Removal efficiency of PM2.5 and PM10 by four types of masks

Binh Nguyen, Independent Researcher, Hanoi, Vietnam

Four types of face masks including fabric, surgical and for-air-mask, particulate respirator masks were evaluated for PMs removal efficiency (RE). Masks are mounted on a Ф6-cm-PVC pipe and the air was pulled out the box by a fan with varying speed. PM2.5 and PM10 in filtered air and in the background were measured by Plantower PMS7003 light-scattering sensors. The removal efficiency was calculated by a ratio of the difference between PMs concentration in filtered air to the background air. Each mask was tested with the ambient air from 2 to 6 days. In four type of masks, removal of PMs of the fabric is 8.2%-14%, the surgical is 29%-39%, the air mask is 0.03% to 41% and 3M masks are 48-58%, AQBlue, Airphin masks are from 90-95%. Analysis of RE with airflow rate and photographs of the internal structure of the masks supports the RE result. A correlation of airflow to RE from -0.76 to -0.89 indicated a lower RE to a higher flow rate with surgical and air masks. The correlation coefficient of -0.51 for 3M Aura suggested improvements to airflow a higher airflow without significant RE reduction. Fabric masks, which are the dominant type on Hanoi streets, is not effective to filter out PMs. 3M-brand masks showed the highest removal but not high as 90%+ with standard tests. The form fitting is not evaluated in this study.
1. Introduction

Particulate matters pollution emitted directly and derived from fossil-fuel burnings in dense and crowded population leaves a few choices to urban residents. At home, air filters installed in air-conditioned units or a seperate unit in close room could reduce fine particles or PM2.5 between 60%-84% [1],[2]. On the streets, motorbikes are the dominant transportation means for commuters in Vietnam. In Hanoi, the total motorbikes were reported about 6M units in 2019, about 10 times more than passenger cars. Protection to respiratory systems for the riders is limited to various types of face masks.

By the author's observation, about 60%-80% riders are wearing face mask, in which more than half is various fabric face masks and surgical masks are the dominant of the lesser half. Brand-name particulate respirators are in few to be seen.

Shakya at. el., 2016 [3] tested fabric and surgical masks with monodispersed aerosol sphere only found a high efficacy from 30% up to 100% with sizes from 30nm to 2.5µm. In combination with diesel exhaust and monodispersed aerosol, the removal efficiency drops to the range of 15%-57% with the fabric mask performed the worst. Tests with respiratory masks such as N95 showed marginal improvement with fine particles (PM2.5).

Concerns about chronical diseases and mortality rate link to air pollution resulted in the 68th World Health Assembly passed a resolution recognized that each year 4.3M deaths attributed to the indoor and 3.7M deaths. attributed the outdoor air pollution. The mortality rates are higher in developing countries. The author analyzed fine particles or particles with dynamic sizes less than 2.5µm (PM2.5) monitored by the US. Embassy in Hanoi and the US. Consulate in HCMC, Vietnam from 2016 to mid-2019, the good air quality level is accounted for 26% in Hanoi and 50% in HCMC by the Vietnam standard for the daily average. The WHO recommmended a limit of 25µg/m3 for the daily average concentration, a 50% lower than the Vietnam standard.

The author was motivated to evaluate the removal efficiency of popular types of masks against PM(s) and mainly PM2.5. Available methods for certificating particulate respirators required by the National Institute for Occupational Safety and Health (NIOSH) in the US under 42 CFR* 84 and guidelines from the Occupational Safety and Health Administration (OSHA), in the EU under EN 149:2001, in China under GB2626:2006. Because of convoluted technical details and expensive standard equipment, the author decided to improvise a testing box to evaluate the removal efficiency of the masks using low-cost sensors. A total of 14 tests on 11 masks were conducted over 45-day periods on four main types of masks including fabric, surgical, airmask and respirator. The form fitting factor is critical to the overall performance of a mask. Because of an improvising setup, it is not accounted in by this study.

2. Methods and Materials

A plastic box was modified to install one 5-V exhaust fan and another end with a Ф6-cm-PVC pipe. The open space was sealed with the silicone sealant except for the exhaust hole and a mounting place for the mask. One dust sensor, Plantower PMS7003, was installed inside the box with a microcontroller, NodeMCU ESP8266, to read the PM2.5 and PM10 concentration (hereafter referred as PMs) and then transmit the data over a wireless network. A potentiometer was installed to manually adjust the fan duty when necessary. The background PMs concentrations were measured by another PMS7003 sensor placed inside a plastic box with a 1mm-stainless steel mesh in the intake. The testing equipment was placed on a balcony on the ~10th floor of a high-rise building in the South of Hanoi. The balcony faced a quiet corner with light to moderate traffic on the ground floor with mostly motorbikes and passenger cars. No visible industrial chimney stack was observed nearby.

Box setup
Fig. 1: Configuration and sensors setup for testing mask and background concentration monitoring.

Each test lasted from 2 to 6 days with fan duty varied from 25% to 100% by automating a target fan duty with the uptime counter by the NodeMCU. Each level of fan duty lasted 3 hours. Before each test, one interval of 3 hours was used as the blank sample with no mask mounted. The PMs concentration was compared with the background concentration to cross check the sensors.

A total of 11 masks divided into 4 groups: fabric, surgical, air mask and 3M brand's. Details of each masks is presented in Table 1. When testing, masks were mounted on PVC pipe and tightened by 3-4 rounds of rubber bands.

Table 1: List of masks tested in this study. Unless undicated, the mask are new and produced in Vietnam.
Group ID Description Price (kVND)
Fabric FU1 used fabric mask, >2 years in use ~20
FU2 used fabric mask, >2 years in use ~20
FN new fabric mask sold in drug counter in carboard box, indicated multiple awards and certifcates 38
Surgical S1 thin colbalt-colored mask, no wrap 1
S2 thick with black color in the outside, individually wrapped, marked as surgical mask 2
S3 thick with black color in the outside, invidually wrapped, sold in a box of five, marked as generic mask 3
Airmask A1 thin, white-colored, marked as made in Japan, indicated for filtering dust and smog, sold in a box of five 3
A2N medium-thick, white-colored, branded as PM2.5, indicated for filtering PM2.5, sold in a box fo five 5
A2U similar to A2N, a used PM2.5 mask in use for 1 month (infrequent) 5
3M brand 3M1 model 9001v with exhale valve, made in China, KN90 grade 10
3M2 model Aura 9332+ with exhale valve, made in UK, FFP3 grade 20
3Mo 1-month in use, model 9001v 10
AQBlue AQ1 AQBlue, gifted from AQBlue rep., June 1, 2019, no wrap, made in Vietnam 49
AQ2 online order, individual wrap with Moto on the package, made in Vietnam 35
Fig. 2: Gallery of masks tested in this study..

After testing, the masks were cut opened and photographed using a USB-microscopic webcam advertised to have up to 1600X magnification.

PMs data from PMS7003 in the active mode were selected with the atmospheric condition as recommended by the manufacturer and no coefficient factor was applied. The interval of measurement is one-minute. Data were then transmitted wirelessly to a home server using the MQTT protocal. Analyzing data were performed on Jupyte Notebook editor with pandas library for data processing and Matplotlib for data visualization.

Removal efficiency (RE) is the primary parameter to evaluate the effectiveness of each mask. RE is calculated by taking the difference of PM2.5 or PM10 between the background concentration (BC) and filtered air concentration (FC) and finally divided to the BC.

RE of each mask was averaged over the testing period and deducted from the difference during first 3-hour windows of cross checking if significant difference was found. Snapshot analysis followed by RE was varied by each fan duty cycle or different between PM2.5 and PM10.

3. Results and Discussion
3.1 Data Cleanup

Over a period of 45 days, about 65,000+ data points were collected by each sensor. The data were cleaned up by removing single peaks using pandas libraries. After 10 rounds, 3.4% of the total rows were removed from the dataset with mask (referred as the mask data), and 1.7% of the total rows were removed from the dataset of the background concentration (referred as the background data). Fig 1 shows the data before and after clean up.

Fig. 3: Data cleaning/wrangle of the mask and background datasets. A lapse in data with mask shows the experiment with mask was suspended.

The PMS7003 with 1-minute interval sampling yielded large datasets. Single peaks were shown in Fig. 4 indicated sudden busts in concentration with no clear reason behind. The author accepted that the peaks as the artifact of low-cost sensors and decided to remove those peaks before analyzing the effectiveness of mask to screen out PMs.

3.2 Removal efficiency (RE)

After peaks were removed, the cleaned dataset was segmented into a pair for each experiments. One included a background and a filtering period. The background period lasted for 90 to 120 minutes to cross check the PM sensor for monitoring filtered air with the PM sensor monitoring the ambient air, also called background PM concentration.

The removal efficiency (RE) was as the portion of PM filtered out to the total PM contained in incomming air. Each of the graps below presented the average RE, RE calculated for the crosscheck windows, and of PM2.5 and PM10 during the experiments.The results are presented in Figs. 4-7. For the boxplot graph, each boxplot marked the medium, 25th, 75th percentile and the outliers that were processed by Seaborn library on the top of Matplotlib of Python package.

Summary of PM2.5 Removal of all tested masks presented as the average values taking all values into consideration. Details of each test presented belows with built-in software function to excluse outliers. The result is a post-analysis of this one and follow-up experiments.

Fig. 4: Removal efficiency during 3-hour crosscheck before experiments. The upper and lower of the box represent the 25th and 75th percentile with the cross inside the box is for 50th percentile (median). The lower and upper whiskers represent the minimum and maxium of the set after removing outliners which are represented by the dots.
Fig. 5: Removal efficiency of PM2.5 by each mask.
Fig. 6: Removal efficiency of PM10 by each mask.

Details of the removal during crosscheck, of PM2.5 and PM10 with fan duty are listed in Table 2. Fan duty indicates the speed of the fan which 1 as on all the time, and 0.5 is 50% on. The fan duty was used as the surrogate for the air flow through one mask.

Removal during the crosscheck is closed to zero as shown Fig. 4 and Table 2 confirmed that the setup is adequate to compare the PMs concentration of the filtered air and the background. Because the numbers are closed to zero and much smaller than the standard deviation, the RE of PMs was reported without deducting the difference during the crosschecks.

Table 2: Removal effiency during the cross check, and of PM2.5 and PM10 including all data (not excluding the outliners as shown by Seaborn boxplot).
Group ID Crosscheck PM2.5 PM10 Fan Duty
Fabric FU1 -0.01 0.13±0.07 0.14±0.08 0.58
FU2 test #1 0.01 0.09±0.06 0.08±0.05 0.63
FU2 test #2 0.01 0.15±0.09 0.12±0.01 0.51
FN 0.00 0.11±0.06 0.12±0.07 0.53
Surgical S1 -0.02 0.29±0.12 0.29±0.13 0.58
S2 test #1 0.01 0.29±0.13 0.34±0.14 0.51
S2 test #2 0.03 0.33±0.06 0.33±0.08 0.55
S3 0.02 0.37±0.13 0.39±0.10 0.62
Air mask A1 -0.03 0.03±0.06 0.03±0.08 0.53
A2N 0.02 0.41±0.09 0.40±0.10 0.58
A2U 0.04 0.37±0.10 0.39±0.11 0.61
3M brand 3M1 0.03 0.47±0.08 0.52±0.07 0.56
3M2 test #1 0.01 0.57±0.09 0.59±0.09 0.60
3M2 test #2 0.01 0.56±0.09 0.62±0.07 0.62
3M old N/A 0.51±0.09 0.52±0.09 0.56
AQBlue AQ1 0.07 0.94±0.04 0.94±0.04 0.60
AQ2 0.09 0.92±0.05 0.92±0.04 0.62
3.3 Mask photographs

The experiment setup with a mask mounted on a PVC pile limits the scope of testing to the filter efficiency of the mask's material. The construction of each mask was presented on Figs. 7-10 with an USB-microscopic webcam.

Fig. 7: Photograph of fabric masks with a phone camera followed by two photo taken from a microcopic webcam. A long in black object is a string of hair for the scale. Human hair has a diameter of 60-70 µm.
Fig. 8: Photograph of surgical masks.
Fig. 9: Photograph of airmasks.
Fig. 10: Photograph of 3M brand particulate respirators.
Fig. 11: Photograph of AQBlue particulate respirators.

The construction of the farbic masks (F1U and FN) includes 3 layers with a cotton-like cloth in the outer layer, a thick polyester layer in the middle and the support layer in the inner layer. The middle layer is loosely packed with made of the thickness of the masks.

Surgical masks are 4-5 layers packed tightly. One middle layer is made of tiny fibers and densely packed. Two out the experiments.The results are presented in Figs. 4-6. Each boxplot marked the medium, 25th, 75th percentile and the outliers that were processed by Seaborn library on the top of Matplotlib of Python package.

In two masks purposedly against particulate pollution, A1 is constructed with a single layer with randomly woven polyester. The size of pore compared to the hair string is not much different. Meanwhile, PM2.5 is about 30 times smaller. This visual supports the measurement on Table 2 as no effective to filter out PMs. The other mask (A2) is constructed with 5 layers with very fine fibers as shown in the bottom row in Fig. 9.

Brand-name particulate respirators, 3M 9001 with KN90 (3M1) and 3M Aura 9332+, are composed of 5 layers. The later model has the middle layer thick and pillow-like shape. The former model has two layers in the middle is composed of very fine fibers.

3.4. Airlow rate and RE

A mask is an equipment that is very specific in use; however, the working principle is as a filter. When airflow moves into the mask, the friction the mask layer creates turbulence around the pores. A higher airflow is translated to a shorter contact time with materials and a lower probability that the particles to be adsorb onto the fiber matrix. For the mechanism of filtering, please refer to this link for more information.

Figs. 12-15 presents a snapshot analysis for 4 candidates of each type. Those graphs provide detailed visuals. The correlation of the fan duty to RE is presented in Fig. 15. The author used the fan duty as the surrogate for flow rate indicator.

Fan Fabric
Fig. 12: Snapshot analysis with a fabric mask (FU2).
Fan Surgical
Fig. 13: Snapshot analysis with a surgical mask (S3).
Fan PM2.5
Fig. 14: Snapshot analysis with an air mask (A2N).
Fan 3M
Fig. 15: Snapshot analysis with a particulate respirator (3M2).

The results in Fig. 16 is consistent with the RE in Table 2. The fabric mask is porous to PMs that leads to marginal RE. The RE is not correlated with airlow because the friction is minimal. With surgical and air masks, a negative correlation indicates at a higher flow rate, a lower RE of PMs. The 3M Aura with FFP3 standard shows a correlation of -0.51, which is in line with the two above but the effect of airflow to the RE is smaller. This suggests improvements in 3M mask that allows a higher airlow with a less nagative on RE.

Fan to RE
Fig. 16: Correlation of fan duty to PMs removal efficiency. The correlation factor was calulcated using Pandas built-in function and presented on the top of each graph.

The problem associated with PMs pollution is not singular to Vietnam. Other countries such as China and India have been experienced unhealthy levels. Exisiting blog posts and journal research are relevant to users in Vietnam to understand the basic and consider the recommendations. For example, Yu at el., 2014 found that the fitting of standard N95 mask to Chinese works is "poor". Informative blogs posts on the testing methods and performance of masks are useful for the "citizen science" approach.

3.5. Limitations

In a research journal, the limitation section is not included. Nevertheless, the goal of this study is to inform the mask users and some limitation the author should spell out.

  • Wearing a mask on the face is different than having a flat layer mounted on the PVC pile. The latter applies to this study. Fitting factor, in plain English, is how the mask fitted on the user's face to prevent leakage or shortcut of airflow. The standard tests included fitting a mask on the face and measure the number of particles of certain size inside and outside the mask. Therefore, the results in Table 2 represent the maximum removal efficiency in the testing conditions in this study.
  • Not all the mask are designed with the same forms. Surgical masks are better to remove PMs but the form is not designed for a close space around the nose. Short airflow from the sides of mask renders the RE to almost none with the ambient air. The fabric masks are better to curse around the nose, but the materials are not adequate to protect users against PMs. This emphasizes the first points. Even with a well-designed mask, how user's face features and actual wearing affects the final effiency of removal, which is out of the scope in this study.
  • The results in this study neither support nor discard the rating by the manufacturer because of different testing conditions, equipment and procedures were applied. The author prefer a testing condition simulated the on-road condition rather with standard reagents. Choosing a brand name and wear the mask properly is one sentence summary.
  • 4. Conclusion

    During a 55-days of evaluating 11 masks for the effectiveness of filtering out PM2.5 and PM10 from the ambient air, the findings indicated the fabric masks which is populuos on Hanoi street is the least effective to remove the PMs, ranging from 8-14%, the surgical masks have moderate effectiveness from 29-39%. The masks marketed for protecting health against PMs is hit and miss which one type can remove 37-52% and the other barely shows the difference. Brand-name masks such as 3M show the highest removal rate ranging from 47% to 62%, and AQBlue is from 83% to 87%. Additional analysis on 4 types indicated that fabric mask is porous and the headloss is marginal. A correlation of airflow to removal efficiency from -0.76 to -0.89 suggested a lower RE to a higher flow rate. The correlation coefficiency of -0.51 for 3M Aura indicated improvements to allow a higher airflow without significantly reduced RE of the 3M mask. The form fitting is not evaluated in this experiment. In addition, this experiment filtered air in the 10th floor with light traffic on the ground only represents an approximate condition on the street.

    5. Acknowledgement

    The author would like to thank Dr. Han Huy-Dung for lending the PMS7003 sensors. Two friends suggested that the author should test air masks and for that additional push tipped the balance, for that the author thanks them for the extra tick. Last but not least, the author would like to thank his wife (H.N) for willingly gave up two-used air masks and switched to a brand name respirator. She also gave up a large portion of the balcony to host the equipment and forgave the author not doing a day job in order to follow his priority, for that, the author thanks her for her consideration.

    ▣ ▣ ▣