Indoor and outdoor air pollution are major contributors to human disease, disability, and premature death globally. Household air pollution which, in developing countries, results primarily from the incomplete combustion of primitive fuels for cooking and heating (e.g., wood, charcoal, etc.), is considered the 6^th leading risk factor for disease and death on the planet (IHME, 2015). Outdoor air pollution from particulate matter, which has numerous anthropogenic and biogenic sources, is considered the 11^th leading risk factor (IHME, 2015).

Despite these alarming risks, our understanding of human exposure to air pollution (whether indoors or outdoors) is limited. Although regulatory agencies such as the U.S. EPA maintain national air quality monitoring networks, the monitors that constitute these networks tend to be relatively sparse, costly to maintain, and report only outdoor pollution levels. Furthermore, data from such monitoring networks are only modestly correlated with an individual’s daily exposure to air pollution (Wilson and Brauer, 2006, Meng et al., 2005), except in cases where the study population is highly sedentary (Janssen et al., 2000). People spend most of their lives indoors, moving from one microenvironment to the next (e.g., at home, at work, or in transit). For these reasons, assessment of personal exposure remains the standard for determining individual risk. Studies that have examined personal exposure to air pollution, however, have consistently demonstrated lognormal variations in exposure that span both space and time (Good et al., 2015, Wallace et al., 2006, Rappaport and Kupper, 2008). Characterizing lognormal exposure distributions (for exposure assessment or exposure modeling) requires studies with relatively large sample sizes; however, most studies that employ personal sampling are limited to modest sample sizes because of limitations in the technologies used to assess personal exposure.

A major constraint on our ability to assess personal air pollution exposure is the cost and physical burden of the monitors themselves. For monitoring exposure to particulate matter (PM), personal air samplers typically consist of a battery-powered diaphragm pump connected by tubing to a size-selective inlet (e.g., a cyclone or impactor) to measure inhalable, respirable, or PM_2.5 size fractions of particulate matter within the wearer’s breathing zone. Such personal air samplers are expensive (typically costing >$1500 each), relatively heavy (>0.5kg in total), and noisy (emitting >60 dB from the pump). The physical burden posed by these monitors (noise, visual aesthetic, and weight) make them difficult to wear for extended periods. Further, the diaphragm pumps must be periodically checked for flow accuracy and the tubing connections often disconnect or become pinched if the wearer is physically active. For these reasons, studies of personal PM exposure often suffer from small sample size and data loss due to poor user compliance and instrument reliability.

Historically, the development of PM exposure monitors was driven by the need to assess occupational intake for aerosol hazards in the dusty trades such as mining, construction, manufacturing, and agriculture (Vincent, 1999, Lioy, 2010). In those workplaces, the weight and noise of the monitors was less of a hindrance. For personal monitoring among the general population (or outside of heavy industry), however, a need exists for technology that overcomes these limitations. The objective of this work was to develop a wearable air pollution monitor to address limitations of the current state-of-the art (cost, noise, weight) without sacrificing precision, accuracy, and reliability. We leveraged recent advances in consumer electronics, open-source software platforms, and additive manufacturing to iterate upon several versions of a prototype wearable PM sampler. The resultant device, an ultrasonic personal aerosol sampler (UPAS), is compact, lightweight, and virtually silent when running. The UPAS was tested for pump and battery performance, flow accuracy, and size-selective sampling efficiency. Device performance was also evaluated against an EPA-certified reference method for PM_2.5 sampling through laboratory tests (and also compared to an off-the-shelf commercial sampling device).

Selected sensors and components for the final UPAS design (Figure 1C) are provided in Table 1; a time-series plot of data collected by these sensors is shown in the Supplementary Information. The retail cost of these sensors, when purchased in single quantity amounts, totals approximately $150. Shown in Figure 2 are data depicting the performance of the UPAS pump at three different (arbitrary) power levels for standard temperature and pressure (STP) conditions. The shaded area on the figure represents the operating envelope of the pump (flow vs. pressure drop across all possible power levels). The operating range of the pump spans 4kPa of water static pressure head and upwards of 3 L/min of flow at the highest power setting. At 1.0 and 2.0 L/min of flow the pump is capable of drawing air against a back pressure of 2.5 and 1.7 kPa (9.0, 5.0 in H₂0), respectively. Shown also in Figure 2 are flow-pressure relationships for the UPAS cyclone inlet operated with three common air sampling filters (nominal 37mm diameter): mixed cellulose ester (MCE; 0.8 μm SKC, Inc.), polytetrafluoroethylene (PTFE; PT37P, MTL Inc.), and PTFE-coated glass fiber (Pallflex Fiberfilm, Pall Inc.). The intersection of a given cyclone/filter curve with the UPAS pump curve represents the operating point for the UPAS at any particular power level.

Tests of flow control accuracy using the integrated mass flow sensor were within 5% when evaluated against a primary flow standard (Table 3). The internal UPAS battery (lithium-ion; rechargeable via a microUSB port on the side of the unit) lasted approximately 25 hr at 2.0 L/min and 45 hr at 1.0 L/min, respectively, when sampling air through a 37mm Pallflex Fiberfilm filter (with all other sensors running). When tested with a TissueQuartz filter (with approximately twice the pressure drop), the battery life decreased by approximately 2 hours. Additional performance specifications are provided in Table 3. At a distance of 20 cm (the approximate length from the ear to a hypothetical sampler mounted on a lapel within the breathing zone) the UPAS produced 40 dB of A-weighted noise. For comparison purposes, the PEM and XR5000 pump combination emitted 60 dB under similar conditions.

The dimensions of the final 1.0 and 2.0 L/min cyclone designs are provided in Table 2, along with d₅₀ and β values for operation at their design air flows. Equation (3) from Hinds (1999) determines within 0.1% the fraction of particles that a cyclone or other device should collect to match the Code of Federal Regulations (CFR) specification for PM_2.5, η(d)_FRM,

Particle collection efficiency of the 2.0 L/min UPAS cyclone is shown in Figure 3, which depicts the corresponding relationship from Equation (2) for the cyclones whose dimensions, d₅₀ and β values are given in Table 2. Figure 3 also depicts the relationship between aerodynamic particle size and the EPA PM_2.5 criterion [η(d)_FRM from Eq (3)] for comparison. Results for the 1.0 L/min cyclone are nearly identical and shown in the supplementary material.

To characterize the adequacy of the two cyclones, the difference between the collection efficiency of an ideal PM_2.5 collector as given by Eq (3) was compared to the efficiency of each cyclone developed here for a series of hypothetical size distributions. This comparison was conducted for 30 log-normally distributed aerosols with median diameters ranging from 0.5 to 8 μm and with geometric standard deviations from 1.5 to 4.0. Results for these simulations are shown in Figure 4 in the form of a bias plot. For the vast majority of particle size distributions encountered in household or outdoor air, the UPAS cyclone catch should match that of an FRM sampler within approximately 5%. Only one size distribution (8 μm median diameter with a geometric standard deviation of 1.5) produced a bias greater than 10% between the UPAS and EPA PM_2.5 criterion.

The performance of the PEM and UPAS relative to the EPA FRM sampler is shown in Figure 5 for PM_2.5 test dust. Chamber concentrations spanned a range from approximately 25 μg/m³ to 800 μg/m³ over these tests. Both the UPAS and PEMs showed strong correlations with the EPA FRM method across this range (Figure 5). A simple linear regression between the UPAS and FRM samplers gave a slope of 0.996 with an intercept of 3.7 μg/m³. For comparison, the PEM sampler, when regressed against the FRM, gave a slope of 0.959 with an intercept of 11.5 μg/m³. Among replicate samples (i.e., instruments co-located within the chamber), the coefficient of variation was 1.4% for the FRM, 5.1% for the UPAS, and 3.4% for the PEM. The average difference (in absolute terms) in measured PM_2.5 mass concentration was 7% between the UPAS and FRM and 6% between the PEM and FRM. A Bland-Altman analysis showed no directional bias between the UPAS and FRM measurements as a function of chamber concentration (data not shown).

The state-of-the-art for personal exposure assessment has long relied upon expensive and burdensome equipment; this paradigm has limited our ability to determine individual risk at scales relevant to a diverse population (especially for epidemiology). The UPAS represents an attempt to address this limitation with a wearable air sampler that is low-cost, lightweight, and low-burden (silent, no-tubing, etc.). Advantages of the UPAS include its compact size and weight (about 1/3 that of the SKC PEM and XR500 pump) and its nearly silent operation. The integration of wireless communications also serves to streamline device programming and data transfer. The inclusion of a cyclone inlet and filter cartridge directly onto the sampler body also removes the need for an external tubing connection; this is an important component of the design as many people complain about the restriction (and visual stigma) of wearing a long piece of tubing across their body.

A point of emphasis for this design was to enable data collection that is comparable to established metrics (i.e., PM_2.5 mass concentration) and health-based exposure guidelines (WHO, 2014, USEPA, 2013). For this reason, the UPAS was designed to estimate personal exposure to PM_2.5 mass across a 24 hr time scale. To achieve this goal, the development path for the UPAS relied heavily upon the recent emergence of low-cost, do-it-yourself electronics (i.e., Arduino) and additive manufacturing (i.e., rapid-prototyping machines) to move rapidly from a proof-of-concept device through an iterative series of informative design steps. This was especially helpful for the cyclone design, as cyclone design equations such as those used here are semi-empirical; thus, some iteration was required to optimize the size-selectivity of the 1.0 and 2.0 L/min UPAS inlets. At 2 L/min of flow, the UPAS will draw approximately 2.9 m³ of air through the filter over a 24hr sampling period. For gravimetric analysis, typical limits of detection are reported in the range of 10–25 μg of mass accumulated onto the filter, which translates to a detectable air concentration of approximately 8 μg/m³ at the upper end of this range. The unit may also be programmed to operate intermittently, should the user wish for longer run times (for a given flow rate).

Across the range of concentrations tested, the UPAS gave PM_2.5 mass concentrations that were in close agreement with the EPA FRM sampler. To note, the agreement at the lowest concentration tested was achieved using pure Teflon filters because at this low a concentration (~25 μg/m³) the adsorption of semi-volatile aerosols in the background air can bias a PM sample collected on a fibrous filter (such as the Pallflex T60A20) by as much as 100% (Volckens and Leith, 2002, Park et al., 2003). When sampling at higher concentrations, such as those encountered indoors for homes burning biomass, this adsorption bias would be negligible with a larger amount of PM mass accumulated onto the filter substrate. The close agreement between the two instruments is also a function of the cyclone performance, which closely matched the EPA criterion (Figures 3 and 4) for PM_2.5 size selection.

The UPAS contains a suite of environmental sensors to improve the utility of the data collected. The light sensor is used primarily for detecting the presence of a UV signal, which is indicative of the sampler being outdoors (Fahrni et al., 2011). The accelerometer is used to gauge participant activity level, which, in addition to confirming user compliance (i.e., that the sampler is physically worn), can also be used to infer daily behavioral patterns (Bussmann et al., 2001). As a mass-based, time-integrated monitor, the UPAS is geared primarily towards assessing one’s cumulative risk from PM exposure. We chose not to include a real-time PM sensor on the UPAS for several reasons. Although several low-cost nephelometers exist on the market (Wang et al., 2015, Austin et al., 2015), these devices suffer from drift and precision issues, in addition to the myriad problems associated with accurate PM measurement by light scattering. These problems include the important loss of signal at sub-micron particle sizes (Hinds, 1999), humidity effects (Chakrabarti et al., 2004), the variation of Mie scattering intensity with size and refractive index (Benton-Vitz and Volckens, 2008), and insensitivity to differences in particle density. However, future versions of the UPAS could easily incorporate direct-reading PM (as well as gaseous) sensors, provided those sensors do not add significant cost. Future versions of this device could also leverage geo-locating technology, which, when coupled with direct-reading sensors, could provide the advantage of resolving both spatial and temporal exposure patterns of exposure. Other wearable PM monitors have been developed in recent years, such as the MicroPEM and MicroAeth, both of which offer direct reading capabilities of PM_2.5 mass and black carbon, respectively. Both devices have also been field-validated, however, their costs have similarly limited their deployment to relatively modest sample sizes (Chartier et al., 2016, Dons et al., 2011).

At 190 g the UPAS poses a reduced but still noticeable weight burden to the user (by comparison, the PEM/XR500 sampler weighs approximately 700 g). Approximately half of the instrument weight is taken up by the battery, which was sized to achieve greater than 24 hr of sample time at either 1.0 or 2.0 L/min of flow. These flowrates were selected to meet detection limits of typical gravimetric analysis systems. However, with recent advances in automated filter weighing systems (CARB, 2011), sensitive, low-cost analytic chemistry (Cate et al., 2014, Rattanarat et al., 2014, Meredith et al., 2016), and non-destructive sample analysis techniques (Ramanathan et al., 2011), it should be possible to develop future versions of the UPAS that operate at lower flowrates. Because the UPAS battery size/weight scales nearly directly with flow (for a pre-determined sample duration), these reductions would reduce the weight burden to the wearer. Future reductions in size and weight should also reduce manufacturing costs.

The UPAS shows promise for increasing our ability to assess personal PM exposures by reducing the size, weight, and cost of the sampler. As a result, sampling can be conducted at scales that are more relevant to epidemiologic and community-based research. Development of the UPAS was largely made possible by the recent revolution in ‘original equipment manufacturer’ electronics and open-source software. Looking forward, these industry trends should continue, and thereby aid the development of low-cost sensors that produce public health gains through applied research, advocacy, and awareness.