ADAPT: A Smart Mask for Active Defense Against Air-Borne Pathogens

Face masks are a primary preventive measure against airborne pathogens. Thus, they have become one of the keys to controlling the spread of the COVID-19 virus. Common examples, including N95 masks, surgical masks, and face coverings, are passive devices that minimize the spread of suspended pathogens by inserting an aerosol-ﬁltering barrier between the user’s nasal and oral cavities and the environment. However, the ﬁltering process does not adapt to changing pathogen levels or other environmental factors, which reduces its effectiveness in real-world scenarios. This paper addresses the limitations of passive masks by proposing ADAPT, a smart IoT-enabled “active mask”. This wearable device contains a real-time closed-loop control system that senses airborne particles of different sizes near the mask by using an on-board particulate matter (PM) sensor. It then intelligently mitigates the threat by using mist spray, generated by a piezoelectric actuator, to load nearby aerosol particles such that they rapidly fall to the ground. The system is controlled by an on-board micro-controller unit (MCU) that collects sensor data, analyzes it, and activates the mist generator as necessary. A custom smartphone application enables the user to remotely control the device and also receive real-time alerts related to recharging, reﬁlling, and/or decontamination of the mask before reuse. Experimental results on a working prototype conﬁrm that aerosol clouds rapidly fall to the ground when the mask is activated, thus signiﬁcantly reducing PM counts near the user. Also, usage of the mask signiﬁcantly increases local relative humidity (RH) levels.


Introduction
The onset of the SARS-CoV-2 pandemic in late 2019 saw an immediate adoption of preventive measures to control the airborne spread of this virus 1 . Citizens worldwide were advised to take precautions such as social distancing, face coverings, and good hygiene practices, while governments adopted technologically advanced measures like contact tracing. Face coverings are effective in reducing the spread of the virus, even with high levels of exposure in closed spaces 2 . However, existing face masks do not adapt to changing exposure levels, i.e., only provide passive protection 3 . As a result, they are inefficient when these levels are unpredictable and fluctuate over time, which is generally the case. Thus, there is a need for smart protection mechanisms that not only act as physical barriers but also use active mitigation techniques to reduce the spread of pathogens whenever necessary.
Regular human activities such as breathing, talking, sneezing, and coughing discharge droplets into the surrounding environment, creating a cloud of aerosol particles ( Fig. 1(a)) 4 . For example, a single sneeze 5, 6 produces as many as 40,000 droplets with diameters between 0.5 − 12 µm that are expelled at speeds up to 100 m/s and can reach distances up to 8 m. Similarly, a single cough produces up to 3,000 droplet nuclei, and comparable numbers are typically recorded after 5 minutes of talking. Such aerosol clouds include various types of cells (e.g., epithelial cells and leukocytes), ions present in mucus and saliva (e.g., Na+, K+, and Cl−), and, potentially, various infectious agents suspended in the droplets (e.g., bacteria, fungi, and viruses). Even weak air currents caused by innocuous daily activities, such as walking 7 or opening doors, can transport these infectious droplets over long distances (> 10 m). Such airborne transmission is the primary driver for virus spread in indoor spaces such as bathrooms, doctor's offices, daycare centers, and public transportation; it is also important in many outdoor environments 8 .
The SARS-CoV-2 virus is 100 − 120 nm in diameter and can remain suspended within droplets of diameter d > 0.2 µm 9 . Droplets with d > 5 µm fall to the ground quickly due to gravity, while very small droplets evaporate 10 Figure 1. (a) Droplets expelled from mouth during various activities like exhalation, coughing, and sneezing create aerosol clouds that travel to different distances. (b) Expelled droplets containing suspended virus particles are small and easily turn into aerosols (left). The proposed ADAPT smart mask sprays mist on these particles (right), which loads them (i.e., increases their size and mass) and thus causes them to rapidly fall to the ground. and aerosolize in a few seconds to droplet nuclei with d ≈ 1 µm 11 . Fortunately, most masks can filter out droplets of this size: many materials have ≥ 96% filtration efficacy for particles > 0.3 µm, including 600 TPI (threads per inch) cotton, cotton quilts, and cotton layered with chiffon, silk, or flannel 12 . Thus, the most critical requirement for the proposed closed-loop smart mask is to eliminate the small (but potentially significant) fraction of virus-laden droplets that are < 0.3 µm in diameter.
Unlike existing masks, the proposed system actively detects the presence of nearby airborne particulate matter (PM) that can contain viruses or other pathogens. It then analyzes the sensor data (including concentration, size distribution, and other PM properties) to determine a smart mitigation strategy that minimizes aerosol spread. In our experimental prototype, mitigation is provided by a cold mist generator that loads the particles, which increases their aerodynamic diameter and mass and makes them quickly fall to the ground ( Fig. 1(b)). The settling time t s of aerosols scales as d −2 , making the proposed mitigation method particularly effective for small particles (d < 0.3 µm) that are not efficiently filtered by masks. For example, the value of t s in both still and turbulent air decreases from ≈ 130 hours for d = 0.3 µm to only ≈ 8.2 min for d = 10 µm 13,14 . However, all particles of a given size have the same settling time in still air, while in turbulent air the probability of settling increases exponentially with time, i.e., as p(t) = 1 − exp(−t/t s ). Additionally, the mist spray creates an airflow pattern that actively blows droplets away from the user. Adaptability is provided by algorithms running on the on-board controller that adjust the mist generator's spray angle, intensity, and duration based on sensor data. Such active closed-loop monitoring and protection can remove viruses (and other pathogens) from the air before they can infect others, thus reducing the need for periodic disinfection of the area while also providing increased protection to both the wearer and others in the vicinity. In fact, groups of smart mask users can collaborate to increase the local relative humidity (RH), which in turn reduces the probability of airborne transmission 15 . Specifically, increased RH allows nasal mucus to more easily carry airborne pathogens to the stomach where they are destroyed by digestive acids.
The proposed smart mask contains two main hardware modules: the PM sensor and the active mitigation device. Standard PM sensing techniques include impactors and filters 16 for collection, and real-time polymerase chain reaction (RT-PCR) 17 or surface-enhanced Raman spectroscopy (SERS) 18 for analysis and identification. However, it is difficult to implement such techniques in real-time within a portable form factor and limited power budget. Thus, we employ a simpler detection approach based on laser scattering. Specifically, the PM sensor (SPS30, Sensirion) detects scattered laser light to count airborne particles and categorizes them in "bins" based on mass concentration (µg/m 3 ) and number concentration (1/cm 3 ). Since aerosol travel distance also depends on environmental factors such as RH and temperature 16 , the on-board control algorithm combines data from PM and environmental sensors to determine air quality in real-time. The proposed algorithm ( Fig. 2(a)) classifies air quality based on health risk (e.g., "very high,", "high", "moderate", and "low"). The output categories are then encrypted for security and sent to the mitigation module via either a wired or wireless (e.g., Bluetooth) connection. A "high risk" output triggers the mitigation module ( Fig. 2(b)), which generates aerosolized mist on-demand using a piezoelectric transducer. The smart mask can also connect to authorized mobile devices through a Bluetooth module. A custom mobile application allows users to monitor current air quality, check system status (e.g., battery life, liquid level, notifications, and alerts), and manually override the on-board automated mitigation algorithm if desired.
While our prototype uses tap water to generate the mist spray, in general a variety of liquids, colloids, or solid suspensions can be used. For example, solid additives can increase droplet density and/or size (thereby decreasing t s ), while disinfectants can actively kill pathogens. The best disinfectant for a given pathogen can be found using guidelines provided by agencies such as the U.S. Centers for Disease Control and Prevention (CDC) 19 ; common choices include diluted bleach, soap, and > 70% alcohol solution. However, since the safety of such disinfectants when inhaled is unknown, their use will be studied in future work. Furthermore, later versions of the device can be equipped with machine learning (ML) algorithms that learn when respiratory droplets are likely to be present in a location and proactively employ the proposed active mitigation mechanism.

Results
We tested a prototype of the ADAPT smart mask in various scenarios. Our experiments confirm the expected increase in particle sizes and decrease in settling time when the mask is turned ON. We also tested our mitigation module with several water-based solutions. Finally, we verified the ability of the mask to increase local RH.

Effects of ADAPT on aerosol concentrations
We emulated the scenario of a person wearing an ADAPT smart mask by mounting it at a height of ≈1.6 m inside an enclosed environment with no significant air currents. The smart mitigation module used water to generate the mist spray, while a separate humidifier generated water droplets to emulate a potentially-infectious aerosol cloud. Finally, PM sensors were used to measure particle counts and sizes in the region around the mask.
Two PM sensors were used to compare aerosol properties with and without the smart mask. One of the sensors ("int") was mounted on the mask along with the mitigation module, while the other ("ext") was placed on the ground. Initially, the humidifier was turned ON for 15 sec with the mitigation module turned OFF. The resulting outputs of both PM sensors (local number and mass concentrations) were monitored until most aerosols had settled out (≈ 160 sec). Next, the tests were repeated with the mitigation module turned ON but the humidifier turned OFF. Finally, the the effectiveness of the smart mask was verified by turning ON both the humidifier (for 15 sec) and the mitigation module (for 15 sec after the PM sensor in the mask detects a significant change in concentration).
The observed trends in total PM concentration (1/cm 3 ) and density (µg/m 3 ) during these tests were described in our previous work 20 and are reproduced in Fig. 3 for completeness. Figs. 3(a)-(c) show that the time-averaged PM concentration around the mask in the d = 0.3 − 1 µm range decreases significantly (∼40%) when the mitigation module is enabled. Fig. 3(d) confirms that the aerosol cloud generated by the humidifier has a relatively uniform size distribution that peaks in the d = 0.3 − 0.5 µm range, similar to that generated during daily activities 5, 6 . Finally, Figs. 3(a)-(c) show a significant increase in PM concentration and density for the "ext" sensor (on the ground) when the mitigation module is turned ON, which confirms the expected aerosol loading effect of the mist spray.

Temporal dispersion of aerosol particles
To visualize the aerosol loading effect with respect to time, we measured PM concentrations over the size range 0.5 − 2.5µm at a sampling rate of 1 sec. Measurements were performed simultaneously at three different levels (i.e., heights) L i , where i ∈ {1, 2, 3}, by interpolating the outputs of four vertically-separated PM sensors. The sensors were placed slightly off the line of sight joining the smart mask and the humidifier to minimize their impact on the aerosol cloud. The experiment was started at t = 0, and the external humidifier was switched ON at t = 10 sec. At t = 15 sec, the humidifier was switched OFF and the mask turned ON. Finally, mist generation by the mask was stopped at t = 20 sec, and PM readings recorded until t = 120 sec. The resulting matrix of PM concentrations is shown in Fig. 4 in two cases: (a) mask OFF (used as a control), and (b) mask ON. These plots allow the dispersion of the aerosol cloud over time to be visualized. Fig. 4(a) shows the presence of a significant aerosol cloud generated by the humidifier for times up to 80 sec 4/9  (particularly at level L 1 ). By contrast, Fig. 4(b) shows no significant aerosol cloud for times > 20 sec. Additionally, concentrations at the lower levels (L 2 , and L 3 ) are significantly higher than in the earlier case, again suggesting that aerosol loading by the mist spray greatly reduces the settling time.

Effects of liquid type on mitigation metrics
Aerosol loading experiments were repeated with aqueous solutions of salt (NaCl) and sugar (sucrose) used within the mitigation module instead of tap water. The resulting time-averaged PM concentrations observed near the mask are summarized in Fig. 5. Both salt and sugar solutions result in significantly larger droplet sizes than pure water, particularly for d > 1 µm. Such increases in mean droplet size, which are likely driven by the well-known increase in surface tension and viscosity of water solutions for certain solutes (including NaCl and sucrose) 21,22 , may provide users with options for further reducing aerosol settling times.

Effects of ADAPT on relative humidity (RH)
Research suggests that maintaining higher RH levels can help improve resistance against influenza infections 15,23 .
Since ADAPT relies on a water-based mitigation module, we expect its use to provide additional health benefits by increasing local RH levels. We verified this behavior by using a capacitive humidity sensor (DHT11, Adafruit) to measure RH at different locations near the mask (Fig. 6(a)) when its mitigation module is ON. The results ( Fig. 6(b)) show significant RH increases in all cases compared to the equilibrium RH of the room (≈55%). For example, the "central" locations exhibit RH values between 64% (at 60 cm) and 77% (at 15 cm).

Discussion
We have demonstrated ADAPT, a new active closed-loop "smart mask" paradigm that can both defend the wearer and reduce the spread of airborne pathogens such as SARS-CoV-2. ADAPT is capable of controlling the duration and intensity of mist generation based on its awareness of the current location (e.g., a hospital, quarantine zone, or care facility), ambient conditions (e.g., temperature, humidity, and human occupancy), and health indicators (e.g., age and preexisting conditions). The mask design can potentially be further extended to provide real-time active protection against other airborne hazards including pollutants, dust, and pollen. The physical design of ADAPT was derived from an open-source mask project, modified to i) add breathable perforations, and ii) house a liquid reservoir, piezoelectric transducer, and PM sensor. The perforations are internally shielded using replaceable air filters to further reduce PM concentrations. The experimental prototype was 3D-printed using polylactic acid (PLA). It consists of two main parts, each weighing ∼110 gm: i) the mask itself, and ii) a belt unit that houses the micro-controller and battery, Tests on volunteers confirmed that the device is comfortable enough for long-term use. Further improvements in the manufacturing process (e.g., using injection molding) and integration of the electronics can reduce system size/weight and eventually eliminate the belt unit.
Test results from a working ADAPT prototype confirm that the mitigation module significantly reduces PM concentrations near the mask via aerosol loading. The use of aqueous salt and sugar solutions increases mean droplet diameter compared to water, and thus has the potential to further reduce PM settling times. Such solutions are also known to have antibacterial properties 24 . In addition, the mask was found to increase local RH, which is beneficial since it enables nasal mucus to filter out pathogens before they enter the respiratory tract 1,23 .
Further work will focus on improving the sensor to provide additional information on pathogen type, thus allowing the mitigation strategy to be appropriately optimized. Laser-induced fluorescence 25 is promising for this purpose. The on-board algorithm for deploying the mitigation technique can also be improved. For example, artificial intelligence (AI) can be used to determine i) when the module should be turned ON, and ii) the optimum mitigation parameters as a function of location, environmental conditions, health data, and other variables.

Smart mask
The current version of ADAPT consists of two main parts: the mask and the belt unit ( Fig. 7(a)). The mask, which is 3D printed, contains the PM sensor, the liquid reservoir for the mitigation module, and the piezoelectric transducer. A flexible cable provides power for the sensor and transducer from the belt unit. The latter contains the micro-controller and battery, and also a relay, oscillator, and amplifier for driving the transducer. The PM sensor uses a small built-in fan for self-cleaning. However, the exhaust from this fan interferes with motion of the aerosol particles being measured. This effect was reduced by adding a small plastic tube to redirect the exhaust.
The piezoelectric transducer is disc-shaped and contains a fine vibrating mesh sandwiched between two electrodes (one facing the water reservoir and the other facing the environment). Fluid pressure in the reservoir was kept low enough to allow surface tension to prevent water leakage through the mesh. The mitigation module generates mist by driving the transducer near its resonant frequency (110 kHz) to generate a pressure drop across the mesh. The transducer consumes 300 mA when driven at 5 V from a single 2200 mAh Li-polymer battery; the corresponding vaporization rate is ≈ 300 ml/hour. During typical use, the module is turned ON (via a relay) for 30 sec every time high PM counts are detected. Assuming 100 such detection events per day, the expected battery life is ≈ 6 days.
The micro-controller (ESP32, Espressif) interfaces with the PM sensor via a wired I 2 C bus. Acquired data is relayed to nearby mobile devices over Bluetooth and accessed via an Android application ( Fig. 7(b)). The app lets users manually turn the mitigation module ON/OFF, thus overriding the built-in automated control loop. The latter uses a threshold-based algorithm (Fig. 2) to analyze data from the PM sensor and then make decisions on turning the mitigation module ON/OFF. The algorithm also monitors the mobile device's built-in microphone to detect relevant audio cues and switch ON the mitigation module if necessary (e.g., when a sneeze or cough is detected).

Experimental setup
During the experiments, the smart mask was operated in manual mode. A humidifier was used for on-demand generation of aerosol clouds in the 0.3 − 2.5 µm range, thus replicating the effect of daily activities like talking, coughing, and sneezing. The experimental setup is shown in Fig. 7(c). Two PM sensors were used during initial tests: one within the smart mask (which is placed at the height of 1.6 m), and the other placed on the ground midway between mask and humidifier (PM sensor #4 in the figure). Both sensors were connected to a laptop to acquire data, while the mobile app was used to control the mitigation module.