Outward and inward protections of different mask designs for different respiratory activities

Aerosol generator

Aerosol typically refers to droplets with diameters less than 5 µm and are thought to be responsible for airborne transmission of diseases, since they can stay suspended in air for a long period of time [15, 16]. Aerosol droplets generated during respiratory activities (such as talking and coughing) and in healthcare settings (such as intubation) are thought to vary between 0.1 and 10 µm [16–18]. Here, we generate polydisperse aerosol droplets spanning this range from 2 wt% NaCl aqueous solution using an ultrasonic humidifier (home appliance) at room temperature of 22°C and 58–65% humidity.

The diameters of the droplets D generated vary between 0.3–5.0 µm, as measured using MET ONE HHPC+ optical particle counter (See histogram in Fig. 1). Note that droplets below 0.3 µm in size cannot be detected by the instrument. NaCl salt is added to mimic the solid content in respiratory fluid droplets and prevent the droplets from evaporating completely [19].

• Download figure
• Open in new tab

FIG. 1.

Distribution of aerosol droplets generated using an ultrasonic humidifier.

The total number of droplets generated can be tuned by adjusting the operation time of the humidifier. Millions of droplets can be produced within seconds.

Simulating different respiratory activities

We connect a mechanical resuscitator or an Ambu bag to a life-sized manikin head with a 3 cm-sized orifice at the mouth region. Previously, we have used a similar set-up to quantify the outward protection of a barrier enclosure against airborne transmission [20].

To simulate exhalation and inhalation, we compress and re-inflate the Ambu bag fully by hand within Δt = 3 s. The volume V expelled and drawn in with each compression and re-inflation cycle is determined to be 0.75 ± 0.05 L, generating a mean flow rate Q = V/Δt = 15 L min⁻¹ (International Organization for Standardization [ISO] standard for a female performing light work [7, 21]) and typical speed U ∼ 30 cm s⁻¹. V is determined by releasing the expelled air into a jar filled with water and measuring the volume of water displaced. A metronome is used to keep the operator in time. We did not find significant differences in V for different human operators.

To simulate coughing, the Ambu bag was compressed by hand as quickly as possible. Experimentally, we found that V is expelled within Δt = 0.30 ± 0.03 s to generate Q = 140 L min⁻¹ or the equivalent U ∼ 3 m s⁻¹, close to reported physiological parameters of human coughs [22]. Δt was determined by recording the sounds of simulated coughs and measuring the time duration when the sound amplitudes are above background noise (Fig. 2). Previously, the sound of human coughs has been used to analyze cough duration and amplitudes [23].

• Download figure
• Open in new tab

FIG. 2.

Sounds recorded during simulated coughs.

Measuring outward and inward protections

The manikin head is placed inside an acrylic chamber with dimensions w × l × h = 0.6 0.5 0.6 m³ and volume V₁ = 180 L (chamber 1 in Fig. 3). An Ambu bag connects the manikin in chamber 1 to a smaller chamber with dimensions w × l × h = 0.4 × 0.5 × 0.4 m³ and volume V₂ = 80 L (chamber 2 in Fig. 3). Air can flow from one chamber to the other with the flow direction controlled by a one-way valve inside the Ambu bag.

• Download figure
• Open in new tab

FIG. 3.

Experimental setup to measure (a) outward and (b) inward protections by different mask designs.

To measure the outward protection of masks P_out, we first generate aerosols of different concentrations ρ₂(D) inside chamber 2 (Fig. 3a). We then simulate four exhalation or cough cycles to transfer a portion of the aerosol N_source = 4V ρ₂ from chamber 2 (the source chamber) to chamber 1. P_out can then be determined by comparing N_source with the number of droplets reaching chamber 1 and not filtered out by the mask N_out = V₁ρ₁, since , where ρ₁ is the droplet concentrations in chamber 1 and T is the transfer rate of the droplets between the two chambers with no mask. T < 1 accounts for droplets that are deposited along the connecting pipes and the Ambu bag during aerosol transfer. To maximize T, we minimize the pipe lengths and avoid any bends in the pipes.

Since N_out ≤ N_source, ρ₁ ≤ 4V/V₁ ρ₂ = 0.02 ρ₂, i.e. ρ₁ << ρ₂. Hence, we used two different optical particle sensor models (MET ONE HHPC+ and Sensirion SPS30) with suitable ρ range to accurately measure ρ₁ and ρ₂. The specifications of the two models are summarized in Table I. The optical particle sensors have four bin sizes which allow us to measure P_out for four different droplet size ranges: 0.4 ± 0.1, 0.75 ± 0.25, 1.5 ± 0.5 µm and 3.5 ± 1.5 µm. Note that the bin widths are slightly different for bins 3 and 4 for the two sensors, which we ignore in our analysis for simplicity. Through-out the experiment, small fans were used to distribute to the aerosol uniformly inside the two chambers.

A similar setup was used to measure the inward protection of masks P_in, except the aerosol is now generated in chamber 1 and the direction of the flow is from chamber 1 to 2 (Fig. 3b). After four inhalation cycles, the protection efficiency is determined by comparing N_source = 4V ρ₁ and the number of droplets reaching chamber 2 unfiltered N_in = V₂ρ₂, where ρ₁ and ρ₂ are measured using Sensirion SPS30 and MET ONE HHPC+ sensors, respectively. As before, we expect . By varying the initial number of droplets from the source N_source, we can obtain the slopes of N_{out, in} against N_source. From equations 1 and 2, we deduce that . The outward and inward protections can then obtained through the relation . Experimental uncertainties in P_{out, in} depends on the uncertainties in the two slope values, i.e.

Mask fit

In our experiments, the cloth/surgical mask is securely fastened using the ear loops to cover both the nose and mouth of the manikin snugly. The nose bridge is also adjusted to minimize gaps above the nose area. It is not possible however to completely eliminate gaps for cloth and surgical masks.

Healthcare workers typically undergo professional mask fitting to eliminate gaps around an N95 mask. To obtain a similar level of mask fit, we use a silicone sealant around the mask perimeter after securing it on the manikin. To simulate the lower level of protection for N95 masks worn by the general public, we mould the mask snugly onto the manikin but without applying the silicone sealant.

Background particle levels

The number density of background particles ρ (D=0.5– 5.0 µm) in our lab is about 1000 L⁻¹. To minimize background ρ, we used a commercially available air ionizer from Zero2.5 Biotech, Singapore. Turning on the ionizer for 1 min before the start of each experiment reduces ρ < 10 L⁻¹ insider chambers 1 and 2. Background ρ remains low for at least 1 hour after the ionizer has been switched off.

Outward and inward protections

We look at four different mask designs in this study: an N95 mask, a surgical mask and two cloth masks widely distributed by the Singapore government to its population. All four masks, including the cloth masks, contain a non-woven filter layer in the middle, whose scanning electron micrographs are shown in Fig. 4. For cloth mask 2, the filter layer can be removed from the middle slot by the wearer. According to the manufacters, the surgical mask and the two cloth masks have particle filtration efficiencies (PFE) in excess of 95% for 100 nm-sized particles. However, the reported PFE values ignore the presence of gaps around the masks.

• Download figure
• Open in new tab

FIG. 4.

Different mask designs used in the study, consisting of three to four layers, including a non-woven filter layer. Scale bars for the scanning electron micrographs are 30 µm.

A typical set of experimental results for the manikin with and without mask is summarized in Fig. 5 (dashed lines, unfilled markers and full lines, filled markers, respectively). In this case, the mask is cloth mask 2 with the filter layer removed. As discussed in the Materials and Methods section, we first generate aerosol droplets with known number densities ρ_1,2 at the source chamber which we use to calculate N_source(Fig. 5a–c). ρ_1,2 values remain relatively unchanged during the four cycles of the simulated respiratory activities, since the volume of aerosol transferred 4V is much smaller than the total volume of the source chamber V_1,2.

• Download figure
• Open in new tab

FIG. 5.

(a–c) Number density of aerosol droplets at the source: ρ₂ in chamber 2 for exhaling and coughing and ρ₁ in chamber 1 for inhaling. Different droplet sizes are coloured differently. The four exhalation, coughing and inhalation cycles are coloured gray and labelled 1 to 4. (d–f) Number density of aerosol droplets reaching the other chamber after simulated respiratory activities with and without mask. (g–i) Plot of total number of droplets N_{out, in} reaching the other chamber against the total number of droplets from the source N_source. ΔN_{out, in} are calculated based on the standard deviation of ρ_2,1(t) in a–c, while ΔN_source is calculated based on the uncertainty ρ values reported by the manufacturer for Sensirion SPS30.

With each cycle, the number densities of droplets reaching the other chamber increases until they reach a plateau (indicated by gray lines on Fig. 5d–f) whose values we use to calculate N_{out, in}. For larger droplets D > 1.0 µm, evaporation is unimportant and their number densities remain constant after the initial rise (red and green lines). In contrast, Ostwald ripening drives the evaporation of smaller droplets and their number densities are only constant for a few seconds before decreasing (blue and black lines) [24, 25].

For each experimental case (different respiratory activities, with and without masks), we varied the initial number of droplets from the source N_source and measured the number of droplets reaching the other chamber N_{out, in} (Fig. 5g–i). In our experiments, N_source ∼ 10⁶ is significantly more than the number of respiratory and aerosol droplets in a single human exhalation/cough ∼ 10³ [1, 26]. However, the aerosol number densities should still be considered low: for ρ_1,2 ∼ 10³ cm⁻³, the typical spacing between neighbouring droplets l ∼ ρ^−1/3 = 1 mm or about a thousand times the droplet diameters.

To a good approximation, the droplets do not interact with one another. Hence, we expect equations 1 and 2 to remain true and that N_{out, in} ∝ N_source which we verified experimentally for all the droplet sizes (Fig. 5g–i). The full and dashed lines are the lines of best fit for a linear regression with zero intercept for cases without and with mask, with lower slopes indicating better aerosol filtration. The slopes correspond to T for the case of no mask and (1 − P_{out, in})T with mask, from which we can calculate P_{out, in} using equation 4. From the various slopes in Fig. 5g–i, we can deduce that P_{out, in} vary widely for different respiratory activities and droplet size ranges, which will be elaborated in the rest of the manuscript.

We repeated the experiments for the four mask designs. Different masks offer different levels of protections P_{out, in}, as evident from the different slopes of N_{out, in} against N_source for droplets of size range D = 0.75 ± 0.25 µm (Fig. 6a–c). Similar experiments can be performed for the other D range to obtain the corresponding P_{out, in} (Fig. 6d–f).

• Download figure
• Open in new tab

FIG. 6.

Plot of N_{out, in} against N_source for (a) exhaling, (b) coughing and (c) inhaling. The droplet size range here is D=0.75 ± 0.25 µm. Lines are best-fit results from linear regression. Results for N95 masks are not shown to avoid cluttering. (d–f) Plot of P_{out, in} for different respiratory activities, mask designs and droplet size ranges D = 0.4 ± 0.1, 0.75 ± 0.25, 1.5 ± 0.5 µm and 3.5 ± 1.5 µm. Error bars in P_{out, in} are calculated using equation 5. Lines are guides to the eye.

Not surprsingly, a well-fitted N95 mask achieves P_{out, in} > 90% for all D range and all respiratory activities, close to the reported 95% filtration efficiency (red solid line, filled markers). In contrast, an N95 mask with sub-optimal fit (red dashed line, unfilled markers) provides little to no additional protection P_out for exhalation as compared to cloth and surgical masks (Fig. 6d); in fact, a poorly fitted N95 mask offers even less protection P_in < 40% for inhalation as compared to cloth and surgical masks (Fig. 6f). This is because more than half of the air flows through the gaps around the N95 mask unfiltered.

Cloth masks (blue lines and markers) offer similar level of protections P_{out, in} ≈ 50% as surgical masks (green lines and markers) as long as they contain a non-woven filter layer. Removing the filter layer significantly reduces P_{out, in} by a factor of 2 or more (compare light blue solid and dashed lines).

For the same flow rate Q = 15 L min⁻¹, we found that P_in for inhalation is higher than P_out for cloth masks, consistent with results by Pan et al. (2021) [12]. This is likely because the negative pressure generated during inhalation draws the flexible mask closer to the face, reducing gaps. In contrast, no such effect was observed for an N95 mask due to its rigid, hard shell.

P_out for coughing are consistently highest for all mask types and D ranges as compared to exhalation and inhalation. This is because coughing with its higher flow rate Q = 140 L min⁻¹ and pressure gradient forces more of the aerosol laden air through the filtration layer. As a result, a poorly fitted N95 mask is able to achieve P_out close to that of a well-fitted N95 mask (red dashed and solid lines in Fig. 6e).

Recently, the CDC reported that the wearing of a cloth mask on top of a surgical mask or double masking can help improve mask fit and reduce a wearer’s exposure by more than 90% [7]. Here, we found similar improvement in P_out for a simulated cough when wearing cloth mask 1 or 2 over a surgical mask (Fig. 7b). However, double-masking provides little to no additional protection P_out for exhaling (Fig. 7a) and less protection P_in for inhaling (Fig. 7c).

• Download figure
• Open in new tab

FIG. 7.

Outward protection by double masking (surgical and cloth mask as inner and outer layers, respectively) when (a) exhaling and (b) coughing. (c) Inward protection by double masking when inhaling.

Interestingly, the performance of double-masking, closely mirrors that for a poorly fitted N95 mask (compare black dashed lines in Fig. 7a–c to red dashed lines in Fig. 6d–f). This suggest that while the addition of cloth mask improves the material filtration efficiency, resulting in improved P_out for coughing, it also decreases breathability and increases material rigidity, resulting in poorer P_in for inhalation.

We can now calculate the combined outward and inward protections for different mask designs and respiratory activities , which is also the risk reduction in airborne transmissions with universal masking in the entire population.

For respiratory activities with low flow rates such as exhaling and inhaling, there is little to no benefit for the general public to wear N95 mask (Fig. 8a) or double masks (Fig. 8b). For respiratory activities with high flow rates such as coughing, there is some additional protection with poorly fitted N95 mask (Fig. 8c) and double masks (Fig. 8d), as compared to a single surgical/cloth mask. For COVID-19, it is not clear which respiratory activity (coughing/sneezing vs. exhaling) is the dominant mode of airborne transmissions.

• Download figure
• Open in new tab

FIG. 8.

Combined outward protection when exhaling and inward protection when inhaling for (a) single masks and (b) double masks. Combined outward protection when coughing and inward protection when inhaling for (c) single masks and (d) double masks.

A well-fitted N95 masks offers the best P_combined close to 100% and should be the default choice for high-risk activities, such as healthcare workers in contact with infectious patients. For the general public, a single surgical and cloth mask offers significant protection with P_combined between 50% and 90%, as long as it contains a filter layer.

Limitations of our study

We have only looked at the effectiveness of mask protections on a single manikin head, even though the mask fit is sensitive to head geometry. Moreover, the manikin is made of a rigid plastic material which is different from deformable human flesh. It is also likely that filtration material will degrade with regular use and washing. In future studies, we plan to address these shortcomings, for example by 3D printing manikin heads of different sizes and covering them with a soft silicone material [27].