Design of a Medium Scale Ambient PM 2.5 Cleaning System

High concentrations of ambient particulate matter (PM) have caused millions of premature deaths annually worldwide. The source control strategy was normally implemented to bring the PM pollution down to meet the standards. However, it took more than 50 years for UK and US. As PM pollution is hurting people’s health on daily basis in many countries, a fast and inexpensive control technology should be developed to remedy the slow pace of source control. It should be applied in highly polluted areas, e


INTRODUCTION
Since the end of 2019, the whole world has been impacted by the COVID-19 pandemic. A total of 6.6 million people died in the past 3 years from this acute coronavirus (SARS-CoV-2) disease (Johns Hopkins, 2022). The acute effect causing severe illness and high mortality rate shocked and scared everyone. With the help of effective vaccines, the daily deaths were significantly reduced from a peak of ~16,000 globally in January 2021 to around 1,500 in November 2022. It is foreseeable a further reduction of mortality in the coming months.
However, the mortality due to long-term exposure to PM was almost forgotten in the past couple of years due to the improved air quality resulting from the reduction of industrial manufacturing (Chen et al., 2020a;Yin et al., 2021). It was the consequence of the disease transmission and mitigation measures that have been implemented in many countries. However, after the recovery of the global economy, industrial and motor vehicle emissions stroke back which could cause a more severe impact than COVID-19, thus attention should be paid seriously.
Just a decade ago, the International Agency for Research on Cancer (IARC) placed outdoor particulate matter (PM) in Group 1 carcinogen in 2013. In the PM, PM10 (particulate matter smaller than 10 µm in aerodynamic diameter) and PM2.5 (particulate matter smaller than 2.5 µm in aerodynamic diameter) are more inhalable and easier to get into the lower respiratory system and deposit in the alveolar area compared to larger size fractions (Hinds, 1999). Therefore, it is considered that the major contributors to carcinogenicity are PM2.5 and PM10 (Chen et al., 2022;Pui et al., 2014). It has been reported that PM2.5 pollution caused 1.8 and 0.67 million premature deaths in China and India, respectively, in the year 2017 . The OECD projected that premature death may reach 6-9 million globally in 2060 if the PM and ozone pollutions are not significantly mitigated. It costs 1% of the global GDP, USD 2.6 trillion annually, as a result of sick days, medical bills, and reduced agricultural output (OECD, 2016). The economic loss on average is about $350,000 per premature death.
Reducing PM10 and PM2.5 to meet or be close to the current air quality standards by source control has been achieved successfully in many developed counties, including European counties and the U.S. (WAQI, 2022). However, the fact is that it took more than 5-6 decades, such as in the U.S. from the 1940s to the early 2000s, to attain the current 24-hour average standard, 35 and 150 µg m -3 for PM2.5 and PM10, respectively (Federal Register, 1997, 2006, 2013. As the extremely high mortality rate of PM causes tragedy and tremendous economic loss (OECD, 2016), not only should strengthening the standards from source control be implemented but also area control is urgently needed in the regions where the air is still highly polluted.
To mitigate high PM in a densely populated area to save many lives and reduce huge economic loss, a solar-assisted large-scale cleaning system (SALSCS, 1 st generation, Fig. 1(a)), consisting of a set of flat-plate solar collectors, a tower, and filter banks was proposed (Cao et al., 2015). A 1:1000 scale SALSCS demonstration unit with a 60 m × 45 m × 50 m dimension was built in Xi'an, China in 2016, as shown in Fig. 1(a) (Cao et al., 2018b). Field measurements were in good agreement with the computational fluid dynamic (CFD) simulation on the natural convective flow produced by this unit, which was 200 m 3 s -1 during sunny daytime in summer. At this flow rate, the filtration system's PM2.5 removal efficiency was ~80%. The clean air delivery rate (CADR = flow rate multiplied by filtration efficiency) was 160 CMS equivalent to more than 1000 home purifiers (with ~0.1-0.2 CMS CADR). For a more sustainable and environmentally friendly consideration, the 2 nd generation water spray-based SALSCS was designed and built at Yancheng Environmental Science Park, China in 2017, as shown in Fig. 1(b). The removal mechanism of PM2.5 was by sprayed water droplets, an analogy of rainfall scavenging (Chen et al., 2020b). This unit has a dimension of 18 m × 18 m × 16 m and about 20 CMS CADR.
The above two demo units applying filtration and wet scavenging mechanisms to remove PM2.5 have been experimentally evaluated for their CADRs, i.e., flow rate and filtration efficiency. Good agreements with the original designs based on theoretical calculations and CFD simulations (Cao et al., 2015(Cao et al., , 2018a(Cao et al., , 2018bChen et al., 2020b) were obtained indicating the accuracy of the CFD simulations. Based on the numerical model, it is found the deployment area of the solar collector as well as the site area should be largely increased for generating a sufficient amount of the Natural convective flow for both 1 st and 2 nd Gen SALSCS. However, land acquisition in a city area is very difficult and the cost is very high. To keep the system size reasonably small, e.g., around 20 m × 20 m base and 20 m tower height, but to largely enhance the CADR to benefit more individuals nearby the cleaning system, a compromised plan is to apply air blowers to produce a large flow, e.g., 1000 m 3 s -1 . Since the blowers are used, the flow is preferably reversed and directly supplying clean air to people having activities nearby. Under a high flow rate, the removal of PM by filtration is favorable because it allows a relatively shorter residence time than the droplet scavenging and other airborne particle removal mechanisms, e.g., electrostatic precipitation and centrifugal cyclone.
PM2.5, the fine-size fractions of PM10 are more difficult to be removed by filtration. Therefore, the effectiveness of PM2.5 removal is majorly shown and discussed throughout the article as it represents the worst scenario of PM treatment by the cleaning system. In this study, a blower-driven filtration-based 3 rd generation cleaning system (the 3 rd Gen) is proposed and designed to directly supply clean air to pedestrians, workers, students, commuters, etc., who are walking to school, office, or market, waiting for a bus or train, and riding a bike or motorcycle in a busy city. The effectiveness and performance of the 3 rd Gen in terms of PM2.5 reduced coverage area and CADR will be predicted numerically by CFD. The filtration system having high efficiency, low-pressure drop, and long service life will be designed based on experiments and theory. The ultimate goal is to minimize the exposure of city residents to PM2.5 and PM10 to save lives. Fig. 2 shows the 3D schematic diagram ( Fig. 2(a)) and 2D top view ( Fig. 2(b)) of the 3 rd Gen. Be noted, the shape of the main body and tower of the system could be square or circular, but the former is easier to be built. Therefore, the following simulations will be based on square geometry. The polluted air (or the ambient air) is dragged into the tower by blowers, purified by the filter bank, and then blown out of the system to benefit the people nearby. The arrangement of the filter bank and blower array in order following the flow direction was to prevent the dust resuspension of deposited PM on the filter. Besides, dragging flow through the filter bank can also reduce the pressure drop due to a more uniform flow field. The most suitable (or lowest) height of the tower based on the minimized flow recirculation will be predicted by CFD simulation.

Main Body of the 3 rd Generation Cleaning System
Minimizing the pressure drop of the cleaning system is very important as it can largely affect the performance of the blower. Normally, tube axial fans are used to provide high flow in industrial applications. There are several ways to reduce the pressure drop of the 3 rd Gen, including designing a bellmouth inlet on the top of the tower, choosing a bird screen with a proper mesh size, smoothing the internal surface of the system, making flow straighteners between the 90-degree bend inside the tower, choosing filters with low pressure drop but high efficiency and holding capacity, etc.
With a side length of 20 m, an outlet height of 4 m, and a desired 4 m s -1 flow speed (~9 mile hr -1 ), the output flow of the 3 rd Gen is 1,280 m 3 s -1 , and the CADR will be around 1000 CMS with a filtration efficiency of 80%. The CARD of the 3 rd Gen is around 6 times the 1 st demo units (50 times of 2 nd Gen) and saves the land size by a factor of 7, resulting in a ~40 times effectiveness.

Filter bank and frame
To accommodate the extremely high flow rate through the filters with limited space, a special design of the filtration system is needed. As shown in Fig. 2 and Fig. 3, the filter bank was designed in a W-shape formed by multiple metallic frames (e.g., aluminum or stainless steel), which was an analogy of pleated filters. The included angle of 2 adjacent frames was ~12°, therefore the total surface area of the W-shape frames was increased by ~10 times compared to the flat one (= 1/sin 6°). The choice of 12° not only can largely increase the filtration area but also remain a wide space (0.4 m) for one to have access to the filters for maintenance and inspection purposes. The frames should be well-sealed on top and bottom with the ceiling and ground for preventing infiltration. In each frame, grooves, rails, and weather stripping are made for fitting and sealing filters. With the increase of area for installing 10 times more filters, the face velocity of the flow through the filters can be reduced by 10 times to enhance the removal efficiency of PM2.5. If pleated filters are also used, such as residential HVAC filters, the flow speed through the filter media can be further reduced by 5-10 times. For example, the media face velocity will be reduced from ~5 m s -1 (smaller cross-sectional area at the filter bank than the outlet of the tower) to 0.05-0.1 m s -1 when applying the W-shape frame and pleated filters compared to that of a flat filter perpendicular to the flow.

Filter media
The atmospheric PM often shows a bimodal in mass and trimodal in number size distribution covering 4-5 orders of particle sizes, e.g., 0.002-100 µm (Chen et al., 2010a(Chen et al., , 2010b(Chen et al., , 2011Pui et al., 2014). A series of a prefilter to remove coarse PM and a final filter to take out the fine PM is normally applied. Chen et al. (2016) challenged electret (respirator) and mechanical (engine intake) filter media with bimodal PM (mass median aerodynamic diameters, MMADs, at ~0.5 and ~6 µm). Results showed that the mechanical engine intake filters had a low efficiency for both mass-and number-based PM2.5, which was only about 25-30%. There was a much lower numberbased PM10 efficiency than that of mass-based due to residing of PM mass in the coarse size range. In comparison, the charged respirator filter media could effectively remove both PM2.5 and PM10, where the mass efficiency was always higher than that of the number. The study provided insights into the performances and characteristics of mechanical and electret filter media against atmospheric PM10 and PM2.5. Due to the add-on electrostatic attraction, electret media had a much higher efficiency, but the same pressure drop compared to mechanical media with the same fiber diameter, thickness, and porosity. Chang et al. (2019) investigated the decline of initial efficiency and particle holding capacity of electret composite filter media consisting of a coarse fiber layer and a fine fiber layer. Mechanical  glass fiber and PTFE media were also tested and compared with the composite media. In the loading, particles mimicked the distribution of ambient PM2.5 were produced and used to challenge these media. The effect of the concentration of the challenging PM2.5 on the holding capacity was also evaluated. The results showed that the efficiency decline can be minimized for the electret composite media if the fine fiber layer has a sufficiently high initial efficiency, such as using electret melt-blown (charging density > 45 µC m -2 ) with fiber diameters of 3-4 µm and ~15 g m -2 basic weight. This was because the loading effect enhanced the filtration efficiency when the efficiency declined due to the loss of fiber charge by the particle shielding effect. The holding capacity from great to least was the electret composite media, glass fiber, and PTFE. The holding capacity of the composite media was about 6 and 2 times of the PTFE and glass fiber media, respectively. Besides, there was a significant effect of PM concentration where the holding capacity was reduced with increasing particle concentration. A relatively low PM concentration close to the ambient PM in the loading test should be applied, such as < 3 mg m -3 . Tien et al. (2020) further investigated the reasons causing the enhancement of holding capacity by the electret media. They found that due to the loss of fiber charge along the loading, the lower layer of media had a chance to expose to a higher concentration of incoming particles. Therefore, the lower layer fibers still containing charges have a higher efficiency and could collect more particles than the upper layers that have lost charges. This phenomenon occurs successively, resulting in a more uniform particle deposition in depth. Therefore, the same filter media with charge in general has a higher holding capacity than the discharged one. Besides, they found that the upstream to downstream of the composite media should consist of the layers having a holding capacity from high to low but efficiency from low to high to maximize the holding capacity. Li et al. (2020) investigated the effect of relative humidity and hydrophilicity of challenging particles on the initial efficiency and holding capacity of electret filter media that are often used in ambient environments. In the initial efficiency test, the electret media were evaluated at 10, 60 and 90% RH. In the loading test, the media performances were characterized at 30, 60 and 80% RH, and challenged by hygroscopic NaCl, non-hygroscopic SiC, and NaCl + SiC mixed particles. The results showed the initial efficiency was independent of the RH. In the aspect of loading, it was found operating the filtration at ~60% RH can greatly enhance the particle holding capacity. Electret filter media operated in the ambient environment could experience high RH during the night due to the reduction of temperature, so the pressure drop of the filter media could reduce by the collapse of dendrites and thus perform a certain level of filter regeneration. However, to quantitatively determine the effect of RH on filter regeneration, future study is needed.
From above, a composite electret filter media consisting of a coarse fiber layer and a fine fiber layer should be applied in the 3 rd Gen to be the final filter. Prefilters are still needed to remove coarse dust to protect the final filters which are mainly collecting PM2.5. Very importantly, it is desired that both filters are electret to enhance the initial efficiency and holding capacity to extend the service life of the filters. In this study, several specially designed filter media were examined for their initial efficiency and PM2.5 holding capacity under 10 cm s -1 face velocity. The experimental system same as that shown by Tien et al. (2020) was applied.

PM Concentration Reduction Zone Simulation
To mitigate PM2.5 for a large area, i.e., Beijing terrain of ~1400 km 2 (within the sixth ring road), eight 1 st Gen each having a diameter of 5 km were proposed (Cao et al., 2015). By the CFD simulation, it was found the natural convective flow rate for each unit could be as high as 2.6 × 10 5 and 3.8 × 10 5 m 3 s -1 under the solar heat flux of 445 and 640 W m -2 , respectively. The weather research and forecasting (WRF) model was applied to consider the effect of weather conditions of historic PM2.5 concentration and wind speed and direction on the PM2.5 reduction and clean air transport from the eight 1 st Gen. Passive tracer scalars were used to simulate PM2.5 pollutant and SALSCS clean air. Air pollution reduction percentages of SALSCS over the urban region were obtained during seven PM2.5 pollution episodes. The results showed that the average PM2.5 reduction percentage is 12 and 15% under the convective flow of 2.6 × 10 5 and 3.8 × 10 5 m 3 s -1 , respectively. For the current 3 rd Gen, the same method can be applied to determine the PM2.5 reduction percentage and covering zone when multiple cleaning systems and a large area are considered. The WRF can be combined with the large eddy simulation (LES) which is the intermediate between the direct simulation of turbulent flows and the solution of the Reynolds-averaged equations.
For predicting the effectiveness of a single system to clean a small area, e.g., 1~2 km 2 , the CFD simulation without applying WRF and large eddy simulation is sufficient (Cao et al., 2018a(Cao et al., , 2018b. In this study, the 3D incompressible Reynolds-Averaged Navier Stokes (RANS) equations and κ-ε 2-equation turbulent model were adopted to uncover the flow field. The mixture of polluted ambient air and clean air from the cleaning system was assumed. Besides, the species transport equation was solved for the polluted ambient air. In addition, the effect of tower height for the 3 rd Gen, i.e., 15 and 23 m, on the air recirculation rate was examined. The background weather conditions, i.e., ambient flow direction and speed, according to historic data were also considered in the simulations.
The boundary conditions for the atmospheric inlet and outlet of the computational domain were velocity inlet and pressure outlet, respectively. A fan boundary condition with a constant pressure drop was applied to drive the system's airflow rate to 1000 m 3 s -1 . The top boundary of the atmospheric domain was chosen to be a shear-free wall. All the remaining wall boundaries in the domain were defined as no-slip wall boundary conditions.
The ANSYS Fluent 17.0 fluid solver (ANSYS, 2016) employs the finite volume method for solving the governing equations discussed above. The pressure-based solver was applied for the incompressible flow. The SIMPLE algorithm was chosen as the pressure-velocity coupling scheme, and the body-force-weighted algorithm was selected as the pressure interpolation scheme. Advection and diffusion terms in the governing equations were discretized by the second-order scheme. A grid-independent study was performed to determine whether the grid number would affect the numerical results. Three simulations with total grid numbers of around 2,000,000, 4,000,000, and 6,000,000 were tested, respectively. It was found the discrepancy in the simulated air pollution concentration fields amongst the three cases was smaller than 3%, indicating that our model had an excellent grid-independence performance. Thus, the grid number of 4,000,000 was chosen for the current study. The absolute convergence criteria were set to be 10 -6 for all the variables to monitor the solution convergent performance of the numerical simulations.

System Pressure Drop and Blower
With a design of tower width of 8 m × 8 m, the Reynolds number under 1000 m 3 s -1 system flow was about 8.3 × 10 8 , resulting in a turbulent flow. Besides, with the assumption of bellmouth inlet loss coefficient of 0.12, bird screen with 0.5-inch mesh, inlet 90-degree bend loss coefficient of 0.18, tower surface roughness of 0.3 cm (concrete), flow straightener loss coefficient of 0.16, fan diffuser loss coefficient of 0.2 and discharge loss coefficient of 1, the pressure drop of the bellmouth, bird screen, top 90-degree bend, internal walls, flow straighteners, fan exit diffuser, and discharge is about 18, 75, 26, 4, 23, 19, and 90 Pa, respectively, a total of 255 Pa (~ 1 in-H2O). The pressure drops from each component and the assumptions of their loss coefficients (K) and physical properties (friction factor, f, or mesh size) are summarized in Table 1. To supply ~1000 m 3 s -1 flow, 40-50 tube axial blowers with ~40 HP are needed. The blower with this size normally can have about 95, 85 and 65% of fan efficiency under 250 (system pressure drop), 500 and 750 Pa static pressure, respectively. Therefore, the allowable pressure drop contributed from the filter bank and PM loading can be as high as ~500 Pa (2 in-H2O). Thus, the initial CADR will be around 800 CMS (1000 m 3 s -1 flow rate × 80% efficiency), and the CADR during loading can be remained as high as 765 (850 m 3 s -1 flow rate × 90% efficiency) and 650 (650 m 3 s -1 flow rate × 100% efficiency) under 1 and 2 in-H2O pressure drop, respectively. The pressure drop of the cleaning system can be further reduced by smoothing the surfaces, making a better flow straightener, enlarging the mesh size of the bird screen, etc. More importantly, a good design of the filtration system will result in the success of this study.

Filter Media
There were more than 20 specially designed filter media tested individually or combined as composite filters for their dynamic efficiency along the loading. In addition to the efficiency, PM2.5 holding capacities were also investigated. Fig. 4 shows the initial efficiency and pressure drop ( Fig. 4(a)) and the worst efficiency ( Fig. 4(b)) and holding capacity (Fig. 4(c)) of the investigated media during PM2.5 loading. The media A, B and C shown in the figure represent the grades of typical electret HVAC filter media with the minimum efficiency reporting values of [11][12][13][14]. The Coarse and 1.5 Coarse in Fig. 4 are electret coarse fiber layers with a thickness of 1.0 and 1.5 mm, respectively, and fiber diameters of ~25 µm. Several composite media combining the coarse fiber layer and HVAC media were also tested and the results are shown in Fig. 4. It is seen that media A, B, C, and the Coarse fiber layer all have a certain level of efficiency decline during the loading, while the decline is negligible for the 1.5 Coarse and three composite media. The composite electret filter media not only have acceptable initial efficiencies with a 70-80% but also negligible efficiency decline, and low pressure drops . To be mentioned, filter media with extremely high initial efficiency such as the HEPA filters should not be considered as their pressure drops are high and holding capacities normally are very low . Fig. 4(c) compares the holding capacities amongst individual and composite filter media. Obviously, the holding capacity was improved by combining the coarse fiber layer and MERV 11-14 media, Fig. 4. Comparison of (a) initial efficiency and pressure drop, (b) the lowest efficiency, and (c) holding capacity during the PM2.5 loading of filter media studied in this study. except for the 1.5 Coarse layers. The extremely high capacity for the 1.5 Coarse layer was due to its relatively low initial efficiency and more void spaces for particle loading. The final holding capacity, at 250 Pa pressure drop, of the 1.5 Coarse+A is also very good, which was 35 g m -2 . Compared to the holding capacity of 5-7 g m -2 by the composite media, 1 g m -2 by PTFE and 4 g m -2 by glass fiber reported in the literature (Tang et al., 2018;Chang et al., 2019), the capacity of the media developed in this study are much higher (factors of 5-35). The above data were obtained under the test condition of ~30% RH and it has been found the capacity can increase by more than 4-5 times if the operation RH is 60-80% . Therefore, the capacity of the 1.5 Coarse+A can be as high as 140-175 g m -2 . Under an average PM2.5 concentration of 300 µg m -3 , the composite filter can serve about 6 months until the next replacement.
Very importantly, before the composite filters, prefilters should be used to remove ambient dust larger than 2.5 µm to protect the main filter. We have also tested the prefilter in an ASHRAE 52.1 wind tunnel using A2 fine dust under 0.5 m s -1 face velocity assuming it was a flat sheet filter. The results showed that with a relatively low pressure drop, i.e., 50 Pa, the prefilter can hold more than 100 g m -2 dust. It is estimated the replacement cycle of the prefilter is around ~3 months when the concentration of total suspended particulate (TSP) is as high as 600 µg m -3 . To conclude, the design of the filtration system should be very adaptive depending on the locations and the areas to be cleaned as the weather conditions and particle size distributions are varying from place to place.

PM2.5 Reduction Zone
To investigate the performance of the 3 rd Gen in a realistic condition, this study assumed the system was sitting in a space with three large buildings surrounded as shown in Fig. 5. The three buildings, assuming the same height of 18 m, could be the complex of government offices, school, or transportation stations (train, bus, and subway), etc. We intentionally sat the system close to one of the buildings to see the effect of obstacles on clean air delivery. Figs. 5(a) and 5(b) show the simulation domain (1 km × 2 km × 0.6 km) and the PM2.5 reduction contour under different wind directions and a constant wind speed of 0.5 m s -1 (an average wind speed in highly polluted cities in wintertime). It is seen PM2.5 was reduced by more than 50% in the whole space (1.5 m height) within the three buildings under all four wind directions, indicating people are protected in this space. This space is as large as 250 m × 300 m, allowing thousands of people to enjoy the cleaned air. It is also observed the buildings could block the transport of cleaned air, nevertheless, which could overflow to the outside of the complex through the opening. Be noted from the Harvard Six City study (Dockery et al., 1993;Pope et al., 1995), a reduction of 30 and 50% of PM2.5  can reduce the mortality rate by 8 and 17%, respectively. The currently designed 3 rd Gen is expected to largely reduce the adverse health effects for the city residents.
To investigate the performance of the 3 rd Gen in a more crowded city area, we assumed the system was placed in a commercial and residential area as shown in Fig. 6(a). The heights of the main surrounding buildings were between 4 and 30 m tall. Fig. 6(b) is the computational domain in the CFD simulation. Similarly, the PM2.5 concentration distributions were calculated under four wind directions and a constant wind speed of 0.5 m s -1 .
The simulation result of the PM2.5 contour shown in Fig. 7(a) indicates the area being cleaned was restricted to a relatively smaller area compared to that shown in Fig. 5. This was due to the shorter distances between the 3 rd Gen and the surrounding buildings in this case. To investigate the effect of the reduced distance between the system and the surrounding buildings on the PM2.5 reduction, the profiles of the PM2.5 reduction ratio in the streamwise and spanwise plane across the center of the system are shown in Figs. 7(b) and 7(c), respectively. Fig. 7(b) indicates   that the adjacent buildings of the system trapped the cleaner air to remain at a low PM2.5. It is seen the PM2.5 concentration increases (reduction percentage reduces) along the distance of the system but it reversely reduces near the building. This was due to clean air recirculation. Outside the nearest buildings, cleaned air could hardly be delivered due to the building blockage. In the spanwise direction shown in Fig. 7(c), cleaned air can transport to the building outside because the flow was not totally blocked. The results shown above reveal that the selection of the site is very important, and obstacles can easily block the delivery of clean air.
To determine what height of the tower is needed to reduce the clean air from the system outlets re-entraining back to the system tower due to air re-circulation, the tracer at the outlets was released and the quantity of the tracer in the tower was examined. In the simulation, the site shown in Fig. 5 was adopted and the tower heights of 15 and 23 m, two ambient wind speeds of 0.5 and 1.0 m s -1 , and two wind directions of southwest and northeast were considered. Fig. 8 shows the simulation results of the 23 m ( Fig. 8(a)) and 15 m ( Fig. 8(b)) tower height under a wind speed of 0.5 m s -1 in the southwest direction. It is seen the re-entrainment increases with reducing tower height. Table 2 summarizes the results of the eight conditions simulated. In general, it is found that re-entrainments were lower than 10% for the 23 m height and it was close to 20% for the 15 m height. The re-entrainment was reduced with increasing ambient wind speed. The above results imply that there was a certain amount of clean air dispersing to the higher layer of air when leaving the outlets, therefore horizontal guide plates at the system outlet are needed to direct and constrain the clean air along the ground to a farther distance. Further CFD simulations are needed to improve the clean air delivery distance.

An Array of 3 rd Gen
This study also applied the LES method to calculate the effectiveness of building 400 units of the 3 rd Gen with 1 km apart on the reduction of PM2.5 concentration. They were uniformly placed on a 400 km 2 (20 km × 20 km) space and operated at the flow rate of 1000 m 3 s -1 with an ambient wind speed of 0.5 m s -1 . The results showed that PM2.5 concentration can be reduced by 65% up to a distance of 700 m in the downwind direction. It is estimated the PM2.5 concentration can be reduced by 60% under the height of 10 m in the whole 400 km 2 .
As discussed in the Introduction section, the OECD estimates the loss of every premature death is about $350,000. The population in a 400 km 2 area can be from 2 to 5 million in densely populated cities. Assuming the cost to build one unit is $2M, the cost per capita is only $200 for building 400 units in a city with a 4 million population. The above calculation has not included the costs of energy, filters and replacements, system maintenance, waste treatment, etc. Nevertheless, it has illustrated that comparing the loss of death, purifying the city air by building the 3 rd Gen to save lives is very cost-effective and necessary.

Optimization of 3rd Gen
The first simulation that should be explored is to see the effect of the guide plate on improving clean air delivery. If fences are used to restrict the general public to enter the area of the system, it is required to evaluate the effects of the fences on blocking air delivery. A possible way to eliminate the effect is to elevate the base of the system to allow free clean air delivery. As discussed in Section 3.3, the delivery of clean air can also easily be stopped by buildings and small obstacles, and an alternative plan is to further reduce the size of the 3 rd Gen to 5 m × 5m. They can be placed at intersections, bus stops, school playgrounds, sports fields, etc. Besides, the system geometry can be changed to be a triangle or circle or reduce the number of outlets according to the condition and position of the sites. The effect of fugitive PM emissions in the treatment area on PM reduction by the cleaning system should be also studied in the future.
In the aspect of operation costs, further reducing the pressure loss of the system is needed. To improve the sustainability of the cleaning system, washable filters may be considered. However, not only the economic analysis of the costs for the labor and water usage to clean the filters but also the durability and reliability of the washable filters should be determined and compared with the disposable filters.

CONCLUSIONS
This study proposed an outdoor cleaning system, named 3 rd Gen, to quickly reduce PM2.5 and PM10 concentration in a highly polluted and densely populated city to remedy the slow pace by source control, which usually takes several decades to meet the air quality standard. This system is a large blower-driven filter-based air cleaner and blows clean air directly to the ground level to benefit people who are walking, jogging, or riding a bike on streets, waiting for buses and trains, etc. The design of this system is based on theoretical calculations, numerical simulations, and laboratory experiments.
In the calculations, the pressure drops of the system, including the bellmouth inlet on the top of the tower, 90-degree bend, bird screen, tower wall, flow straighteners, filter bank, etc., were determined and the results were to select the appropriate blower. The core technology of this study was the design of the filter bank. With the aid of the theoretical filtration models and laboratory experiments, we developed a W-shape filter bank that can largely increase the filtration area, therefore, lowering the media face velocity, increasing the filtration efficiency, reducing the pressure drop, increasing the PM holding capacity, and elongating the service time of the filters. The optimal main filters were designed and evaluated in the laboratory, which had a holding capacity 10 times higher than traditional glass fiber filters. With the results of minimized pressure drop of the system, an aim of delivering 1000 m 3 s -1 flow rate, PM2.5 filtration efficiency higher than 80%, system size with 20 m × 20 m × 23 m, a cost-effective system was invented.
With the 1000 m 3 s -1 flow rate, the effectiveness of the 3 rd Gen to reduce PM2.5 in a busy city was numerically predicted. In the model, two possible building plans were assumed, and ambient wind direction and speed were considered in the simulations. The results showed that the system can effectively clean the air in an area of 300-400 m in diameter. We also found there were ~10 and 20% of re-entrainment clean air circulating into the tower, thus, a guide plate should be installed at the outlet to prevent its dispersion to the upper air. A simple calculation shows to build 400 units of the 3 rd Gen in a 400 km 2 densely populated city is very cost-effective. The cost per capita is only $200 and it can save tens of thousands of people's lives.