of powdered CNFs within a small laboratory chamber, photometer response was compared against respirable mass concentration (GK2.69 cyclone; BGI Inc., Waltham, MA, USA), with CNF mass determined by gravimetric analysis [NIOSH 0600; National Institute for Occupational Safety and Health (NIOSH), 1998]. Photometer response was found to be proportional to the gravimetric mass concentration for respirable CNFs in the range 0–8 mg m⁻³. However, some variation was noted and likely due to a combination of the efficacy of powder dispersion (i.e. some variation in particle size distribution between individual tests), the relative performance of the two respirable cyclones, and nonuniformity in concentration initially introduced into the chamber through dispersion. Linear regression provided a one-to-one relationship (n = 9, gradient of 0.99), with an R² value of 0.79, indicating that the factory calibration (performed shortly prior to this study) provided a reasonable estimate of respirable CNF mass concentration. No further calibration or correction factor was required for data collected in the facility.

Particle and gas emissions measured at various locations (A through D) are presented as time series in Fig. 4. Notable events (I through IV) in the processing area (Location C) are marked as shaded areas and merit further examination and discussion with respect to corresponding instrument responses.

With the notable exception for CO₂, the conference room was the cleanest of indoor locations monitored for contaminants. Rapid elevation and greatest concentrations of CO₂ were due to a number of individuals cleaning/packing equipment at the end of the day. The control room, where sampling commenced, was generally elevated in contaminant concentrations relative to the conference room but less contaminated than locations in the manufacturing area. Of the production and processing areas, particle number, transient respirable mass concentration elevations, and active surface area appeared greatest in the processing area. In contrast, general (nontransient) photoelectric response and CO concentrations were generally greatest in the fiber production area.

Both the dryer and the furnace/stripper (thermal treatment) were in close proximity, without physical barriers between them and operations ran concurrently over a period of several hours. However, the dominant source of ultrafine particles appeared to emanate from the furnace/stripper during high-temperature thermal treatment. Because ultrafine particles emitted from the dryer likely contained PAHs (Fig. 4, Events II and III), the absence of general elevation in photoelectric response in the processing area (similar to photoelectric response in the control room) suggests that for the most part, general elevations in ultrafines were not likely from the dryer.

Only on the noted occasions that the dryer was opened, did simultaneous and transient elevations in photoelectric response and active surface area concentrations result in addition to modest increases in particle number. Furthermore, the last batch of CNFs passed through the thermal treatment at ∼14:45, and the ultrafine particle concentrations in the processing area were subsequently observed to decline from a general maximum observed during that period. General elevation in photoelectric response was additionally observed in the fiber production area, close to the reactors. Elevation in photoelectric response was likely due to particles with a PAH component in this locale, emitted directly from fiber synthesis operations.

It should be further noted that observed elevations in ultrafine particle concentration in the processing area were not from CNFs but rather from by-products formed through high-temperature thermal processing of the CNFs (Birch et al., in preparation). Because measures were in effect to exclude oxygen and maintain an inert atmosphere during high-temperature thermal treatment, ultrafine particles were not combustion derived. Relatively consistent baseline CO₂ concentrations observed in both the fiber production and the processing areas further supports this argument.

From the perspective of direct monitoring of airborne CNF release into the workplace, respirable particle mass estimated by the photometer appeared to be the most useful and practical metric. Elevations or peaks in respirable mass concentration closely coincided with the transfer or handling of material in the open workplace. The strongest of those elevations were associated with the dumping of product after drying (Fig. 4, Event III) and the manual change-out and closing of bags of final treated CNF product (Fig. 4, Event I, and see photograph in Fig. 5a). Note that dark visible airborne clouds of CNFs were also observed on these occasions indicating the presence of large agglomerated particles. Although data presented in this article cover only a single shift, further stationary direct-reading monitoring on subsequent days indicated that CNF handling activities were directly responsible for transient mass concentration increases in the workplace. In addition, through full/partial shift integrated personal and area monitoring, sufficient respirable CNF mass was collected for quantitative and selective personal exposure determinations (Birch et al., in preparation). Direct-reading measurements conducted and reported in this article provide insights into how these exposures occurred, yet worker exposures are best quantified through more selective approaches in the breathing zone. Furthermore, microscopy analysis through size-selective area sampling (Birch et al., in preparation) indicated that large fiber bundles were present, further supporting findings here that particle mass may be a more practical metric for monitoring CNF emissions and exposure in this facility.

Particle number concentration and active surface area concentrations were generally greatest in the processing area. In general, active surface area concentration follows particle number more closely than particle mass (Heitbrink et al., 2009) because the diffusion charger responds most strongly to sub-100-nm particles (Ku and Maynard, 2005).

In summary, integrated particle number and active surface area concentrations (determined by a CPC and DC in this study) were not particularly useful in assessing the contribution of emissions from CNFs in this workplace since these measurements were dominated by ultrafine particle emissions rather than by CNFs. A less expensive and simpler instrument such as a photometer, providing estimates of particle mass, may be more appropriate in this application. Particle number concentrations spanned from 9.0 × 10⁴ to 1.15 × 10⁶ cm⁻³ in the manufacturing areas. In a similar manner, active surface area concentrations due mainly to ultrafine particle emissions ranged from 430 to 1440 μm² cm⁻³. When assessing emissions and potential worker exposure, multiple constituents of the mixed exposure should perhaps be considered, and these metrics may warrant further merit in this instance.

A current emphasis appears to be placed on direct-reading particle counting within the industrial hygiene community for screening nanotechnology workplaces and emission/exposure assessment (e.g. Methner et al., 2010) and/or assessing the effectiveness of engineering controls (e.g. Old and Methner, 2008). This approach, although straightforward, may be a rather too simplistic for universal application. Potential errors are introduced when transient changes in particle background, simultaneous emission of ultrafine particles, or the influence from other particle sources are present. These confounding factors may not be recognized at the time, and in such instances, it may be difficult to properly proportion the nanomaterial contribution from a particular process or task. The suggested approach of deriving a mean average background concentration from several spot measurements and subtracting from the ‘process’ (e.g. Methner et al., 2010) does not adequately address any of these issues.

Further potential errors relate to an expectation that all particles observed as a transient concentration elevation in response to a particular event or task are due to engineered nanomaterials. An example from the present study is the dryer dump event earlier discussed. A large transient particle number concentration increase was observed with this event as CNFs were dumped from the dryer to a collection vessel. The transient increase in particle number was not due to CNFs, but rather, condensable ultrafine particles emitted as the dryer apparatus was opened and fugitive emission controls were temporally breached. Using only particle counting as a guide, this critical distinction between particle sources would not have been determined using current approaches (e.g. Methner et al., 2010). Opening of the dryer for manual redistribution of partially dried CNFs also generated ultrafine particle emissions, yet little if any CNFs. A good understanding of workplace emissions is an essential prerequisite to reducing exposure because recommended approaches to mitigate these related but very different particle emissions may vary considerably (see Appendix 1 for discussion of exposure controls). Careful interpretation is required when relying on simple measurements in the assessment of complex systems.

Results for CO monitoring are also presented as a time series in Fig. 4. Elevated CO concentrations in this facility were found to be of particular concern but were not the primary focus of this study; therefore, further discussion is provided in the Appendix 1. The peak concentration observed was 265 ppm in the production area, with mean average concentrations of 44, 130, 99, and 47 ppm observed in the control room, production area, processing area, and conference room, respectively, over the periods sampled in these locations. Thermal decomposition of the carbonyl catalyst precursor during fiber synthesis was the likely source.

Measurements conducted with the ELPI (aerodynamic diameter) and FPSS (mobility equivalent diameter) provided number-weighted particle size distributions over the entire sampling period, giving a combined size range spanning 5 nm to 10 μm. By careful observation and interpretation of data presented in Fig. 4, as earlier described, particle sources and plumes were identified. Size distributions were obtained as mean average measurements over periods present in these locations and are presented in Fig. 6. In the case of thermal treatment (at the furnace/stripper), averaging was taken over those periods not influenced by Events I through IV since elevations observed during these periods were from other sources.

In contrast, since relatively transient, event timing was used to identify pertinent plume data in the dryer plume and gas burner plume cases, and the results are presented over shorter duration (several minutes) in Fig. 6. Concentration range bars were not included for clarity, but concentrations varied considerably in the case of the processing area, as evidenced by the variability in the CPC-derived particle number concentrations presented in Fig. 4, but relatively stable, for instance, for the outdoor background measurements.

Dominant sub-10-nm modes were noted for the thermal treatment and production emissions, suggesting freshly nucleated particles. Dryer plumes were dominated by the thermal treatment emissions when present, so they exhibited a similar size profile. However, small concentration increases between 50 and 200 nm are just perceivable, consistent with condensation particle emissions from the dryer. Gas burner emissions exhibited an additional mode between 15 and 25 nm and concentration elevations in the 20–80 nm size range above those seen previously for thermal treatment or dryer emissions. Similarly sized modes were previously observed (Heitbrink et al., 2007). A distribution from outdoor background (semirural location and upwind of the facility) provided the lowest concentrations observed in the study. FPSS data above ∼100 nm (Fig. 6) were not plotted because data were influenced by an erroneous electrometer offset; however, the ELPI provided sufficient data to adequately cover this size range.

Since other particle sources in the facility dominated on a particle number basis, extracting specific size information for CNFs (by number) was challenging. However, careful time alignment with observations assisted in identifying events when CNF plumes were known to be present and respirable mass estimates, provided by the photometer, were particularly useful in identifying these periods. By careful selection of appropriate baseline distributions, net changes in particle size distributions from both the FPSS and the ELPI were derived and are presented in Fig. 7. Subtraction of the CNF plumes from baselines negated the effect of the erroneous electrometer offset for the FPSS data >100 nm since this was consistent throughout. Baseline distributions were derived as a mean average during a relatively stable period (typically a few minutes) just prior to known CNF emissions. Raw CNF distributions were averaged over a period of several minutes when the plume was known to be present.

A larger quantity of CNFs was released on a respirable mass basis (see Fig. 4) for the dryer dump event over the bag change event. Respectively, a greater elevation in particle number concentrations by size was correspondingly observed in the size distribution data. During both events, CNF plumes exhibited dominant modes weighted by number between 200 and 250 nm as both mobility equivalent (FPSS) and aerodynamic diameters (ELPI). ELPI measurements exhibited shoulders in the overall distribution at this size but corresponded closely to measurements from the FPSS, with the latter providing greater sizing resolution. The close agreement between these two different, but complementary, sizing techniques provides a greater level of confidence in the presented CNF size data. A secondary mode (observed as shoulders) between ∼1 and 3 μm aerodynamic diameter was also observed in the ELPI size distributions weighted by number for both CNF plumes. The photometer (response proportional to mass) likely responded most strongly to particles in this size range.

Because both aerodynamic and mobility equivalent diameters appear very similar for the 200- to 250-nm CNF mode, effective particle densities at or close to unity are implied. For a theoretical discussion on the interrelationships between mobility equivalent, aerodynamic diameters, and effective densities, see DeCarlo et al. (2004). In a prior study (Ku et al., 2006), CNF particles generated through laboratory agitation of bulk powder provided effective densities between 0.85 and 0.75 g cm⁻³ at 200 and 250 nm, respectively. Our findings are generally consistent with these earlier observations. A decrease in the effective density of particles with larger diameters is anticipated (greater divergence in mobility and aerodynamic diameters), as more complex and open structures are observed (Ku et al., 2006). CNF particle structures in the 200–250 nm size range consisted mostly of clusters of fibers with some single fibers also observed (Ku et al., 2006). Similar structures were observed from CNF plumes in this workplace (see Birch et al., in preparation).

Since integrated particle number concentration measurements provided by the CPC were dominated by other sources (Fig. 4), it is impossible to obtain the contribution from CNFs solely through these measurements. However, by utilizing size distribution data, one may estimate particle number concentration increases for the two notable events that precipitated major releases of airborne CNFs. FPSS particle size distributions for CNF plumes were utilized (Fig. 7) since these data possess finer size resolution than corresponding ELPI data. The bag change event resulted in an estimated 230 cm⁻³ transient elevation in particle number concentration from CNFs and the dryer dump event an estimated 3130 cm⁻³ elevation. These values need to be compared to a mean average particle number concentration of 6.53 × 10⁵ cm⁻³ (obtained with the CPC) observed in the processing area over the period measurements that were conducted in this location. On a particle number basis, CNF plumes made up ∼0.035 and 0.48%, respectively, of mean average particle number concentrations, illustrating the challenges and potential inaccuracies in making such measurements. For example, had the transient particle number concentration increase of between 3.0 and 3.5 × 10⁵ cm⁻³ above baseline associated with the dryer dump event been erroneously and solely attributed to CNFs, a potential two orders of magnitude error in estimated CNF number concentration would have resulted.