You are on page 1of 10

Building and Environment 187 (2021) 107389

Contents lists available at ScienceDirect

Building and Environment


journal homepage: http://www.elsevier.com/locate/buildenv

Assessing the impact of filtration systems in indoor environments


with effectiveness
Tianyuan Li a, Jeffrey A. Siegel a, b, *
a
Department of Civil and Mineral Engineering, University of Toronto, Toronto, ON, Canada
b
Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: Effectiveness is commonly used to evaluate the particle removal performance of residential filtration systems.
Residential HVAC system Previous investigations have used various effectiveness metrics to assess system performance through modeling
Exposure reduction efforts. However, the model assumptions often lead to an overestimation of effectiveness. Further, none of the
Runtime
studies considered the impact of realistic residential heating, ventilation, and air conditioning (HVAC) system
Cycle on-times
Effectiveness metrics
operation patterns. In this investigation, we evaluated the strength and limitations of various effectiveness
metrics. We identified exposure reduction as the most appropriate metric for modeling analyses because it does
not rely on any steady-state assumptions. A time-varying mass balance model that accounts for system operation
parameters including runtimes and cycle on-times, and the impact of indoor emission sources is used to simulate
the changes in indoor particulate matter smaller than 2.5 μm (PM2.5) concentration with and without system
operation. The effectiveness results show that runtime has the biggest impact on exposure reduction. At a fixed
runtime, short cycle on-times and the coincidence of system operation with emission events can further improve
the removal performance of the system. Thus, the particle removal performance of a filtration system can be
optimized by changing its operation pattern while still maintaining similar fan energy use.

1. Introduction the impact of such a system on particle concentration in an indoor


environment, which is dependent on the filtration efficiency, as well as
Exposure to high levels of PM2.5 (particulate matter smaller than 2.5 the airflow rate through the filtration system and the indoor environ­
μm) has been associated with adverse human health effects (e.g., mental properties including particle source position, deposition, air
Ref. [1–4]). Because North Americans spend nearly 90% of their time change rate, and airflow pattern. Based on this definition, previous in­
indoors and almost 70% of the time at home [5–7], much of this expo­ vestigations have used various effectiveness metrics to assess the impact
sure occurs indoors and in residences. Particles in indoor environments of filtration and air cleaning (i.e., devices that removes particles through
can be transported from outdoor air through enclosure leaks and/or mechanisms other than filtration) technologies through a combination
mechanical ventilation systems (e.g. Ref. [8–11]), and generated from a of measurements and modeling efforts (e.g., Ref. [17–29]).
variety of indoor sources including cooking, smoking, cleaning, indoor While these earlier studies have helped to characterize the perfor­
chemical reactions, and various biological sources. Particle filtration mance of the filtration systems, most of the modeling investigations
systems, including filters in central forced-air heating, ventilation, and relied on relatively simple input parameters, steady-state or time-
air conditioning (HVAC) systems and portable air cleaners are widely averaged assumptions, and/or accounted only for particle loss mecha­
used to reduce indoor particle concentrations. nisms. The temporal variations in both particle sources and losses were
The particle removal performance of these filtration systems is often not considered and this often leads to an overestimation of filtration
characterized by their efficiency and effectiveness. Efficiency is the effectiveness. These studies also mainly focused on outdoor particle
fraction of particles that are removed from the air on a single pass sources because the epidemiological literature has focused on health
through a filtration system, which is mainly dependent on filter prop­ effects associated with ambient particle levels. Few of them accounted
erties (e.g. Ref. [12,13]), and system and dust loading conditions for the particles generated by indoor activities, which depends on
including face velocity and relative humidity [14–16]. Effectiveness is occupant behavior, is time-varying, and could contribute significantly to

* Corresponding author. Department of Civil and Mineral Engineering, 35 St. George St., Toronto, ON, M5S 1A4, Canada.
E-mail address: jeffrey.siegel@utoronto.ca (J.A. Siegel).

https://doi.org/10.1016/j.buildenv.2020.107389
Received 18 July 2020; Received in revised form 14 October 2020; Accepted 17 October 2020
Available online 20 October 2020
0360-1323/© 2020 Elsevier Ltd. All rights reserved.
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

indoor particle concentrations (e.g., Ref. [30]). Moreover, the existing this metric also represents the reduction in occupant inhalation expo­
studies, time-averaged or time-varying, generally ignore the impact of sure to airborne contaminants.
realistic residential HVAC system operation pattern, which can be /
characterized by both system runtime and cycle on-time. The system R = 1 − Csystem Cnosystem (1)
runtime (also referred to as duty cycle in some studies) is the fraction of
time the HVAC system operates. In most North American homes, the where:
forced-air systems are typically designed to cycle on and off to meet
conditioning demand. The runtimes vary greatly among these homes Csystem = mean concentration of the scenario with system operation
because of differences in climate, system design, and user preference. (μg⋅m− 3),
Although fan-only operation (i.e., independent of conditioning) is Cno system = mean concentration from the baseline, which is the same
allowed by most thermostats, the runtimes of these systems are gener­ scenario but without any system operation (μg⋅m− 3).
ally low. In a recent study, Touchie and Siegel [26] reported an 18%
median runtime in over 7000 North American homes based on smart Equation (1) has been used in some previous modeling studies [19,
thermostat data. The cycle on-time, which is the duration of the system 24,25,28,29,31], although with different baselines. In these studies, the
on-time of each cycle, also varies greatly among homes because of dif­ baseline scenarios have the same system operation status as the evalu­
ferences in system design and conditioning mode. For systems that ated interventions but with lower efficiency filters installed in the sys­
operate to meet conditioning needs, the cycle on-times are also generally tems. In this work, we selected no system operation as the baseline
too short for the indoor particle concentration to reach a steady state. because the primary focus of this work is to evaluate the impact of
Because of these two parameters, the realistic impact of HVAC systems filtration system operation, instead of filter efficiency which has been
on indoor particle concentration may not be well-simulated by some well investigated in the literature.
previous models; thus, questions remain about the actual particle
removal performance of HVAC systems in real residences. Further,
because effectiveness is a relative metric, even small effectiveness 2.2. Other effectiveness metrics in the literature
changes could lead to a big difference in particle concentration and
occupant exposure, especially in homes with high particle In addition to exposure reduction, we summarized the other four
concentrations. effectiveness metrics that have been used in the literature and their
The present investigation evaluates the exposure reduction impact of corresponding studies in Table 1. The first three metrics in Table 1 uti­
residential forced-air systems on PM2.5 through modeling. Our approach lize time-averaged assumptions, while the last one, similar to exposure
uses a time-varying mass balance model that accounts for input pa­ reduction (R), could also be applied to transient scenarios. The theo­
rameters including indoor sources, runtimes, and cycle on-times to retical effectiveness (H) is the most commonly used metric in the liter­
simulate the indoor concentration changes in a transient-state. The ature for modeling analyses. Without consideration of runtime, it is used
primary objective of this work is to investigate the impact of filtration- to estimate the theoretically maximum impact of the operation of a
system related parameters on exposure reduction. The secondary filtration system compared to the no system scenario. In real life, H can
objective is to evaluate the strength and limitations of the effectiveness be achieved if a system is operating continuously. However, as discussed
metrics used in earlier work. The ultimate goal of this work is to provide earlier, it is rarely the case for residential filtration systems. The runtime
a transient-state framework with an adequate effectiveness metric to influenced effectiveness (RIE) is therefore used to account for the impact
evaluate the actual performance of the residential filtration system and of runtime (t) by multiplying the filtration system loss rate by the run­
provide guidance on performance improvement. time in the denominator. It could provide a more realistic estimation of
system impact as it does not rely on the assumption of continuous
operation. A similar form of this metric is the runtime factored effec­
2. Effectiveness metrics
tiveness (H′ ). It has a distinct way to account for the impact of runtime
by multiplying the theoretical effectiveness of the system by its frac­
In the literature, previous studies have applied various effectiveness
tional operation time (i.e., runtime). Both RIE and H′ are equivalent to H
metrics on field measurements or modeled parameters to evaluate
when the runtime is either 0 or 100% but differ for any runtime values in
filtration system performance. In these metrics, the baseline is either
between. Although H′ was not used in previous studies, it is included in
defined as the concentration (or particle loss rate) without any system
Table 1 for comparison purposes. Lastly, median effectiveness (Hmedian)
operation, or with system operation but with a lower efficiency filter
is similar to exposure reduction (R) except for the fact that it uses
installed in the system, depending on a specific investigation. The reason
for the metric selection was rarely discussed. In the following section, we
summarized five metrics and discussed the assumptions, strengths, and Table 1
limitations of these metrics when used in modeling analyses. Summary of effectiveness metrics used in previous modeling studies.
Metric Equations Study
2.1. Exposure reduction (R) Theoretical effectiveness H = 1 − (λi + β)/(λi + β + λr η) (2) [17,20,22,
(H) 25,27,31]
Equation (1) shows exposure reduction (R), the filtration effective­ Runtime influenced RIE = 1 − (λi + β)/(λi + β + λr ηt) (3) [18,26]
effectiveness (RIE)
ness of the system, calculated by subtracting the ratio of mean indoor
Runtime factored H = H× t

(4)
concentration with filtration system operation to the mean indoor con­ Effectiveness (H′ )
centration for the same scenario but without any system operation Median effectiveness Hmedian = 1 − Csystem, median / (5) [21]
(baseline) from unity. One key strength of Equation (1) in modeling (Hmedian) Cno system, median
analyses is that it does not rely on the assumptions of constant sources/
where:
losses parameters or continuous filtration system operation. The use of λi = air exchange rate due to natural ventilation (hr− 1),
mean concentrations in Equation (1) allows exposure reduction to ac­ β = particle deposition rate (hr− 1),
count for the changes in concentrations due to temporal variations in λr = recirculation rate through a central forced-air HVAC system (hr− 1),
sources and losses and reflect them in the resulting filtration effective­ η = filtration system efficiency (%),
t = runtime (%),
ness. Further, if we assume that the breathing rate of an individual Csystem, median = median concentration of the scenario with system operation (μg⋅m− 3),
occupant is constant, the ratio of the mean concentration in Equation (1) Cno system, median = median concentration from the baseline, which is the same scenario
equals to the ratio of the total inhalation exposure of the occupant. Thus, without any system operation (μg⋅m− 3).

2
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

median concentrations instead of mean concentrations, to evaluate throughout the simulation and a full cycle is considered to be the system-
filtration performance. on time (i.e., cycle on-time) plus the system off-time. Because runtime is
In this work, we applied all the five effectiveness metrics (R, H, RIE, the fraction of time the HVAC system operates, the duration of a full
H′ , and Hmedian) to simulated concentrations to explore the differences cycle can be calculated by dividing the cycle on-time by runtime. For
between metrics. For all five metrics discussed above, we consider R to each simulation, the system is assumed to start operation at the begin­
be the most appropriate metric when estimating particle removal per­ ning of the simulation. The potential impacts of this assumption of
formance of a filtration system for the following two reasons. First, R system operation on the simulation results are discussed in more detail
does not rely on the time-averaged assumptions like H, RIE, and H′ . in Section 4.2. The system operation status (SO) for each time step is
Second, R uses the mean concentration, which is the integral of the determined based on the cycle on-time and full-cycle duration.
concentration over time divided by the duration of the simulation.
L1 = λi + β (7)
Because the total inhalation exposure of an occupant is the integral of
the product of concentration and breathing rate over time, R better
L2 = λr η (8)
represents the actual inhalation exposure reduction of occupant than
Hmedian, which uses the median concentration. S = λi PCout + E/V⋅ES (9)

3. Methodology where:

To calculate exposure reduction and the effectiveness metrics sum­ Cout = mean daily outdoor particle concentration (μg⋅m− 3),
marized in Section 2, we estimated the indoor PM2.5 concentration for P = building envelope penetration factor (fraction),
varying indoor scenarios with and without filtration system operation ES = emission status for time-varying indoor sources in the previous
through modeling analyses. time step, determined by the duration and the start time of each
emission event, 1 if there was a source present, 0 if there was none.
3.1. Mass balance model framework

3.2. Model input parameters


A time-varying mass-balance model was used to estimate the indoor
concentrations (C) of PM2.5 based on the previous time step (tn-1) in
We divided our input parameters into three categories: filtration
residences using Equation (6). The model assumes a well-mixed indoor
system-related, non-system related, and particle source-related. In doing
environment with both indoor and outdoor particle sources, first-order
so, we were able to isolate the impact of the filtration system from other
loss process, and a central forced-air HVAC system without dedicated
influencing factors. For each input parameter, a statistical distribution
outdoor air supply (as is typically the case in North American homes).
was collected and used in a Monte Carlo simulation of Equation (6) to
We also assumed that all the input parameters, except for the filtration
examine the range of the impact of the HVAC systems with varying
system-related parameters (L2 and SO), are the same for each pair of
parameters in homes. In addition to the Monte Carlo simulation, single-
with and without (i.e., baseline) system operation simulations. This
value simulations that used the mean or geometric mean of normally- or
assumption is to ensure that difference in concentration is caused only
lognormally-distributed parameters, respectively, were conducted to
by the system operation. The initial concentration (C0) of the model is
reduce variations and clearly illustrate the impact of individual pa­
assumed to be the time-averaged concentration of the same scenario
rameters on effectiveness.
without any system operation and the length of each simulation is set to
be 24 h. In addition, this model framework does not consider the impact
3.2.1. Filtration system-related parameters
of window or door opening. Although window opening plays an
The filtration system-related parameters include runtime, cycle on-
important role in exposure modeling due to its impact on the air ex­
time, recirculation rate (λr), and filtration system efficiency (η). In a
change rate, we neglected this behavior to reduce variations and focus
recent study, Touchie and Siegel [26] reported runtimes measured by
on the filtration system impact. The input parameters of this model are
smart thermostats from over 7000 homes across most Canadian prov­
described in detail in Section 3.2. All dynamic indoor air mass balance
inces and US states. It is by far the largest dataset available for measured
simulations in this work are conducted using Stata Version 16 [32].
residential runtime and is consistent with other runtime measurement
/ ( )
C(tn ) = C(tn− 1 ) × e− (L1 +L2 ⋅SO)Δt + S (L1 + L2 ⋅ SO)⋅ 1 − e− (L1 +L2 ⋅SO)Δt (6) studies. A lognormal distribution (GM = 18.2%, GSD = 2.3, limited to
the range of 0–100%) was fit to their reported data by minimizing the
where: sum of the squared errors between the reported values and the
lognormal distribution function. It is important to note that with the
C(tn) = indoor PM2.5 concentration of particles at the current time boundaries, the resulting distribution is no longer lognormally distrib­
step (μg⋅m− 3), uted, but it is a reasonable simulation of the realistic runtime mea­
L1 = the summation of particle loss rate due to particle removal surements. Two concerns of this dataset are that it might not be entirely
mechanisms including indoor and outdoor air exchange and depo­ representative as only a fraction of homes in North America currently
sition without system operation (hr− 1), calculated from Equation (7), uses a smart thermostat and that the use of smart thermostat likely
L2 = particle loss rate due to filtration system operation (hr− 1), underestimated runtimes [26]. This distribution is slightly lower than
calculated from Equation (8), the runtime distribution (GM = 24.6%, GSD = 1.85) used in El Orch
SO = HVAC system operation status in the previous time step, 1 for et al. [33] possibly due to the underestimation associated with smart
system on and 0 for system off, characterized by system runtime and thermostats, but it is more likely because El Orch et al. [33] relied on a
cycle on-time, sample of homes and light commercial buildings located in
S = the summation of volume normalized indoor and outdoor par­ cooling-dominated US states, which generally have higher runtimes. The
ticle sources (μg⋅m− 3hr− 1), calculated from Equation (9), chosen distribution is higher than the median runtime (9.6%) from 20
Δt = the time interval between time steps, 1 min. heating-dominated homes in Toronto, Ontario, Canada in Li et al. [23]
due to mild weather during the yearlong field study.
In the model framework, systems are set to cycle on and off to As discussed earlier, cycle on-time is an important parameter that has
simulate a typical residential forced-air HVAC system. The length of been overlooked in previous similar studies. It is defined as the duration
each system-on and system-off time is assumed to be constant of time the system (or the fan) operates during each cycle. In the

3
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

literature, cycle on-time (or sometimes referred to as cycle time) is indoor sources are often not considered in previous studies (e.g. Ref. [9,
mainly investigated to evaluate the dehumidification performance of air 33]), likely because the epidemiological literature has focused on
conditioners (e.g. Ref. [34]), instead of particle removal performance, ambient particles. Indoor PM2.5 is emitted from a wide variety of
and the measurements are scarce. Thornburg et al. [35] reported a mean emission events and occupant activities such as cooking, smoking,
cycle on-time of 10 min based on data collected for a total of 215 days of dusting and vacuuming, use of humidifiers and aromatic products, and
measurements from homes in both heating and cooling mode in North secondary aerosol formation from ozone/terpene reactions [47–50]. The
Carolina. Stephens et al. [36] reported a longer mean cycle on-time of frequency and strength of these sources also vary greatly among
18.3 min for a total of 54 days of measurements from 8 residential households and events within the same household. As a result, it is
buildings in the cooling season in Austin, Texas. It is worth noting that extremely challenging to accurately characterize time-resolved indoor
both previously reported on-times are generally not long enough for the emissions without a large dataset of source frequencies and schedules.
indoor particle concentration to reach a steady state. For lack of a more Chan et al. [30] used an identification and characterization algorithm to
robust dataset, we rely on the measurements of cycle on-time from the quantify time-resolved PM2.5 emission events from 18 apartments in
20 Toronto homes in Li et al. [23] and fit a lognormal distribution with a California over 224 days. The authors quantified the duration and
GM of 6.11 min (GSD = 2.28). Although the lognormal distribution does strength of 836 distinct emission events and established a diurnal fre­
not fit these data perfectly, it fits better than other reasonable distri­ quency of the events. In this work, we rely on the findings from Chan
butions that we have tried. The range of this distribution is limited to be et al. [30] and simulated the time-varying indoor emission events based
at least longer than 2 min based on observations of the shortest cycles in on three parameters including emission rate (E), home volume (V), and
the study. On-times longer than 100 min, which were caused by the emission status (ES). The emission rate (E) follows a lognormal distri­
continuous fan-only operation, were also excluded because they repre­ bution with a GM of 40 mg/h (GSD = 3.9). Because emission rate is
sented a small fraction (0.8%) of the operation patterns and are due to found to be significantly correlated with home volume (V), we used the
user preference. Similar to the runtime distribution, the resulting mean floor area (94.7 m2) reported in Chan et al. [30] and an assumed
on-time distribution with the boundaries is no longer lognormally mean ceiling height of 2.74 m to estimate the mean home volume (259
distributed, but it reflects the realistic operation patterns of HVAC sys­ m3). The emission status (ES) is determined by both the duration and the
tems. It is worth to note that these data might not be entirely repre­ start time of each emission event. The former parameter was retrieved
sentative of the residential systems in North America because they are from the diurnal fraction of hours with identified emission events for
solely collected from a small sample of homes in a heating-dominated weekday and weekend reported in Chan et al. [30]. The latter parameter
climate and the heating cycle on-times are generally shorter when was randomly selected for each hour from a uniform distribution of
compared to cooling cycle on-times. We will further discuss the impact possible start times (ranging from the first minute of the hour to the last
of this selection of cycle on-time distribution on effectiveness in Section minute that an emission event could happen to satisfy the former
5.2. parameter) to model the events stochastically. It is important to note
Recirculation rate (λr) is the volume of airflow through the HVAC that a constant emission rate (randomly selected from the aforemen­
system relative to the volume of the space it serves in. We rely on a tioned emission rate distribution) is assumed for all events that
lognormal distribution (GM = 4.36 h− 1, GSD = 1.44) calculated by Fisk happened within an individual 24-h simulation scenario. This assump­
and Chan [37] from a total of 32 residential and small commercial tion is not realistic, but it was made to minimize the uncertainties
buildings from two studies [39,40]. This distribution is similar to the introduced by the temporal variations in emission rate (as well as other
values (GM = 4.3 h− 1, GSD = 1.4) used in Touchie and Siegel [26] from parameters) to focus on filtration system impact.
54 homes, and slightly lower than the distribution (GM = 5.7 h− 1, GSD We accounted for particles of outdoor origin by using the mean daily
= 1.26) used in El Orch et al. [33]; which only relied on the sample of outdoor particle concentration (Cout). To put the impact of filtration
cooling-dominated buildings in Stephens et al. [40]. system in context, we considered two cities, Toronto, Ontario, Canada,
which generally has low ambient PM2.5 concentrations, and New Delhi,
3.2.2. Non-system related parameters India, which on the contrary, is heavily polluted. For both Toronto and
The non-system related parameters include air change rate (λi), New Delhi, a lognormal distribution was fit to the mean daily concen­
deposition rate (β), and penetration (P). Although home volume is a tration measurements over a year [51,52], resulting in a GM of 6.5 μg
building characterization parameter, it is categorized as a source-related m− 3 (GSD = 1.77) and a GM of 92.5 μg m− 3 (GSD = 2.07), respectively.
parameter as emission rate (E) is correlated with home volume while all
the other non-source related parameters are independent of it, and thus 4. Results
described in Section 3.2.3.
Air change rate (λi) due to infiltration varies widely across buildings Based on the model framework and input parameters established, we
and temporarily even within the same building [41–44]. Because of the estimated 24-h concentration changes through Monte Carlo simulations
large variations, we rely on the modeled distribution of air change rate first with a time-averaged continuous indoor particle source and then
estimated by Persily et al. [45] based on a sample of 209 dwellings that with time-varying indoor emission events. The effectiveness metrics
represent 80% of U.S. housing stock. A lognormal distribution (GM = were determined based on Equations (1)–(5) and the differences be­
0.44 h− 1, GSD = 2.04) was fitted by El Orch et al. [33] to the distribution tween effectiveness metrics were investigated. We also specifically
in Persily et al. [45]; and was used for simulation in this work. examined the impact of individual parameters on exposure reduction
For deposition rate (β) to indoor surfaces, like some of the previous (R), runtime influenced effectiveness (RIE), and median effectiveness
PM2.5 simulation studies (e.g. Ref. [26]), we rely on a normal distribu­ (Hmedian) with both single-value simulations as well as Monte Carlo
tion of deposition rates (mean = 0.42 h− 1, SD = 0.19) calculated by simulations.
Williams et al. [46] in 37 homes using air change rate measurements.
The distribution is limited to a range of 0.1–0.8 h− 1 based on the min­ 4.1. Difference between effectiveness metrics
imum and maximum values. For PM2.5 penetration (P), we relied on the
same study [46] and used a normal distribution with a mean (±SD) of Because time-varying sources introduce fluctuations into the effec­
0.72 (±0.21), limited to the range of 0 to 1. tiveness results due to concurrence of system operation and emission
events, we first conducted a set of simulations with a constant, contin­
3.2.3. Particle source-related uous indoor source to clearly illustrate the differences between effec­
We categorized particle source-related parameters based on the tiveness metrics. The impact of this assumption is discussed in more
origin (i.e., indoor or outdoor) of the particles. As discussed earlier, detail in Section 4.3. The emission rate of this continuous source is

4
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

Fig. 1. Effectiveness results from five effectiveness


metrics (R, H, RIE, H′ , and Hmedian) as a function of (a)
system runtime; and (b) cycle on-time, over 24-h
simulation with time-averaged indoor emission
source for Toronto, Ontario, Canada. The central solid
line represents the median effectiveness values and
the shaded areas represent the 25th and 75th
percentile range from the Monte Carlo simulations.
All the other input parameters were sampled 1000
times for each value of the independent parameter
from distributions described in Section 3.2 to calcu­
late concentrations.

assumed to be the time-averaged hourly emission rate (12.67 mg h− 1) no noticeable difference between RIE and R; however, at a cycle on-time
calculated by spreading the geometric mean of the emission rates (40 of 100 min (although unlikely to happen in most systems operating to
mg h− 1) of the emission events (GM = 19 min) reported in Chan et al. meet conditioning demand only), the absolute difference between the
[30] over an hour. To evaluate the impact of input parameters, we median values of RIE (or Hmedian) and R is greater than 10%. This trend
conducted a Monte Carlo simulation to calculate concentrations with explains why there is minimal difference between RIE and R in Fig. 1 (a),
and without system operations as a function of each independent as the cycle on-time distribution used in the simulations has a geometric
parameter. For each value of the independent parameter, the concen­ mean of 6.1 min.
trations were calculated 1000 times with random sampling of the other In both Fig. 1 (a) and (b), we can also observe a large variation in R.
input parameters from the distributions described in Section 3.2. Fig. 1 This large range shows that even for systems with a MERV 14 electret
(a) and (b) show the simulated range (the middle two quartiles) of filter installed in them, a wide range of effectiveness is expected among
effectiveness results as a function of runtime and cycle on-time, varied homes because of variations in recirculation rate, deposition, and air
over the range from 0 to 100% (at 1% step) and 0 to 100 min (at a 1-min change rate. When comparing the two figures, we can also observe
step) for Toronto, respectively (companion Fig. S1 a and b for New Delhi greater variations in R for a given cycle on-time (in Fig. 1 b), indicating
are available in the Supplementary Materials). The cycle on-time in that runtime has a greater influence on filtration system performance
Fig. 1 (a) and runtime in Fig. 1 (b) had a lognormal distribution with a when compared to cycle on-time.
GM of 6.11 min (GSD = 2.28, bound between 2 and 100 min) and a The most important point demonstrated by this set of simulations is
lognormal distribution with a GM of 18.2% (GSD = 2.3, limit to 100%), that R decreases as a function of cycle on-time (as shown in Fig. 1b).
respectively. Thus, for any system, at a given runtime and consequently, with similar
In Fig. 1 (a), all effectiveness metrics except theoretical effectiveness energy consequences, we can get more benefits from filtration (i.e.,
(H) increase as a function of runtime. H is considerably higher than the higher exposure reduction) with shorter system-on cycles. However, it is
rest of effectiveness metrics, especially at lower runtimes. This trend is also important to note that even with high efficiency filters installed in
expected as H is the theoretically maximum effectiveness of system them, the exposure reduction of systems with relatively low runtime
operation and is runtime independent. Both the runtime influenced (GM = 18.2%) is bounded at approximately 30%, and this value is
effectiveness (RIE) and median effectiveness (Hmedian) are very similar to consistent with previous investigations [18,26]. Further, short
exposure reduction (R), although the former always slightly over­ system-on cycles can also cause other issues related to the heating and
estimate effectiveness at all runtime values. This overestimation is cooling devices such as waste of residual heat [53], degradation of
caused by the steady-state assumption that is utilized by RIE. With this moisture removal capacity [34], and equipment wear.
underlying assumption, RIE calculates an equivalent mean concentra­
tion for the system operation scenario based on the particle loss rate in
the denominator. This mean concentration underestimates the actual
mean concentration that is calculated as the time-weighted average
concentration (i.e., the integral of concentration over time divided by
the duration of the simulation). Surprisingly, there is such little differ­
ence between RIE and R given that the steady-state assumption, which
RIE is dependent on, is not met in our model framework. The reason for
this trend is discussed in the paragraph below. Fig. 1 (a) also shows that
runtime factored effectiveness (H′ ), while it accounts for runtime by
multiplying the theoretically maximum performance (i.e., H) by the
runtime, is only a close estimation of R at extremely low or high run­
times, making it a less than ideal metric for system performance
evaluation.
Fig. 1 (b) shows greater differences between R, RIE, and Hmedian as a
function of cycle on-time. Similar to Fig. 1 (a), RIE always overestimates
R because of the steady-state assumption. In Fig. 1 (b), it is also clear that
Hmedian always underestimates R. It is due to the fact that for the with
system operation scenario, the median concentration was always higher
than the mean concentration because of the low runtimes (GM = Fig. 2. Exposure reduction (R), runtime influenced effectiveness (RIE), and
18.2%). As the cycle on-time increases, the difference between the median effectiveness (Hmedian) as a function of L2/L1 ratio over 24-h simulation
metrics also increases. At lower cycle on-times (below 10 min), there is for Toronto.

5
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

4.2. Impact of parameters on effectiveness (single-value simulation) emission rates will impact Hmedian and R. We will discuss the impact of
the time-varying emission rate in more detail in Section 4.3.
As shown in Fig. 1, among all effectiveness metrics, runtime influ­ In addition to these three input parameters, we also investigated the
enced effectiveness (RIE) and median effectiveness (Hmedian) yield the impact of simulation artifacts including initial concentration (C0), initial
most similar results when compared to exposure reduction (R). To system operation status, incomplete operation cycles, and simulation
further examine the role of each parameter beside system runtime and time. In our model framework, we assumed the system-off time-aver­
cycle on-time on these three metrics, we conducted 24-hr single-value aged concentration to be the initial concentration. A higher initial
simulations using the mean of normally-distributed parameters and concentration would have no impact on RIE, but it would lead to higher
the geometric mean of lognormally-distributed parameters. The goal of R and Hmedian, indicating that operating the filtration system in a home
these single-value simulations was to smooth the effectiveness curve by with higher initial concentration is more beneficial. Initial system
reducing variations associated with the input parameters to clearly operation status has no impact on the three effectiveness metrics when
demonstrate the impact of the independent parameter in question. Fig. 2 the system-off time-averaged concentration is used as the initial con­
shows the results of R, RIE, and Hmedian, as a function of L2/L1 ratio (i.e., centration. However, if a higher (or lower) initial concentration was
system loss rate over non-system loss rate). Similar to Section 4.1, a used, assuming system operation at the start of the simulation would
constant indoor emission rate (12.67 mg h− 1) is assumed for these also lead to higher (or lower) R and Hmedian when compared to RIE,
simulations. respectively. Through our investigation, we found that incomplete
Fig. 2 shows that all three effectiveness metrics (Hmedian overlaps operation cycles (i.e., the system-on cycle is not completed by the end of
with R) increase as L2/L1 ratio increases, but this increase reduces the simulation) cause periodic fluctuations in both R and Hmedian since
gradually at higher L2/L1 ratios. This trend is expected as effectiveness the actual runtime of the simulation was lower than the assumed value.
increases when the removal due to the filtration system is more Fig. S3 (a) shows an example of the impact of incomplete cycles with 24-
important than other removal mechanisms. However, RIE constantly h single-value simulation. Lastly, simulation time has no impact on any
overestimates R, especially at higher L2/L1 ratios. As discussed earlier in of the three effectiveness metrics. However, a longer simulation time
Section 4.1, the overestimation of RIE is caused by the underestimation would reduce the impact of initial concentration, initial operation sta­
of the equivalent mean concentration when compared to the time- tus, and incomplete operation cycles (shown in Figure S3 b).
weighted average concentration. Further, because of the low runtime
(18.2%), the time-weighted average concentration is dominated by the
system-off concentration. Consequently, at high L2/L1 ratios, the in­
crease in R becomes smaller because the performance of a filtration
system is bounded by its runtime, as opposed to its loss rate. Even with
the low runtimes, RIE becomes less realistic and gradually approaches
100% as the total loss rate in the denominator in Equation (3) continues
increasing. The difference between R and RIE would be even greater
with increases in cycle on-time and would be smaller with increases in
runtime (the impact of these two parameters is shown in Fig. S2). This
trend shows that an increase in filtration system removal rate may not
necessarily lead to greater exposure reduction. Both high runtimes and
short cycle on-times are important for filtration system performance.
From the single-value simulations, we also found that neither the
mean daily outdoor concentration nor the time-averaged hourly indoor
emission rate has any impact on all three effectiveness metrics. These
results show the effectiveness (regardless of metric) of a given system in
a given home is not influenced by the strength of indoor or outdoor
sources. This trend is expected as the steady-state and time-averaged
assumptions are satisfied in these single-value simulations, and thus
both the R and Hmedian equations (Equations (1) and (5)) can be Fig. 4. Exposure reduction (R), runtime influenced effectiveness (RIE), and
simplified to the RIE equation (Equation (3)). However, it is important to median effectiveness (Hmedian) as a function of cycle on-time over 24-h simu­
note that any temporal variations in outdoor concentration or indoor lation with time-varying emission events for Toronto, Ontario, Canada.

Fig. 3. Effectiveness results from five effectiveness metrics (R, H, RIE, H′ , and Hmedian) as a function of (a) system runtime; and (b) cycle on-time, over 24-h simulation
with time-varying indoor emission source for Toronto, Ontario, Canada. The central solid line represents the median effectiveness values and the shaded areas
represent the 25th and 75th percentile range from the Monte Carlo simulations.

6
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

4.3. Impact of time-varying emission events different. Both system runtime and cycle on-time of the portable air
cleaner would likely be higher. Care should be taken when applying this
We used emission status (ES) to simulate the indoor emission events model to portable air cleaners because one of the central assumptions of
stochastically and assess the impact of the time-varying emission events. this model is a well-mixed environment. It is less likely to be achieved
As described in Section 3.2.3, ES is determined based on the fraction of with the use of portable air cleaners for the entire home due to lack of air
hour with identified emission events reported in Chan et al. [30] and a distribution system and even for a single room due to short-circuiting (i.
randomly selected emission start time for each hour. Similar to Section e., re-entrainment of air from the outlet of the air cleaner into the inlet)
4.1 but with the time-varying emission events, we calculated the particle [25,27,54]. As a result, the simulated exposure reduction would likely
concentrations for 1000 scenarios with and without system operation overestimate the actual performance of the air cleaners. However,
per runtime and cycle on-time value through the Monte Carlo simula­ portable air cleaners are also likely to be positioned near people and
tion. The effectiveness results are shown in Fig. 3 (a) and (b) as a result in a greater reduction than the result of the simulation. Thus, the
function of runtime and cycle on-time, respectively. For both input pa­ uncertainties associated with this model would be greater when applied
rameters, Fig. 3 shares a similar trend and range for the effectiveness to portable air cleaners.
metrics as Fig. 1. Further, it seems that the time-varying emission events
have a very limited impact on exposure reduction (R) and median 5.2. Limitations
effectiveness (Hmedian) that should reflect changes in concentration in
transient state. This is due to the fact that the temporal variation indoor There are several important limitations to this work. For one, the
emission rate is overshadowed by the larger variations in other input central assumption of our model is a well-mixed environment. Although
parameters among simulated scenarios. the assumption of perfect mixing is more likely to be achieved in homes
To clearly illustrate the impact of time-varying indoor emission with central HVAC systems with a recirculation rate greater than 4 per
events and isolate its impact from other input parameters, we conducted hour [24], this assumption is still a simplification of the mixing condi­
a single-value simulation (similar to those in Section 4.2) with varying tion in real homes as the pollutant concentration could be different
cycle on-time. As shown in Fig. 4, while runtime influenced effectiveness across rooms, especially when the system is off. The actual reduction in
(RIE) remains constant, R and Hmedian fluctuate over cycle on-time, occupant exposure depends on the proximity to the source, the room
although still follow the general decreasing trend. Hmedian also gener­ mixing condition, and the airflow pattern between the rooms. The
ally overestimates R because of the elevations in concentration from the simulated exposure reduction could overestimate the actual reduction in
background due to the emission events. Further, the periodic fluctua­ some instances while underestimating in other instances. Because all
tions caused by incomplete operation cycles (shown in Figure S3 a) in five effectiveness metrics rely on the well-mixed assumption in the same
both R and Hmedian are replaced by random fluctuations in Fig. 4. These manner, we expect that the metric bias associated with imperfect mixing
fluctuations are caused by the coincidence of system operation and conditions would be small. In-situ concentration measurements in
emission events. If time-varying outdoor concentrations were used, the controlled environments and computational fluid dynamics (CFD)
temporal variations in outdoor concentration would have a similar modeling should be considered for future studies to explore the un­
impact. In both cases, the magnitude of the fluctuations in effectiveness certainties introduced by imperfect mixing, although the specifics of the
is dependent on the magnitude and duration of changes in indoor test environments and the boundary conditions are challenging and
concentration. important aspects to resolve in such future work.
By closely examining concentration profiles from the simulation, we Secondly, the system runtime and cycle on-time were not retrieved
found that R was higher when the system operation was more syn­ from the same sample of homes. As a result, the potential correlation
chronized with the emission events (i.e., system was operating more between runtime and cycle on-time is not reflected by the simulation
when the concentration was elevated). For instance, as shown in Fig. 4, results. For instance, both runtime and cycle on-time could be strongly
R is 21.7% at a cycle on-time of 42 min and 29.0% at a cycle on-time of influenced by the size of the conditioning system relative to the condi­
43 min. Despite the longer cycle on-time and the expected decreasing tioning load. To explore this potential correlation, we plotted the
trend, R at a cycle on-time of 43 min was higher because there happened monthly mean cycle on-time from the 20 Toronto homes in Li et al. [23]
to be a higher fraction of system operation during the higher concen­ over monthly mean runtime for heating and cooling (Fig. S5). The re­
tration periods. This finding is consistent with the impact of initial sults show that there are large variations in cycle on-times and runtimes
concentration (C0) on R that we discussed in Section 4.2. The concen­ across all homes and between operation modes. Further, in most homes,
tration profiles of these two examples are shown in Figs. S4(a) and (b), the cycle on-time is independent of runtime. However, we observed that
respectively. One important implication of this finding is that in addition for cooling operation, many homes with runtimes lower than 10% also
to cycle on-time, the start time and the coincidence of system operation have cycle on-times shorter than 10 min, likely because the air condi­
with emission events could also influence the performance of a filtration tioning unit in these homes is oversized [55]. In homes with an under­
system. Thus, the particle removal benefit of a filtration system can be sized air conditioning unit, we would expect the runtimes to be generally
potentially increased by closely monitoring the indoor particle concen­ high with longer system on-times. Another example would be that older
tration and turning on the system at the right time. systems tend to have longer cycle on-times [56]. These systems are often
installed in older homes that are more prone to duct leakage, leaky
5. Discussion construction, and poor insulation, which in turn, leads to larger condi­
tioning loads and higher runtimes. It is hard to predict the exact impact
5.1. Application to portable air cleaners of these two types of combinations (low runtimes with short on-times
and high runtimes with long on-times) on the difference between
While the primary focus of this work is to investigate the exposure exposure reduction (R) and runtime influenced effectiveness (RIE)
reduction benefit of residential forced-air systems on PM2.5, the central because runtime and cycle on-time have opposite impacts; however, it is
model framework can also be used to evaluate the impact of portable air important to note that in both types of homes, a combination of high
cleaners. The current model can be modified by replacing the filtration runtimes and short cycle on-times, which is important for better filtra­
system loss rate (L2) with a ratio of the clean air delivery rate (CADR, a tion system performance, is less likely to occur without specified user
product of airflow rate, and PM2.5 filtration efficiency) of the air cleaner settings.
to the volume of its served space. Because the main purpose of portable Sampling these two parameters in pairs from the same dataset may
air cleaners is to reduce particle concentration, as opposed to satisfying provide a more realistic simulation of system performance. However,
conditioning demand, the system operation status would also be because of the large variations across homes and the difference between

7
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

heating and cooling, this approach may also introduce large un­ 5.3. Implications and future studies
certainties to the simulation results. Further, the impact of this corre­
lation is likely to be overshadowed by the more dominant parameters, In this work, we examined the difference between exposure reduc­
such as air change rate, which are generally higher in older homes. tion (R) and four other effectiveness metrics including theoretical
Future studies could stratify the homes in heating and cooling domi­ effectiveness (H), runtime influenced effectiveness (RIE), runtime
nated climate based on operation mode (i.e., heating or cooling), con­ factored effectiveness (H′ ), and median effectiveness (Hmedian). The
ditioning system size, construction year, operation preference (i.e., if the simulation results show that RIE and Hmedian generally provides a close
fan-only operation is used) to evaluate the typical system performance in approximation of R, especially with cycle on-times shorter than 20 min
different types of homes. and L2/L1 ratios lower than 10. These two conditions are generally met
Thirdly, the runtime distribution that we relied on was from smart for residential systems operating to meet conditioning demands.
thermostats which logged calls of fan operation instead of the actual fan Consistent with previous investigations [18,26], the runtime has the
operation [26]. The actual fan operation could be different from the biggest impact on R. At high runtimes, all effectiveness metrics approach
recorded data if the fan was controlled by the conditioning unit or H (i.e., the theoretical maximum performance). Thus, in the absence of
continued operating after the conditioning system switched off. As a concentration measurements, RIE can be used to predict the maximum
result, this runtime distribution may not be an accurate representation particle removal performance of a typical residential filtration system in
of the fan operation that influences the particle removal performance of a given home. However, care should be taken because both longer cycle
filtration systems. This limitation is also shared by cycle on-times. Our on-time and temporal variations in concentration, especially at low
distribution of cycle on-times was determined by a combination of runtimes, would lead to RIE overestimating the actual performance of
measurement approaches that record fan or conditioning unit operation the system.
[53]. Because the fan cycles may differ in duration and offset from Further, it is important to note that RIE does not capture changes in
conditioning cycles in some homes [57], this cycle on-time distribution concentration in the transient state. From our time-varying emission rate
may not be an accurate representation of the fan cycle on-times. How­ analysis (Section 4.3), we found that both the start time and coincidence
ever, these differences are likely to have a limited impact on exposure of system operation and concentration elevation could greatly impact
reduction results from our Monte Carlo simulations. the R of the system. Thus, any analyses that focus on concentration
A further limitation associated with cycle on-time is that we relied on changes in the transient state should use real-time concentration mea­
cycle on-times from homes in a heating-dominated climate, which surements, as opposed to any effectiveness metric (H, RIE, or H’) that
generally have shorter cycles on-times when compared to their coun­ utilizes steady-state assumption. Given that most indoor emission events
terparts in cooling-dominated climates. Based on the mean cycle on-time were generated by human activities, with the same runtime (and similar
(18.3 min) reported in Stephens et al. [36] and the general decreasing energy consequences), an optimized system operation schedule which
trend shown in Figs. 1 and 3, the absolute difference in exposure targets the periods with elevated concentrations will result in greater
reduction is 3% higher in our sample of heating-dominated homes when exposure reduction than an operation schedule just cycles on and off to
compared to the homes in Stephens et al. [36]. However, the impact of meet the conditioning demand. The transient-state framework presented
cycle on-time on exposure, especially at low on-time, is much smaller in this work and R could be used to evaluate system operation schedules
than the impact of runtime. and identify optimized schedules to reduce indoor exposure of particu­
Another limitation of this work is the interaction between runtime, late matter.
filter efficiency, and airflow over the service life of a filter. Because the
efficiency of electret filters generally decreases with particle loading 6. Conclusions
[12,13,15], the MERV 14 electret filters that are installed in systems
with higher runtimes are likely to experience greater declines in effi­ In this work, we evaluated the particle removal performance of
ciency. A more heavily loaded filter is more likely to cause greater filter residential filtration systems through time-varying transient-state mass
pressure drop and reduce the airflow rate (although the airflow rate can balance models and compared the effectiveness metrics used in previous
be maintained in some systems at a greater energy penalty, depending studies. Among these metrics, exposure reduction (R) provides the most
on the presence of fan speed controls). Because simulation time in our appropriate system performance evaluation as it accounts for concen­
model is only 24 h, transferring the results from this work directly to a tration changes in transient-state. The runtime influenced effectiveness
longer period (e.g., 3 months) would likely result in overestimation of (RIE) provides the closest approximation of R, but it overestimates the
the actual impact of the filtration systems. system performance, especially at high cycle-on times and low runtimes.
We should also note that in this work, we limited our assessment on The results show that system runtime has the biggest impact on exposure
the removal performance of the filtration systems by effectiveness, reduction. At a given runtime, and consequently, with similar energy
which is a relative metric that describes how filtration systems compete consequences, short cycle on-times and the coincidence of system
with other removal mechanisms including deposition and ventilation. operation with emission events can improve the removal performance of
The amount of reduction in indoor particle concentration is thus not the system. Even this improvement in exposure reduction is relatively
reflected by the effectiveness metrics. Parameters such as mean daily small at low runtimes (less than 10% at a runtime of 18%), but it could
outdoor concentration and time-averaged emission rate, although have lead to a large reduction in absolute concentration and occupant expo­
no impact on exposure reduction, would have an impact on indoor sure in indoor environments with high particle concentrations. Thus,
particle concentration. Since health outcomes depend on the concen­ future studies should focus on the optimization of operation strategies to
trations, it is important to keep in mind that the same effectiveness value increase R without causing increases in energy consumption.
does not transfer to the same indoor concentration. Homes in more
heavily polluted regions or with stronger indoor emission sources will Declaration of competing interest
benefit more from filtration. Our analysis also relied on a relatively
narrow range of input parameters for North American homes only The authors declare that they have no known competing financial
(except the outdoor particle concentration) because of the prevalence of interests or personal relationships that could have appeared to influence
forced-air heating and cooling systems. Future modeling work should the work reported in this paper.
consider a wider range of parameters, and the use of portable air
cleaners, especially from homes in heavily polluted regions where Acknowledgments
exposure control is mostly needed to provide more information on the
exposure reduction benefits of filtration systems. The authors would like to thank NSERC (RGPIN-2014-06698) for

8
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

financial support. Tianyuan Li is partially supported by an Ontario Report, American Society of Heating, Refrigerating and Air-conditioning
Engineers, Atlanta, GA, 2019.
Graduate Scholarship.
[24] A. Novoselac, J.A. Siegel, Impact of placement of portable air cleaning devices in
multizone residential environments, Build. Environ. 44 (2009) 2348–2356,
Appendix A. Supplementary data https://doi.org/10.1016/j.buildenv.2009.03.023.
[25] R.J. Shaughnessy, R.G. Sextro, What Is an effective portable air cleaning device? A
review, J. Occup. Environ. Hyg. 3 (2006) 169–181, https://doi.org/10.1080/
Supplementary data to this article can be found online at https://doi. 15459620600580129.
org/10.1016/j.buildenv.2020.107389. [26] M.F. Touchie, J.A. Siegel, Residential HVAC runtime from smart thermostats:
characterization, comparison, and impacts, Indoor Air 28 (2018) 1–11, https://doi.
org/10.1111/ina.12496.
References [27] M.S. Waring, J.A. Siegel, R.L. Corsi, Ultrafine particle removal and generation by
portable air cleaners, Atmos. Environ. 42 (2008) 5003–5014, https://doi.org/
10.1016/j.atmosenv.2008.02.011.
[1] R.D. Brook, S. Rajagopalan, C.A. Pope, J.R. Brook, A. Bhatnagar, A.V. Diez-Roux,
[28] Y. Zhang, T. Li, J.A. Siegel, Investigating the impact of filters on long-term particle
F. Holguin, Y. Hong, R.V. Luepker, M.A. Mittleman, A. Peters, D. Siscovick, S.
concentration measurements in residences (RP-1649), Sci. Technol. Built Environ.
C. Smith, L. Whitsel, J.D. Kaufman, Particulate matter air pollution and
Online. (2020), https://doi.org/10.1080/23744731.2020.1778402.
cardiovascular disease: an update to the scientific statement from the American
[29] D. Zhao, P. Azimi, B. Stephens, D. Zhao, P. Azimi, B. Stephens, Evaluating the long-
heart association, Circulation 121 (2010) 2331–2378, https://doi.org/10.1161/
term health and economic impacts of central residential air filtration for reducing
CIR.0b013e3181dbece1.
premature mortality associated with indoor fine particulate matter (PM2.5) of
[2] N. Fann, S.Y. Kim, C. Olives, L. Sheppard, Estimated changes in life expectancy and
outdoor origin, Int. J. Environ. Res. Publ. Health 12 (2015) 8448–8479, https://
adult mortality resulting from declining PM2.5 exposures in the contiguous United
doi.org/10.3390/ijerph120708448.
States: 1980–2010, Environ. Health Perspect. 125 (2017), https://doi.org/
[30] W.R. Chan, J.M. Logue, X. Wu, N.E. Klepeis, W.J. Fisk, F. Noris, B.C. Singer,
10.1289/EHP507.
Quantifying fine particle emission events from time-resolved measurements:
[3] J. Lelieveld, J.S. Evans, M. Fnais, D. Giannadaki, A. Pozzer, The contribution of
method description and application to 18 California low-income apartments,
outdoor air pollution sources to premature mortality on a global scale, Nature 525
Indoor Air 28 (2018) 89–101, https://doi.org/10.1111/ina.12425.
(2015) 367–371, https://doi.org/10.1038/nature15371.
[31] S. Miller-Leiden, C. Lohascio, W.W. Nazaroff, J.M. Macher, Effectiveness of in-
[4] P. Saini, M. Sharma, Cause and Age-specific premature mortality attributable to
room air filtration and dilution ventilation for tuberculosis infection control, J. Air
PM2.5 Exposure: an analysis for Million-Plus Indian cities, Sci. Total Environ. 710
Waste Manag. Assoc. 46 (1996) 869–882, https://doi.org/10.1080/
(2020) 135230, https://doi.org/10.1016/j.scitotenv.2019.135230.
10473289.1996.10467523.
[5] J.A. Leech, W.C. Nelson, R.T. Burnett, S. Aaron, M.E. Raizene, It’s about time: a
[32] StataCorp, Stata Statistical Software, StataCorp LP, College Station, TX, 2015.
comparison of Canadian and American time–activity patterns, J. Expo. Sci.
[33] Z. El Orch, B. Stephens, M.S. Waring, Predictions and determinants of size-resolved
Environ. Epidemiol. 12 (2002) 427–432, https://doi.org/10.1038/sj.jea.7500244.
particle infiltration factors in single-family homes in the U.S. Build, Environ. Times
[6] C.J. Matz, D.M. Stieb, K. Davis, M. Egyed, A. Rose, B. Chou, O. Brion, Effects of age,
74 (2014) 106–118, https://doi.org/10.1016/j.buildenv.2014.01.006.
season, gender and urban-rural status on time-activity: Canadian human activity
[34] D.B. Shirey, H.I. Henderson, Dehumidification at part load, ASHRAE J. 46 (2004)
pattern survey 2 (CHAPS 2), Int. J. Environ. Res. Publ. Health 11 (2014)
42–49.
2108–2124, https://doi.org/10.3390/ijerph110202108.
[35] J.W. Thornburg, C.E. Rodes, P.A. Lawless, C.D. Stevens, R.W. Williams, A pilot
[7] U.S. Bureau of Labor Statistics, Table A-1. Time spent in detailed primary activities
study of the influence of residential HAC duty cycle on indoor air quality, Atmos.
and percent of the civilian population engaging in each activity, averages per day
Environ. 38 (2004) 1567–1577, https://doi.org/10.1016/j.atmosenv.2003.12.019.
by sex, 2018 annual averages [WWW Document], Am. Time Use Surv (2018). URL,
[36] B. Stephens, J.A. Siegel, A. Novoselac, Energy Implications of Filtration in
https://www.bls.gov/tus/#tables. accessed 3.19.20.
Residential and Light-Commercial Buildings (RP-1299), SHRAE Research Project
[8] C. Chen, B. Zhao, Review of relationship between indoor and outdoor particles: I/O
1649-RP Report. American Society of Heating, Refrigerating and Air-Conditioning
ratio, infiltration factor and penetration factor, Atmos. Environ. 45 (2011)
Engineers, ASHRAE, Atlanta, GA, 2010.
275–288, https://doi.org/10.1016/j.atmosenv.2010.09.048.
[37] W.J. Fisk, W.R. Chan, Health benefits and costs of filtration interventions that
[9] W.J. Riley, T.E. McKone, A.C.K. Lai, W.W. Nazaroff, Indoor particulate matter of
reduce indoor exposure to PM2.5 during wildfires, Indoor Air 27 (2017) 191–204,
outdoor origin: Importance of size-dependent removal mechanisms, Environ. Sci.
https://doi.org/10.1111/ina.12285.
Technol. 36 (2002) 200–207, https://doi.org/10.1021/es010723y.
[39] D.A. Jump, I.S. Walker, M.P. Modera, Field measurements of efficiency and duct
[10] D. Rim, L. Wallace, A. Persily, Infiltration of outdoor ultrafine particles into a test
retrofit effectiveness in residential forced air distribution systems, in: Proceedings
house, Environ. Sci. Technol. 44 (2010) 5908–5913, https://doi.org/10.1021/
of the 1996 ACEEE Summer Study on Energy Efficiency in Buildings. Presented at
es101202a.
the American Council for an Energy Efficient Economy, American Council for an
[11] B. Stephens, Building design and operational choices that impact indoor exposures
Energy Efficient Economy, Washington, DC, Pacific Grove, CA, 1996, pp. 147–155.
to outdoor particulate matter inside residences, Sci. Technol. Built Environ. 21
[40] B. Stephens, J.A. Siegel, A. Novoselac, Operational characteristics of residential
(2015) 3–13, https://doi.org/10.1080/10789669.2014.961849.
and light-commercial air-conditioning systems in a hot and humid climate zone,
[12] R.C. Brown, D. Wake, R. Gray, D.B. Blackford, G.J. Bostock, Effect of industrial
Build. Environ. 46 (2011), https://doi.org/10.1016/j.buildenv.2011.04.005,
aerosols on the performance of electrically charged filter material, Ann. Occup.
1972–83.
Hyg. 32 (1988) 271–294, https://doi.org/10.1093/annhyg/32.3.271.
[41] M. Alavy, T. Li, J.A. Siegel, Exploration of a long-term measurement approach for
[13] P.C. Raynor, S.J. Chae, Dust loading on electrostatically charged filters in a
air change rate, Build. Environ. 144 (2018) 474–481, https://doi.org/10.1016/j.
standard test and a real HVAC system, Filtr (2003) 35–39, https://doi.org/
buildenv.2018.08.051.
10.1016/S0015-1882(03)80068-X. Sep.
[42] D.M. Murray, D.E. Burmaster, Residential air exchange rates in the United States:
[14] J.T. Hanley, D.S. Ensor, D.D. Smith, L.E. Sparks, Fractional aerosol filtration
empirical and estimated parametric distributions by season and climatic region,
efficiency of in-duct ventilation air cleaners, Indoor Air 4 (1994) 169–178, https://
Risk Anal. 15 (1995) 459–465, https://doi.org/10.1111/j.1539-6924.1995.
doi.org/10.1111/j.1600-0668.1994.t01-1-00005.x.
tb00338.x.
[15] T. Li, J.A. Siegel, In situ efficiency of filters in residential central HVAC systems,
[43] L.A. Wallace, S.J. Emmerich, C. Howard-Reed, Continuous measurements of air
Indoor Air 30 (2020) 315–325, https://doi.org/10.1111/ina.12633.
change rates in an occupied house for 1 year: the effect of temperature, wind, fans,
[16] J.F. Montgomery, S.I. Green, S.N. Rogak, Impact of relative humidity on HVAC
and windows, J. Expo. Anal. Environ. Epidemiol. Princet. 12 (2002) 296–306.
filters loaded with hygroscopic and non-hygroscopic particles, Aerosol Sci.
http://10.1038/sj.jea.7500229.
Technol. 49 (2015) 322–331, https://doi.org/10.1080/02786826.2015.1026433.
[44] N. Yamamoto, D.G. Shendell, A.M. Winer, J. Zhang, Residential air exchange rates
[17] M. Alavy, T. Li, J.A. Siegel, Energy use in residential buildings: analyses of high-
in three major US metropolitan areas: results from the relationship among indoor,
efficiency filters and HVAC fans, Energy Build. 209 (2020) 109697, https://doi.
outdoor, and personal air study 1999–2001, Indoor Air 20 (2010) 85–90, https://
org/10.1016/j.enbuild.2019.109697.
doi.org/10.1111/j.1600-0668.2009.00622.x.
[18] M. Alavy, J.A. Siegel, In-situ effectiveness of residential HVAC filters, Indoor Air 30
[45] A. Persily, A. Musser, S.J. Emmerich, Modeled infiltration rate distributions for U.
(2020) 156–166, https://doi.org/10.1111/ina.12617.
S. housing, Indoor Air 20 (2010) 473–485, https://doi.org/10.1111/j.1600-
[19] P. Azimi, D. Zhao, B. Stephens, Modeling the impact of residential HVAC filtration
0668.2010.00669.x.
on indoor particles of outdoor origin (RP-1691), Sci. Technol. Built Environ. 22
[46] R. Williams, J. Suggs, A. Rea, L. Sheldon, C. Rodes, J. Thornburg, The Research
(2016) 431–462, https://doi.org/10.1080/23744731.2016.1163239.
Triangle Park particulate matter panel study: modeling ambient source
[20] S. Batterman, C. Godwin, C. Jia, Long duration tests of room air filters in cigarette
contribution to personal and residential PM mass concentrations, Atmos. Environ.
smokers’ homes, Environ. Sci. Technol. 39 (2005) 7260–7268, https://doi.org/
37 (2003) 5365–5378, https://doi.org/10.1016/j.atmosenv.2003.09.010.
10.1021/es048951q.
[47] W.J. Fisk, Health benefits of particle filtration, Indoor Air 23 (2013) 357–368,
[21] K.W. Brown, T. Minegishi, J.G. Allen, J.F. McCarthy, J.D. Spengler, D.L. MacIntosh,
https://doi.org/10.1111/ina.12036.
Reducing patients’ exposures to asthma and allergy triggers in their homes: an
[48] C.M. Long, H.H. Suh, P. Koutrakis, Characterization of indoor particle sources
evaluation of effectiveness of grades of forced air ventilation filters, J. Asthma 51
using continuous mass and size monitors, J. Air Waste Manag. Assoc. 50 (2000)
(2014) 585–594, https://doi.org/10.3109/02770903.2014.895011.
1236–1250, https://doi.org/10.1080/10473289.2000.10464154.
[22] W.J. Fisk, D. Faulkner, J. Palonen, O. Seppanen, Performance and costs of particle
[49] L.A. Wallace, F. Wang, C. Howard-Reed, A. Persily, Contribution of gas and electric
air filtration technologies, Indoor Air 12 (2002) 223–234, https://doi.org/
stoves to residential ultrafine particle concentrations between 2 and 64 nm: size
10.1034/j.1600-0668.2002.01136.x.
distributions and emission and coagulation rates, Environ. Sci. Technol. 42 (2008)
[23] T. Li, M. Alavy, Y. Zhang, J.A. Siegel, IAQ and Energy Implications of High-
8641–8647, https://doi.org/10.1021/es801402v.
Efficiency Filters in Residential Buildings, ASHRAE Research Project 1649-RP

9
T. Li and J.A. Siegel Building and Environment 187 (2021) 107389

[50] C.J. Weschler, H.C. Shields, Indoor ozone/terpene reactions as a source of indoor [54] D.L. MacIntosh, T.A. Myatt, J.F. Ludwig, B.J. Baker, H.H. Suh, J.D. Spengler, Whole
particles, Atmos. Environ. 33 (1999) 2301–2312, https://doi.org/10.1016/S1352- house particle removal and clean air delivery rates for in-duct and portable
2310(99)00083-7. ventilation systems, J. Air Waste Manag. Assoc. 58 (2008) 1474–1482, https://doi.
[51] Government of Ontario, Toronto downtown weather station [WWW document], org/10.3155/1047-3289.58.11.1474.
Minist. Environ. Conserv. Parks (2017). URL, http://www.airqualityontario.com [55] D.B. Shirey, H.I. Henderson, R.A. Raustad, Understanding the Dehumidification
/history/station.php?stationid=31103&submit_station=Choose+Station. accessed Performance of Air-Conditioning Equipment at Part-Load Conditions, 2006,
3.15.20. https://doi.org/10.2172/881342. No. 881342.
[52] U.S. Embassy & Consulates in India, Air quality data [WWW document], US [56] American Society of Heating, Refrigerating and Air-Conditioning Engineers,
embassy consulates India (2017). URL, https://in.usembassy.gov/embassy-consul ASHRAE Handbook: Fundamentals, 2009, American Society of Heating,
ates/new-delhi/air-quality-data/. accessed 3.15.20. Refrigerating and Air Conditioning Engineers, Atlanta, GA, 2009.
[53] T. Li, M. Alavy, J.A. Siegel, Measurement of residential HVAC system runtime, [57] T. Li, J.A. Siegel, Impacts of HVAC sequencing on runtime, in: Indoor Air 2018,
Build. Environ. 150 (2019) 99–107, https://doi.org/10.1016/j. 2018. Philadelphia, PA, USA.
buildenv.2019.01.004.

10

You might also like