Professional Documents
Culture Documents
Design and Laboratory Evaluation of An Inexpensive Noise Sensor
Design and Laboratory Evaluation of An Inexpensive Noise Sensor
Summer 2017
Recommended Citation
Hallett, Laura Ann. "Design and laboratory evaluation of an inexpensive noise sensor." MS (Master of
Science) thesis, University of Iowa, 2017.
https://doi.org/10.17077/etd.qun4tq56
DESIGN
AND
LABORATORY
EVALUATION
OF
AN
INEXPENSIVE
NOISE
SENSOR
by
Laura
Ann
Hallett
A
thesis
submitted
in
partial
fulfillment
of
the
requirements
for
the
Master
of
Science
degree
in
Occupational
and
Environmental
Health
in
the
Graduate
College
of
The
University
of
Iowa
August
2017
Thesis
Supervisor:
Professor
Thomas
M.
Peters
Copyright
by
Laura
Ann
Hallett
2017
All
Rights
Reserved
Graduate
College
The
University
of
Iowa
Iowa
City,
Iowa
CERTIFICATE
OF
APPROVAL
____________________________
MASTER'S
THESIS
_________________
This
is
to
certify
that
the
Master's
thesis
of
Laura
Ann
Hallett
has
been
approved
by
the
Examining
Committee
for
the
thesis
requirement
for
the
Master
of
Science
degree
in
Occupational
and
Environmental
Health
at
the
August
2017
graduation.
Thesis
Committee:
____________________________________________
Thomas
M.
Peters,
Thesis
Supervisor
____________________________________________
T.
Renée
Anthony
____________________________________________
Geb
W.
Thomas
ACKNOWLEDGEMENTS
I would first like to express a sincerest thank you to my advisor, Dr. Tom Peters. Thank
you for helping me realize my academic capabilities and encouraging me to keep going through
the cloud of uncertainty. I would have been unable to accomplish any of this without your
support. I would also like to thank my committee members, Renée Anthony and Geb Thomas. I
am appreciative of all the guidance that you have both shown me throughout my time here and
I feel privileged to have the opportunity to work with you.
To my low-‐cost project partners: Sinan, Alyson, Marcus, Mitch, Xiaoxing, Kirsten, Chris,
and Nima. It has been honor to work on this project with you the past 2 years. I enjoyed learning
from you all and seeing the project progress to where it is today. Marcus, this project would not
have been a success without you. Thank you for working so persistently and for always
answering my questions, even when they required last minute action on your part. Sinan, I am
so grateful for all your help with experiments and data analysis. I truly enjoyed working
Thank you to the faculty and staff in the OEH department that made my experience at
Iowa better than I could have ever imagined. Thank you to my lab mates and classmates for the
consistent feedback and friendship. I am fortunate to have built lasting relationships with so
many of you and hope that we can continue to encourage one another throughout our careers.
I would like to thank my family, friends, and derby team for all of their love and support.
You have all provided me much needed balance through this process, and I am lucky to have
Finally, this research was generously supported by The Heartland Center for
Occupational Safety and Health Training Grant No. T42OH008491 and funded by the National
ii
ABSTRACT
Noise is a pervasive workplace hazard that varies spatially and temporally. Hazard
mapping is a useful way to communicate intensity and distribution of noise sources in the
workplace. These maps can be created using a stationary network of sensors, although the cost
of noise measurement instruments has prohibited their use in such a network. The objectives
for this work were to (1) develop an inexpensive noise sensor (<$100) that measures A-‐weighted
sound pressure levels within ±2 dBA of a Type 2 sound level meter (SLM, ~$1,800); and (2)
evaluate 50 noise sensors before field deployment as part of an inexpensive sensor network.
The inexpensive noise sensor consists of an electret condenser microphone, an amplifier circuit,
and a microcontroller with a small form factor (28mm by 47 mm by 9 mm) than can be operated
as a stand-‐alone unit. Laboratory tests were conducted to evaluate 50 of the new sensors at 5
test levels. The testing levels were (1) ambient sound in a quiet office, (2) a pink noise test signal
from 65 to 85 dBA in 10 dBA increments, and, (3) 94 dBA using a SLM calibrator. The difference
between the output of the sensor and SLM were computed for each level and overall. Ninety-‐
four percent of the noise sensors (n=46) were within ± 2 dBA of the SLM for sound levels from
65 dBA to 94 dBA. As sound level increased, bias decreased, ranging from 18.3% in the quiet
office to 0.48% at 94 dBA. Overall bias of the sensors was 0.83% across the 75 dBA to 94 dBA
range. These sensors are available for a variety of uses and can be customized for many
iii
PUBLIC
ABSTRACT
Exposure to noise, particularly in occupational environments, can result in irreversible
hearing loss. Traditionally, occupational noise exposures are measured on a worker using
devices that cost over $1,500. A stationary network of these devices could be used to
continuously monitor and visualize how noise varies over space and time in the workplace,
helping to comprehensively assess the hazard and prioritize actions to reduce noise sources.
However, the cost of these devices limits the feasibility of their use. Availability of an
inexpensive, but accurate, noise sensor would alleviate some of the cost concerns.
There were two objectives to this study. The first was to develop an inexpensive noise
sensor that can detect noise levels within ±2 A-‐weighted decibels (dBA) of the more expensive
reference device. Second, fifty inexpensive noise sensors were evaluated prior to deployment as
The inexpensive noise sensor (<$100) performed similarly to the reference device
(~$1,800) across a 75 to 94 dBA range. These noise levels are typical of what is expected in the
heavy-‐vehicle manufacturing facility where they will be deployed. The versatility of these noise
sensors coupled with their accuracy is promising for future applications in different occupational
environments.
iv
TABLE
OF
CONTENTS
v
LIST
OF
TABLES
Table
1.
Mean
sound
pressure
levels,
standard
deviation,
and
coefficient
of
variation
at
five
target
dBA
levels
(side
by
side,
65,
75,
85,
94
dBA).
P-‐values
are
resultant
from
one-‐sided
t-‐test
with
an
alpha
of
0.05.
Acceptance
criteria
(AC)
is
defined
as
monitor
output
within
±2
dBA
of
reference
output,
n=46
monitors.
...............................................................................
30
Table
2.
Slope
and
intercept,
correlation,
bias
and
coefficient
of
variation
results
for
lab
checkout
of
noise
sensor
from
65
dBA
to
94
dBA
and
from
75
dBA
to
94
dBA,
n=
46
monitors.
30
vi
LIST
OF
FIGURES
Figure
1.
The
inexpensive
sound
sensor.
A
simplified
circuit
diagram
(top
panel)
shows
that
the
electrical
signal
from
the
microphone
is
amplified
twice
before
reaching
the
sensor
microcontroller
(Teensy).
The
actual
components
of
noise
sensor
incorporated
into
multi-‐
hazard
monitor
are
shown
in
the
bottom
panel.
An
electret
microphone
extends
from
the
exterior
of
the
grey
enclosure
and
the
sensor
microcontroller
is
shown
inside
the
enclosure.
On
the
red
circuit
board,
a
monitor
microcontroller
communicates
with
the
noise
and
other
hazard
sensors
and
with
a
database
via
WiFi.
..............................................................................
27
Figure
2.
Setup
for
laboratory
validation
of
noise
sensor.
NTI
XL2
and
noise
sensor
microphones
are
located
within
2.5
cm
of
one
another,
centered
30
cm
from
the
center
of
the
amplifier.
The
amplifier
is
connected
to
the
laptop
by
auxiliary
cable.
.................................
28
Figure
3.
Mean
A-‐weighted
sound
levels
from
noise
sensor
versus
reference
SLM
at
five
target
sound
intensities.
Points
represent
mean
reading
of
all
noise
sensor
and
reference
measurement
pairs.
Error
bars
represent
the
standard
deviation
of
sound
pressure
levels
at
each
target
sound
intensity,
n=46
sensors.
..................................................................................
29
Figure
4.
Box
plot
of
distributions
of
differences
in
sound
sensor
output
from
the
reference
SLM
at
five
target
sound
levels.
Error
bars
represent
the
distribution
of
sound
pressure
level
differences
at
each
target
sound
level,
n=46
sensors.
.................................................................
31
vii
Chapter
I:
Literature
Review
Overview
Noise is a prevalent occupational hazard, with the potential to irreversibly damage
worker hearing if not adequately controlled. One way to determine sources of noise in the
workplace is through the development of a hazard map. Hazard maps can be generated using a
handheld sound measurement device, a roving cart of direct reading instruments, or a network
of stationary monitors. Data collection with a roving cart is subject to temporal uncertainties,
making a network of stationary monitors attractive. However, direct-‐read instruments, like a
sound level meter (SLM), can be expensive (~$1,800), which severely limits opportunities to
Recent innovations in smartphone technology have led to the development of a number
of SLM applications (apps). Under certain conditions, some of these apps have shown accuracy
within ±2 decibels of a reference SLM (Kardous 2016). However, the microphones incorporated
into smartphones have been determined to have substantial amounts of variability in
performance, dependent on the software version and phone hardware. External microphones
can be connected to smartphones to increase accuracy and precision of the noise readings, but
this fix does not work for all phone types (Roberts 2016). Additionally, concerns with data
privacy, accuracy over time, phone battery life, and data export and storage limit the versatility
of these applications.
This thesis describes the development of an inexpensive noise sensor that is
independent of a smartphone. The noise sensor was designed to measure A-‐weighted sound
pressure levels within ±2 dBA of a reference SLM with component costs below $100. Then, 50
noise sensors were evaluated for accuracy and precision over a range from 65 to 94 dBA, sound
1
levels
consistent
with
their
intended
use
as
part
of
a
stationary
sensor
network
in
heavy-‐vehicle
manufacturing facility.
Noise
Workplace noise is unwanted sound produced in occupational environments (NIOSH, 1998).
Noise is one of the most common workplace hazards, prevalent in most occupational
environments (NIOSH, 1998). The effects of noise on human health depend on the amplitude,
Sound pressure levels are used to describe the amplitude of pressure change associated
with a specific sound, expressed in decibels (dB). Decibels are a conversion of this pressure
change relative to the hearing threshold pressure and are measured on a logarithmic scale
(OSHA, 2013). The amplitude of a sound directly relates to the listeners perception of loudness
(NIOSH, 1998). Frequency of a sound relates to the pitch perceived by the listener, expressed in
hertz (Hz). Humans with normal hearing are able to discern a frequency range of 20 Hz to 20
kilohertz (kHz). Human speech is typically generated at frequencies in the 500 Hz to 4,000 Hz
Noise rarely occurs as a pure tone, but rather consists of a number of different frequencies
occurring at different sound pressure levels. An octave band describes a range of frequencies in
which the highest frequency is twice the lowest frequency. Particular octave bands contain
frequencies that are more harmful to human hearing than others. For this reason, weighting can
be applied to sound pressure levels to understand the influence each frequency has on the
perception of loudness. There are three different types of weighting, A, C and Z. A-‐weighting
attenuates low frequencies, so that the sound pressure level output is indicative of that
perceived by a human (OSHA, 2013). When a sound pressure level has been modified using a
2
particular
weighting
network,
the
units
are
written
as
dBA
or
dBC,
the
last
letter
indicating
the
Duration of noise exposure is described as either continuous or impulse. Continuous noise is
characterized by small fluctuations in the overall noise level over the observation period.
Impulse, sometimes called impulsive, noise occurs for less than 1 second and results from a
sharp rise and rapid decay in sound levels (NIOSH, 1998). The effects of impulse noise on
hearing are dependent on the repetition rate, duration of exposure, and personal characteristics
that may enhance susceptibility. Impulse noise levels greater than 119 to 125 dB are considered
critical, and hearing loss is more prevalent with exposure to these sound levels (Dunn 1991). In
many industrial environments, there is a combination of both continuous and impulsive noise
Hearing loss can develop when there is a disruption in the ability of the ear to convert
physical vibration into a nerve impulse (Alberti 1992; WHO 2001). In the United States, an
estimated one in four adults has evidence of noise-‐induced hearing loss, as indicated by a notch
on audiometric test results. Additionally, one third of workers who identified as working in a
noisy environment had evidence of noise-‐induced hearing loss (Carrol 2017). The three main
types of hearing loss resulting from occupational noise exposures are conductive, sensorineural,
and mixed hearing loss (OSHA, 2013). Conductive hearing loss occurs when sound waves are
unable to pass through the outer or middle ear. Sensorineural hearing loss occurs with damage
to the physiological processes in the inner ear. Mixed hearing loss occurs when there is a
combination of both conductive and sensorineural hearing loss (CDC, 2015).
3
Over
22
million
workers
in
the
United
States
experience
occupational
exposures
to
hazardous noise each year (Tak, Davis, & Calvert, 2009). In 2014, the incidence of hearing loss in
all industries was 1.9 events per 10,000 full-‐time workers, with 20,900 reported cases of hearing
loss (BLS, 2016). Industries that have higher proportions of worker exposures to hazardous noise
levels include mining, lumber and wood product manufacturing; rubber, plastics, and leather
production; utilities, and repair and maintenance. While a lower proportion of workers in the
manufacturing industry are exposed to hazardous noise levels, the greatest number of workers
exposed to hazardous noise levels work in manufacturing, estimated at 5.7 million workers at
Regulations
A number of occupational exposure limits (OELs) have been developed to protect worker
hearing. The Occupational Safety and Health Administration (OSHA) sets legally enforceable
noise exposure limits. OSHA requires that employers implement a hearing conservation program
when employees are exposed to sound levels greater than 85 decibels (dB), A-‐weighted (A), as
an 8-‐hour time weighted average (TWA). An exchange rate of 5-‐dB is used for OSHA dose
calculations. OSHA also specifies a Permissible Exposure Limit (PEL) not to be exceeded of 90
dBA TWA (OSHA, 1974). If worker noise exposures exceed 90 dBA TWA, employers must provide
hearing protection devices to attenuate noise levels to at least the PEL.
The National Institute of Occupational Safety and Health’s (NIOSH) Recommended Exposure
Limit (REL) for noise is 85 dBA TWA over an eight-‐hour work shift. Additionally, NIOSH
recommends that no noise exposure, continuous or impulse, exceed 140 dBA. The NIOSH REL is
based upon a 3-‐dB exchange rate, which represents how sound intensity doubles, compared to
the less protective 5-‐dB used by OSHA (NIOSH, 1998; OSHA, 1974).
4
Hearing
Conservation
Programs
The most effective way to prevent hearing loss is to directly eliminate the noise hazard (Tak
et al., 2009). Elimination is not always the most feasible solution, so other programs can be
implemented to protect workers from noise hazards. Both OSHA and NIOSH recommend
hearing conservation programs when workers are exposed to sound levels louder than 85 dBA
TWA (NIOSH, 1998; OSHA, 1974). Exposure monitoring in these environments must include all
continuous, intermittent, and impulsive noise within an 80 to 130 dB range and be collected
during a typical working day (OSHA, 1974). OSHA 1910.95(c)(1) and 1910.95 Appendix A specify
the requirements for collection of noise dose information. There is no required instrument type,
but audio dosimeters are most commonly used to collect information on worker dose.
Another important component of a hearing conservation program is the establishment of an
audiometric testing program. Clinically, hearing loss presents itself as a reduction in the ability of
a person to recognize high frequency tones between 4000 to 6000Hz on pure tone audiometry
testing (Kurmis & Apps, 2007). In order to discern a loss in hearing ability, employers must
conduct baseline audiometric exams within 6 months of exposure to sound levels at or above
the action limit. Then, annual audiometric testing is performed to monitor hearing health and
ensure that the program is effective in preventing hearing loss (OSHA, 2013). Employees that
record an average shift in either ear of 10 decibels or more at 2000, 3000, and 4000Hz have
experienced an OSHA-‐defined standard threshold shift (STS). Per OSHA CFR 1904.10(a), if an
employee’s audiogram indicates that they have an STS in one or both ears, and the total hearing
level is 25 dB or more above audiometric zero in the same ear as the STS, the hearing loss is a
reportable illness. State worker’s compensation for hearing loss varies in frequency and
5
Evaluating
Noise
Levels
There are a number of instruments suited for noise exposure measurements. Most
commonly, a sound level meter or a dosimeter is used to determine worker noise exposures.
Sound level meters (SLM) measure sound pressure levels in units of dB. A SLM consists, at a
minimum, of three components: microphone, a frequency selective amplifier, and an indicator
(NIOSH, 1998). The American National Standards Institute specifies standards for performance
of SLMs in ANSI S1.4 and indicates three different grades of instrument: Type 0, 1, and 2. Type 0
SLMs are used as laboratory reference standards. Type 1 instruments are designed for field or
laboratory use and are allotted an overall error of +/-‐ 1.5 dB from the Type 0 instrument. Type 2
instruments, or general-‐purpose, are designed for field use and allowed an overall error of +/-‐
2.3 dB. General field measurements are assumed to be collected in environments in which high
frequency sounds do not dominate (ANSI, 1983). OSHA regulations require the use of at least a
Type 2 SLM for noise exposure measurements. If a SLM is used to determine personal noise
exposure, time-‐weighted averages must be computed using time and motion information on
workers to ensure that the exposure estimate accurately reflects worker experience of noise.
Dosimeters are the primary instrument used for noise exposure compliance measurements
(OSHA, 2013). Noise dosimeters measure sound pressure levels similarly to the SLM, then record
and process this information over the sampling period to compute the noise dose and TWA
experienced by the wearer (NIOSH, 1998). Different threshold and exchange rates can be
programmed into a dosimeter to measure noise dose based upon either OSHA or NIOSH criteria.
By attaching the microphone directly to the worker, the calculated dose is representative of the
noise experienced by the worker as they move throughout their work environment.
6
The
work
environment
can
inform
the
type
of
sampling
needed.
In
a
workplace
with
homogenous noise levels, area samples can be collected using a sound level meter, and worker
exposures estimated by the percentage of their workday they spend in each area. For workers
that have more variety in tasks or are exposed to a more complex environment, dosimeters are
a better tool to measure their exposure dose, as they are worn by the worker at all times (OSHA,
2013).
Hazard Maps
A method growing in popularity is the use of hazard maps to display hazard information
as it varies over space and time. Hazard maps provide an easily understandable, graphic
representation of hazard location and intensity in the workplace, aiding in communication of
workplace risks to management and workers. Noise mapping can aid in the understanding of
noise levels in context of sources. Within the OSHA technical manual, there are guidelines for
the creation of noise maps and the designation of the “hazard radius” (OSHA 2013).
Previous studies have generated hazard maps using a mobile cart equipped with a
number of direct-‐read instruments (Liu & Hammond, 2010; Peters et al., 2012). These
instruments have the ability to quickly provide information on contaminant levels. However,
portable direct-‐read instruments have the potential to suffer from a number of measurement
errors, including poor accuracy, lack of precision, and bias from interferences (Kirsten A Koehler
& Volckens, 2011). Additionally, generating hazard maps with this process requires considerable
time and personnel support. Collecting measurements point-‐by-‐point also reduces the ability to
make temporal inferences on hazard levels in the mapped area (Hakala, Kivela, Ihalainen,
7
Noise
hazard
maps
have
been
created
to
display
noise
hazards
in
both
environmental
and occupational applications. Environmental noise maps are generated to identify and monitor
noise sources present in the non-‐occupational environment. While there is no standard to
create environmental noise maps in The United States, the European Noise Directive
2002/49/EC requires the development of environmental noise maps every five years in each of
the EU member states. The goal of this mapping directive is to provide communities with the
information they need to preserve noise levels where they are not harmful and prioritize
controls for areas that exceed applicable exposure levels through the creation of strategic noise
maps (Directive, 2002/49/EC). Environmental noise maps are commonly created for areas
Occupational noise maps have been generated for a variety of different workplace
types, from rice mills in India to engine testing facilities in Colorado (Lake, Zhu, Wang, Volckens,
& Koehler, 2015; Prasanna Kumar, Dewangan, Sarkar, Kumari, & Kar, 2008). A mobile individual
measures sound pressure levels at each point on a grid, and then the measurements are
integrated in mapping software to generate hazard maps. This method provides a good visual
interpretation of the distribution of hazards in the workplace. However, generating a hazard
map with this method requires personnel support and time and can be cost prohibitive for a
large workplace. This method also provides a brief snapshot of sound levels, without capturing
The representativeness of a hazard map can differ dramatically, dependent on the path
that the researcher takes through the facility and the timing of the measurements (K. A. Koehler
& Peters, 2013; Lake et al., 2015). Hazard maps make communication of risks easier, but this
benefit is lost if the data were collected in a way that fails to capture the whole story. A study
conducted by Lake et al. (2015) found that noise hazard maps produced for workplaces with
8
steady
sound
levels
looked
similar,
whether
the
data
were
collected
using
static
or
roving
monitors. On the contrary, hazard maps produced for workplaces that had more variety in
sound levels and intermittent peak noises were very different, dependent on the method by
which the data were collected. If hazard maps are to be used to inform decisions on controls or
interventions, it is crucial that they are representative of the risks. An underestimation of risk
may lead to increased risk for the worker, while and overestimation may result in additional
Historically, noise hazard maps have been used to identify areas of a workplace that
have sound levels exceeding 85-‐dB. Within these identified hazard zones, all workers are
required to wear hearing protection devices. While a valuable tool to characterize noise in a
work environment with stable sound levels, collecting information based on a short duration
average does not provide adequate information on the variability of sound levels over a longer
time period. This method limits the true understanding of worker noise exposures. By mapping
sound levels over time, not just as snapshots, the temporal variability in sound levels can be
identified as they change over the work shift, day, or month.
Stationary networks of noise sensors have been used to collect information on
environmental noise pollution, particularly in urban areas. A study conducted by Santini,
Ostermaier, and Vitaletti (2008) investigated the feasibility of a stationary sensor network for
measurement of environmental sound levels, specifically along roadways. The investigators
found that the evaluated noise sensors could capture information on noise pollution levels, but
they are subject to errors based on directionality. A study by Filipponi, Santini, and Vitaletti
(2008) investigated the suitability of wireless sensor networks to capture information on noise
pollution and established requirements for a sensor network to comply with noise pollution
monitoring applications. The investigators recommend a dense network of sensors deployed
9
over
a
small
area,
like
the
façade
of
a
building,
to
collect
the
most
accurate
information
on
noise
pollution levels.
Many low-‐cost noise measurement devices have been developed and deployed for
environmental noise monitoring purposes. A study conducted by Van Renterghem et al. (2011)
investigated the suitability of a number of different microphones for environmental noise
monitoring, ranging in price from less than $1 to over $2,000 USD. The researchers found that a
number of inexpensive microphones placed in an outdoor setting were able to detect sound
pressure levels with only small additional error compared to the error experienced by the much
Electret microphones, like those evaluated in the Renterghem study, are frequently
incorporated into phones because of their low cost. The increased use of smart phones in recent
years has improved access to low-‐cost noise measurement applications (apps). An estimated
77% of Americans owned a smartphone in 2016, an increase from just 35% in 2011 (Pew
Research Center, 2017). Smart phone users can download noise measurement apps on their
phones and infer information on sound pressure levels. User-‐friendly interfaces make it possible
to collect noise survey information in many different locations, from the outdoor environment
A popular use for SLM apps is the generation of crowd-‐sourced environmental noise maps.
Nast, Speer, and Le Prell (2014), evaluated how accurate smart phone sound level meter apps
were for environmental noise monitoring. The investigators tested the accuracy of SLM apps
using six octave bands, generated at three intensities. The criterion for a good response from
the app was ±5 dB from the Type 1 reference SLM. The results showed differences in accuracy of
10
each
app
across
the
different
octave
bands
evaluated.
Murphy
and
King
(2016)
used
broadband
white noise to measure SLM app response at background, 50, 70 and 90 dBA sound levels. Mean
differences in app-‐reported sound levels from the reference SLM differed across the range, with
the poorest performance at the extremes of the range. A significant difference between the
types of phone platform used, iOS or Android, and the mean output at each sound level was
found, indicating possible inconsistencies in the microphone quality of each phone. The
investigators determined that these noise apps are potentially a good tool for use in crowd
sourced environmental noise monitoring, but should not be used to replace reference SLMs.
The potential for use of smartphone apps in occupational settings has been evaluated a
number of times by Kardous and Shaw (2014; 2016). The first study with this team evaluated the
accuracy of 62 smartphone apps for occupational noise measurements. SLM apps were
evaluated at sound levels from 65 to 94 dBA, indicative of the typical occupational noise
exposures encountered in the workplace. Apps were not calibrated prior to use. The
investigators identified two apps that had unweighted mean differences within 1 dB of the
reference SLM across all testing conditions. The investigators also found that smartphones
performed differently based on model, similar to those results shown by Murphy and King.
Three of the 62 tested apps were found to report mean differences within ±2 dBA of the
reference SLM, leading to the conclusion that these apps are adequate for noise assessments in
A more recent study by Roberts, Kardous, and Neitzel (2016), further investigated the
accuracy of smart phone apps, this time incorporating an external microphone and calibration
procedure. The apps were downloaded onto 5th generation Apple iPods, devices virtually
identical to the iPhone save the ability to communicate with a cellular network. The internal
smartphone microphones reported values exceeding their acceptance criterion of ±2dB
11
deviation
from
the
reference
microphone.
They
also
found
an
upper
noise
limit
around
90
dB
on
a number of the apps, limiting the potential to use the app in an occupational or loud
recreational environment. While the external microphones were found to improve
The hazards associated with noise exposure are well documented, and regulations and
best-‐practice recommendations exist to protect workers from occupational noise. A popular
way for employers to communicate noise hazards to workers is through the generation of
hazard maps. The research conducted on noise hazard mapping has indicated that the
representative of noise hazard maps is greatly dependent on the method of collecting sound
levels, the variability of sound levels in the workplace, and the specific path taken through the
area of interest. A method for mapping that can account for temporal variations would improve
understanding of how sound levels fluctuate over time in a workplace.
The advantages that increased static sound measurements can have on the
representativeness of noise maps for both occupational and environmental applications is clear.
In the past, mapping required the movement of a single, expensive SLM through a workplace.
This method fails to capture adequate information on the variability of noise sources over time.
Inexpensive microphones, like those incorporated into smartphones, are stable and accurate.
The advent of the smartphone and the abundance of sound level meter apps available to
consumers has provided hope that there are accurate and inexpensive ways to collect dynamic,
time-‐dependent information on sound levels over a large work area without purchasing an
expensive reference SLM. However, there are limitations that arise from the model of
smartphone used, version of app, and the condition of the smartphone. With the constant
12
development
of
these
applications
and
their
updates,
there
needs
to
be
a
standardized
way
to
evaluate app accuracy. Additionally, technology is rapidly changing, so it can be challenging for
app developers to keep up with phone operating system updates.
It would be valuable to separate the noise sensor from the smartphone and optimize
the circuitry to deploy in an occupational environment. By developing the noise sensor as a
stand-‐alone component, opportunities open to use the sensor technology for a variety of
The objectives of this thesis were twofold: (1) develop an inexpensive noise sensor that
measures A-‐weighted sound pressure levels within ±2 dBA of a reference SLM with component
costs below $100; and (2) evaluate 50 noise sensors for precision and accuracy prior to their
deployment as part of a stationary sensor network. The sensor developed and evaluated in this
study was optimized for a heavy-‐vehicle manufacturing environment, where sound levels were
13
Chapter
II:
Design
and
Laboratory
Evaluation
of
an
Inexpensive
Noise
Sensor
Introduction
Each year, over 22 million workers in the United States experience occupational
exposures to potentially hazardous sound levels (Tak et al., 2009). Occupational exposure to
hazardous sound levels can result in disabling hearing loss (Hong, Kerr, Poling, & Dhar, 2013).
Noise-‐induced hearing loss occurs when there is disruption in the ability of the ear to convert
physical vibration in the air into a nerve impulse (Alberti, 1992). Between 2003 and 2012, the
prevalence of hearing loss across all industries approached 13 percent ({Masterson, 2016 #116}.
In order to protect worker hearing, the Occupational Safety and Health Administration (OSHA)
sets a legally enforceable permissible exposure limit (PEL) of 90 decibels, A-‐weighted (dBA), as
an 8-‐hour time weighted average (TWA) (OSHA, 1974). Additionally, OSHA requires employers
to implement a hearing conservation program when sound levels exceed an action level of 85
dBA TWA.
Traditionally, occupational noise exposure is quantified using a dosimeter or a sound
level meter (SLM). These instruments can cost upwards of $2,000 USD. Dosimeters are affixed
to the worker’s collar and provide a percentage of the full noise dose experienced by the
worker. Sound level meters provide information on area noise levels and are primarily used for
screening purposes to determine where dosimetry should be performed or to designate “high
noise” areas. OSHA requires that sound level meters meet the American National Standards
Institute (ANSI) Standard S1.4, “Specifications for Sound Level Meters,” in which three different
types of SLM performances are identified. Type 2 meters are most commonly used in
occupational environments and have an accuracy of ±2 dBA, the minimum level of precision to
14
An
increasingly
common
way
to
visualize
hazards
in
the
workplace
and
determine
sources of noise is through the development of hazard maps. Hazard maps have be generated
with a handheld direct reading instrument, a roving cart of direct reading instruments, , or a
network of stationary sensors (Lake et al., 2015; Liu & Hammond, 2010; Peters et al., 2012).
Uncertainties can arise when measuring sound levels by mobile mapping, especially concerning
the temporal distribution of hazards. Systems that can map sound both spatially and over time,
such as a stationary sensor network, are more attractive (Lake et al., 2015). However, the price
of direct reading sound instruments, like a dosimeter or SLM, can be cost-‐prohibitive, severely
limiting the opportunity to collect simultaneous information from many locations necessary to
characterize a workplace.
microphones, like the electret condenser microphones incorporated into cellular phones, to
measure sound pressure levels. A number of SLM applications (apps) are available for download
onto personal smartphones. However, concerns have arisen to the applicability of these apps for
occupational environments. Studies assessing the validity of these apps have found that the
accuracy and precision is dependent upon the phone model and the software version (C. A.
Kardous & Shaw, 2014; Murphy & King, 2016; Nast et al., 2014). The variability in performance
of some apps may be improved with the installation of an external microphone, but this is not a
viable solution for all phone types (Roberts et al., 2016). Additionally, the battery life of a
smartphone is not sufficient to keep a computationally demanding app continuously running
While the advancements in smartphone SLM apps are promising, there are a number of
limitations to their use in a stationary sensor network. Roberts et al. (2016) found an upper limit
noise limit around 90 dBA on a number of the apps they evaluated. With the OSHA PEL for noise
15
set
at
90
dBA
TWA,
it
is
essential
that
a
sound
sensor
designed
for
use
in
an
occupational
environment provide accurate sound level readings at levels near and above the PEL.
Additionally, the reliance of these SLM apps on the smartphone increases the cost. Isolating the
microphone and computational components of the SLM from the phone would allow for the use
of this low-‐cost technology in a variety of different environments and for a variety of uses, like a
Thus, the primary objective of this work was to design a noise sensor that measures A-‐
weighted sound levels within 2 decibels of a Type 2 reference SLM with components costing less
than $100. The second objective was to evaluate 50 of these sensors prior to field deployment
Methods
Sensor Design
We developed the noise sensor to integrate into a multiple hazard monitor that includes
sensors for sound, gasses, aerosols, temperature, and relative humidity (Figure 1). These
monitors will be incorporated into a sensor network to map multiple hazards in an occupational
setting, specifically a heavy-‐vehicle manufacturing facility. Preliminary sound pressure levels in
this facility were all greater than 65 dBA. Thus, we optimized the sensor circuitry to measure 60
dBA and greater. Additional design criteria for the sensor include compact design, independence
The noise sensor consists of a microphone, amplifier circuitry, and a sensor
microcontroller. A 20 Hz to 20 kHz omnidirectional analog electret condenser microphone
(CMA-‐4544PF-‐W, CUI Inc, Tualatin, Oregon, USA) projects from the exterior of the monitor
enclosure (Figure 1). Analog voltage from the microphone passes through two amplifiers in
16
series
(4.93x)
and
is
then
acquired
by
a
sensor
microcontroller
(Teensy
3.2,
PJRC,
Sherwood,
Oregon, USA). The sensor microcontroller samples 1024 amplified voltages (N) over a 0.023-‐sec
period (∆t). The acquired sample voltages are then processed with Fast Fourier Transform (FFT)
to a power spectrum with 512 frequency bins, ranging from 0 to 22.05 kHz to encompass the
range of human hearing. The FFT code was developed using the Teensy Audio Library. An A-‐
weighting coefficient is then applied to each bin of the power spectrum to determine the A-‐
weighted power spectrum (XA). The A-‐weighted decibels for the sampling period is obtained by
!
! !
Signal Level in dBA = 10𝑙𝑜𝑔!" !
!!! 𝑋! 𝑘 + 𝐶
Equation
1
!∆!
The calibration constant was determined using a 1 kHz tone and side-‐by-‐side
comparison of the sensor output with a Type 2 SLM (XL2 Audio and Acoustic Analyzer, NTi
Audio, Tigard, Oregon, USA) and microphone (M4260, NTi Audio, Tigard, Oregon, USA). This SLM
was selected because it was compatible with a previously-‐developed data logging program. A
sample of three noise sensors were evaluated to determine if the same calibration constant
could be applied to all sensors. A calibration constant of 127.5 was applied to all sensors to
adjust the A-‐weighted frequency spectrum to the signal level in dBA.
The noise sensor (microphone and sensor microcontroller) can be operated as a stand-‐
alone device with a form factor of approximately 28 mm by 47 mm by 9 mm. In this work we
evaluated the noise sensor integrated into the multi-‐hazard monitor. Within the monitor, a
program on the monitor microcontroller (Seeeduino Cloud, Seeed Development Limited, San
Leandro, CA, USA) requests data from the sensor microcontroller once every 2.5 s, triggering the
sensor microcontroller to send the most recently calculated A-‐weighted decibel sound level.
17
This
2.5-‐s
data
can
be
accessed
directly
from
the
monitor
by
USB
connection.
For
purposes
of
the sensor network, the monitor microcontroller collects the 2.5-‐s data from the sensor
microcontroller, averages, and then sends 15-‐minute averages of the sound levels to a monitor
database.
Laboratory Evaluation
The sound pressure levels measured with 50 noise sensors embedded within the larger
monitors were compared to those measured with a SLM using a pink noise test signal. Each
noise sensor was tested individually inside an isolated office. The microphones of the SLM and
noise sensor were placed within 2.5 cm of each other and 30 cm from the center of an amplifier
(2311000000, Fender Musical Instruments Corporation, Scottsdale, AZ, USA). The amplifier was
connected to a laptop computer with an auxiliary cord. The standard operating procedure for
The SLM used for laboratory comparison was newly purchased at the start of
experiments and calibrated prior to each testing day using a 114 dB test tone generated by a
sound level meter calibrator (SCAL1356, General Tools & Instruments, Secaucus, New Jersey,
USA). The manufacturer stated performance range for the SLM and microphone is 29 dB to 144
dB with a resolution of 0.1 dB. For all experiments, the SLM was programmed to report the A-‐
weighted equivalent sound level (LAeq). No threshold sound level was designated in the SLM
settings.
Five target sound levels, ambient, 65 dBA, 75 dBA, 85 dBA, and 94 dBA, were generated
to test each noise sensor. These sound levels were selected to match the range used by Kardous
and Shaw (2014) in their evaluation of smartphone applications and to meet the needs of the
heavy-‐vehicle manufacturing project. For levels from 65 dBA to 75 dBA, pink noise was
18
produced
by
playing
a
computer
sound
file
(NTi
Audio
Test
Signals
for
Audio
and
Acoustic
Analyzers V1.0). The sound level was adjusted using the volume settings on the laptop and then
verified on the reference SLM before each testing period began. The 94-‐dBA tone was
generated using a sound level calibrator (SCAL1356, General Tools & Instruments, Secaucus,
New Jersey, USA). Each test tone was generated for 30 s before moving on to the next. Both the
noise sensor and SLM reported one sound pressure level measurement every 2 s over each 30-‐s
testing period, n=50 sensors x 5 test levels x 1 measurement every 2 s x 30 s= 3,750 paired
measurements.
Data Analysis
For each sensor at each test level, we calculated the mean and standard deviation of the
sound level reported by the sensor (dBA !"#!$! ; SD!"#!$% ) and the SLM (dBA !"# ; SD!"# ). The
difference between the means of the noise sensor and the SLM was compared to our
acceptance criterion of ±2 dBA. This criterion was adapted from ANSI standard S1.4 -‐ 1983
(R2006) for Type 2 SLMs as an indication of accuracy (ANSI, 1983).
Further analyses were conducted on sensors passing this acceptance criterion. For test
level over all sensors, we calculated the mean and standard deviation of individual sensor
means. As an indication of sensor precision, we calculated the coefficient of variation as:
!"!"#!$%
% CV!"#!$% = ∗ 100%
Equation
2
!"#!"#!$%
As an indicator of accuracy, we calculated bias for each sensor and test level as:
!"#!"#!$% !!"#!"#
% Bias = ∗ 100%
Equation
3
!"#!"#
19
For
each
sensor,
simple
linear
regression
analysis
was
conducted
to
determine
the
slope
and intercept between the sensor and SLM outputs. As an indicator of the correlation between
the sensor and SLM, the Pearson correlation coefficient was calculated.
To determine the overall performance of the noise sensors, data were also aggregated
over two ranges relevant to occupational exposure assessments (65 dBA to 94 dBA and 75 dBA
to 94 dBA). Overall statistical measures were calculated for coefficient of variation, bias, slope
and intercept as the mean of individual values. Calculations were conducted done using
Microsoft Excel.
A paired sample t-‐test was conducted to test the hypothesis that the noise sensor sound
level was equivalent to the SLM sound level at each testing level with an alpha of 0.05.
Significant p-‐values indicate a difference in the mean of the noise sensor from the SLM. All t-‐
tests were conducted on SAS (Cary, NC). Codes and outputs from the SAS procedures are
Results
Ninety-‐two percent of the noise sensors (n=46) passed our acceptance criterion of ±2
dBA deviation from the SLM from 75 dBA to 94 dBA. The noise sensors that did not meet criteria
were removed from the monitor pool, repaired, and retested prior to field deployment. In this
manuscript, we report only on the results from the initially accepted sensors.
The mean A-‐weighted sound level measurements of the noise sensor versus the SLM are
presented in Figure 3 and Table 1. At the ambient test level, the difference between the noise
sensor mean and SLM mean was nearly 12 dBA, the noise sensor over reporting. As the sound
level increased, the difference between mean sensor and SLM sound pressure levels decreased.
However, the mean measurement of the noise sensors was statistically different from that of
20
the
SLM
at
all
testing
levels
(p
<
0.05).
Variability
in
the
measurements
decreased
as
sound
level
increased, as indicated by the reduction in the standard deviations as the testing level increased.
Mean percent bias at the target sound levels ranged from 18.3% at ambient to 0.48% at
94 dBA, with monotonic decrease in bias as sound level increased (Figure 1). Percent bias close
to zero indicates only slight differences between the values reported by the inexpensive noise
sensor and the SLM. The percentage of monitors that met acceptance criteria also increased as
sound level increased. Very few of the noise sensors met acceptance criteria at ambient sound
levels, less than 60 dBA, where 100 percent of the monitors met acceptance criteria for 85 dBA
Overall performance statistics of the noise sensors across two different ranges of sound
pressure levels are presented in Table 2.While the 95% confidence intervals for both slopes
included unity, slope measures were improved in the more restricted 75 dBA to 94 dBA range
(0.98 ± 0.03 dBA) as compared to the 65 dBA to 94 dBA range (0.94 ± 0.05 dBA). This result
indicates that the noise sensor was in better agreement with the SLM at sound pressure levels
greater than 75 dBA. The mean correlation between the sensor and SLM was slightly stronger
from 75 dBA to 94 dBA (1.00 ± 0.001) as compared to the 65 dBA to 94 dBA (1.00 ± 0.004) test
range. The mean percent bias of the noise sensors improved from 1.47% to 0.83% when the 65
dBA testing level was excluded from the range. The mean coefficient of variation, CV, improved
from 1.18% to 0.76% when the 65 dBA testing level was excluded from the range.
A box plot of the distributions of differences between the reference SLM and the noise
sensor at each target testing level is presented in Figure 4. Differences between the output of
the noise sensor and the reference SLM ranged from nearly 25 dBA at ambient sound levels, to
less than 2 dBA at the higher target noise levels. The range of differences decreased as noise
21
level
increased,
similar
to
the
trend
observed
in
Figure
3.
A
difference
of
zero
between
the
noise
sensor sound level and the SLM sound level is indicative of agreement between the two devices.
Discussion
The inexpensive noise sensor (~$30 for components) provided similar sound pressure
level measurements in dBA to a substantially more expensive, Type 2 reference SLM (~$1,800)
from 75 to 94 dBA. The stand-‐alone nature of the noise sensor developed in this work with
embedded measurement and processing on-‐board, coupled with their small size and low cost,
offers great potential for use in a variety of applications. Ninety-‐two percent of the noise
sensors were within ±2 dBA of the reference SLM from 65 to 94 dBA. This range encompasses
the sound levels expected in manufacturing, for which the noise sensors were designed.
The accuracy of the noise sensors, represented as mean bias compared to the reference
SLM was 1.47% from 65 to 94 dBA. Bias further reduced to 0.83% across 75 to 94 dBA. The bias
determined in this study is equivalent to less than ±1 dBA deviation from the SLM across the 75
to 94 dBA range. Kardous and Shaw (2016), found similar accuracy for two SLM apps with the
addition of an external microphone, SoundMeter and SPL pro. Differences for these apps from
the reference were between ±1 dBA across the 65 dBA to 95 dBA test range. SPLnFF and Noisee
apps had wider variations in sound pressure levels, especially at 65 dBA and 75 dBA (Chucri A.
Kardous & Shaw, 2016). Nast, Speer, and Le Prell (2016) found a number of SLM apps that
reported differences ranging from 3 to 10 dBA higher than the Type 1 SLM across their selected
testing frequencies.
The reference SLM equipped with a Type 2 microphone has an allowable error of ±2 dBA
deviation from a Type 0 laboratory standard. The noise sensor designed in this study performed
well compared to the Type 2 reference, but this reference may be off by 2 dBA from the actual
22
sound.
Our
sensor
introduces
additional
uncertainty
and
bias
(1.15%
for
sound
levels
greater
than 65 dBA) that must be considered. While the designed noise sensor is not intended for use
in determining compliance, if decisions were to be made regarding worker health a +4 dBA
safeguard should be applied to the reported sound pressure level output.
The sensor response at sound levels lower than 75 dBA was not consistently within our
acceptance criterion. There was a large distribution in the differences between the sensor and
SLM at ambient and 65 dBA target levels, as observed in Figure 4. As the sound level increased,
the spread of the distribution of differences between the sensor and SLM output decreased.
Preliminary measurements collected at the heavy-‐vehicle manufacturing facility that these
sensors were designed for resulted in sound level measurements consistently greater than 75
dBA. Additionally, sound levels less than 80 dBA are not of primary concern for controlling noise
hazards that may cause noise-‐induced hearing loss. The sensors are within our acceptance
criterion for the sound range they were customized for and perform similarly to the Type 2 SLM.
Modifications can be made to the sensor circuitry to allow for application in a variety of
workplaces with differing sound levels. The circuitry of these sensors could be customized using
high or low-‐pass filters to improve accuracy at the tail ends of the range. While sound levels less
than 80 dBA are not a primary concern for noise exposure in the workplace, they may be
relevant for other environments, such as patient care in a hospital. Additional refinement could
be made to improve the accuracy over the entire 65 to 94 dBA range, or a custom range for a
desired environment. Adjustments can be made to the sample period of the FFT to collect
voltages according to the specifications for impulse, slow and fast weighting.
Inexpensive microphones, like the electret condenser microphone used in this noise
sensor, are incorporated into a variety of consumer products, namely smartphones. Some SLM
23
applications
on
phones
have
shown
good
agreement
with
reference
SLMs,
especially
with
the
introduction of an external microphone (Roberts et al., 2016). However, the accuracy of these
applications varies significantly with smartphone and app software as well as smartphone
hardware (Roberts et al., 2016). Additionally, a number of studies have shown that SLM apps
may not be adequate for noise assessments in occupational environments exceeding 90 dBA
(Murphy & King, 2016; Roberts et al., 2016). The noise sensor evaluated in this study showed
excellent agreement with the reference SLM at sound levels higher than 75 dBA, indicating that
it is more acceptable for use in a noisy work environment than the previously evaluated SLM
apps.
While a number of accurate SLM apps have been identified, there are limitations to
their deployment into the workplace for noise monitoring. The smartphone required to utilize
the app can be expensive and bulky, posing cost and functionality challenges. Moreover, the
SLM apps are computationally demanding, and require a great deal of battery power.
Additionally, concerns have been raised regarding data privacy, accuracy over time, and data
export and sharing (Chucri A. Kardous & Shaw, 2016; Nast et al., 2014). Removing the noise
sensor from the smartphone alleviates a number of the concerns associated with SLM apps and
opens the door for application to other projects beyond a stationary monitor network. For
instance, the small form factor offers potential for repackaging the device into a wearable,
personal model. Additionally, the availability of FFT output enables access to data required to
provide the end user with octave band information that may help identifying determinants of
exposure rather than just sound level. For this study, the noise sensor was incorporated into a
multi-‐hazard monitor. The noise sensor components could stand alone with the addition of a
power source and data storage capabilities. The microcontroller selected cannot accommodate
24
a
USB
flash
drive,
but
an
alternative
microcontroller
with
USB
storage
capacities
could
be
The purpose of this inexpensive noise sensor is to provide reasonably accurate sound
pressure level measurements in real-‐time so that an occupational health professional can collect
more spatial and temporal information on sound levels in their facility. The importance of
temporal and spatial resolution for the representativeness of hazard maps has been established,
especially in non-‐homogenous workplaces (Lake et al., 2015). A clearer understanding of the
variations in sound levels throughout the facility and across shifts with a stationary monitor
network will allow for better allocation of resources when conducting personal noise exposure
measurements. However, the cost of SLMs has been prohibitive to the widespread applicability
of a stationary monitor network. The noise sensor we evaluated provides a relatively accurate
and inexpensive alternative to the reference SLM, and is versatile for use in a variety of
There were several limitations in this study. Testing of the noise sensors took place in a
quiet office, not a reverberant chamber as other noise sensor testing studies have used (C. A.
Kardous & Shaw, 2014; Murphy & King, 2016; Roberts et al., 2016). Reverberant chambers
create a diffuse sound field, removing the influence that positioning or reverberation off
surfaces can have on the sound pressure level readings. In our study, the microphones of the
noise sensor and reference SLM were setup close to each other to limit the influence of position
and reverberation. The pink noise test signal selected for this study allowed insight into the
sensor performance over a range of frequencies, but is not a perfect surrogate for real world
noise (Nast et al., 2014). Additional evaluation of the sensor in an occupational environment
should be conducted to account for differences in temperature, humidity, increased
intermittent noise, and object interference-‐ variables not present during the laboratory testing.
25
These
noise
sensors
are
also
not
individually
calibrated
prior
to
use.
The
calibration
offset,
added on the sensor microcontroller during development of the sensor, improved accuracy of
the sensors and allowed for consistent measurements of noise levels within 2 dBA of the
reference SLM. Future work may look to incorporate a calibration procedure for individual noise
sensors, further improve the accuracy of the sound measurements across the target range and
Conclusions
A new, inexpensive noise sensor (~$30 for components) developed in this work
responded similarly to a substantially more expensive reference SLM from 75 dBA to 94 dBA.
The independence of this noise sensor from a smartphone, coupled with the small form factor
and low price, allows for use in a variety of applications, including incorporation into a sensor
network. Increasing the number of measurement points in a stationary network will improve the
representativeness of hazard maps and better inform decisions on where further sampling
should occur. Future work may look to optimize the noise sensor for different work
environments or further evaluate the sensor performance in field conditions.
26
1@I78<?!69B!-64;<?!
! !
!
!
4<=>?8!9:!-><!@9<VC<9?@J<!?E79B!?<9?E8R!(!?@AC;@D@<B!H@8H7@=!B@6I86A!c=EC!C69<;d!?>EL?!=>6=!=><!
<;<H=8@H6;!?@I96;!D8EA!=><!A@H8EC>E9<!@?!6AC;@D@<B!=L@H<!4<DE8<!8<6H>@9I!=><!?<9?E8!
A@H8EHE9=8E;;<8!c-<<9?5dR! -><!6H=76;!HEACE9<9=?!ED!9E@?<!?<9?E8!@9HE8CE86=<B!@9=E!A7;=@[
>6Z68B!AE9@=E8!68<!?>EL9!@9!=><!4E==EA!C69<;R!(9!<;<H=8<=!A@H8EC>E9<!<V=<9B?!D8EA!=><!
<V=<8@E8!ED!=><!I8<5!<9H;E?78<!69B!=><!?<9?E8!A@H8EHE9=8E;;<8!@?!?>EL9!@9?@B<!=><!<9H;E?78<R!+9!
=><!8<B!H@8H7@=!4E68BS!6!AE9@=E8!A@H8EHE9=8E;;<8!HEAA79@H6=<?!L@=>!=><!9E@?<!69B!E=><8!>6Z68B!
?<9?E8?!69B!L@=>!6!B6=646?<!J@6!X@1@R!!
! !
MP !
!
!
!
4<=>?8!;:!$<=7C!DE8!;64E86=E85!J6;@B6=@E9!ED!9E@?<!?<9?E8R!'-%!2)M!69B!9E@?<!?<9?E8!A@H8EC>E9<?!
68<!;EH6=<B!L@=>@9!MRa!HA!ED!E9<!69E=><8S!H<9=<8<B!bN!HA!D8EA!=><!H<9=<8!ED!=><!6AC;@D@<8R!-><!
6AC;@D@<8!@?!HE99<H=<B!=E!=><!;6C=EC!45!67V@;@685!H64;<R!
! !
M_ !
!
!
9PP!
TO!
TP!
.8IHK?!.K>IJ!(8F87U!J/"!
SO!
SP!
RO!
RP!
QO!
QP!
OO!
OP!
OP! OO! QP! QO! RP! RO! SP! SO! TP! TO! 9PP!
08M8?8IV8!.K>IJ!(8F87U!J/"!
!
4<=>?8!@:!G<69!([L<@I>=<B!?E79B!;<J<;?!D8EA!9E@?<!?<9?E8!J<8?7?!8<D<8<9H<!$)G!6=!D@J<!=68I<=!
?E79B!@9=<9?@=@<?R!3E@9=?!8<C8<?<9=!A<69!8<6B@9I!ED!6;;!9E@?<!?<9?E8!69B!8<D<8<9H<!
A<6?78<A<9=!C6@8?R!#88E8!468?!8<C8<?<9=!=><!?=69B68B!B<J@6=@E9!ED!?E79B!C8<??78<!;<J<;?!6=!
<6H>!=68I<=!?E79B!@9=<9?@=5S!9k^j!?<9?E8?R!
! !
M` !
!
Table
1.
Mean
sound
pressure
levels,
standard
deviation,
and
coefficient
of
variation
at
five
target
dBA
levels
(side
by
side,
65,
75,
85,
94
dBA).
P-‐values
are
resultant
from
one-‐sided
t-‐test
with
an
alpha
of
0.05.
Acceptance
criteria
(AC)
is
defined
as
monitor
output
within
±2
dBA
of
reference
output,
n=46
monitors.
Table
2.
Slope
and
intercept,
correlation,
bias
and
coefficient
of
variation
results
for
lab
checkout
of
noise
sensor
from
65
dBA
to
94
dBA
and
from
75
dBA
to
94
dBA,
n=
46
monitors.
30
!
! !
4<=>?8!A:!*EV!C;E=!ED!B@?=8@47=@E9?!ED!B@DD<8<9H<?!@9!?E79B!?<9?E8!E7=C7=!D8EA!=><!8<D<8<9H<!$)G!
6=!D@J<!=68I<=!?E79B!;<J<;?R!#88E8!468?!8<C8<?<9=!=><!B@?=8@47=@E9!ED!?E79B!C8<??78<!;<J<;!
B@DD<8<9H<?!6=!<6H>!=68I<=!?E79B!;<J<;S!9k^j!?<9?E8?R!
! !
bO !
!
Chapter
III:
Conclusions
The inexpensive noise sensor (~$30) designed in this study performed similar to a Type 2
sound level meter (SLM, ~$1,800) over a test range from 65 dBA to 94 dBA. Ninety-‐two percent
of the sensors passed our acceptance criterion, reporting sound pressure level readings within
±2 dBA of the SLM across the test range. The passing sensors will be deployed as part of a
stationary sensor network in a heavy-‐vehicle manufacturing facility. These sensors will collect
information on the temporal and spatial variability of noise levels in the workplace and provide
measurements to help inform the allocation of control and exposure assessment resources.
This noise sensor was developed as a potential solution for the lack of spatial and
temporal information in workplace noise hazard maps. The high cost of traditional direct-‐
reading instruments places budgetary limits on the installation of a stationary network of
monitors. Inexpensive noise measurement applications (apps) have been developed for
smartphones and provide reasonably accurate noise level measurements, especially with the
installation of an external, calibrated microphone. However, there are limitations to the use of
smartphones in a stationary network, such as data privacy, accuracy over time, and data export
and sharing (Chucri A. Kardous & Shaw, 2016; Nast et al., 2014). The inexpensive noise sensor
developed in this work is independent of a smartphone with a small form factor, offering
There were a number of limitations in this study. Other noise sensor testing studies have
conducted their evaluation inside of a reverberant chamber, while our testing took place in a
quiet office environment. Reverberant chambers create a diffuse sound field, removing
potential for position and reverberation to affect results. To address this concern, we placed the
32
microphones
of
the
sensor
and
SLM
within
2
cm
of
each
other,
and
positioned
the
microphones
The selected pink noise test signal, while providing an insight into sensor performance,
is not a perfect surrogate for real world noise. Workplace noise comes from various sources
ranging in frequency, intensity, and duration (NIOSH, 1998). Additionally, environmental
conditions in a factory are not comparable to those in the testing office. Differences in
temperature and humidity in the heavy-‐vehicle manufacturing facility could have an impact on
the sensitivity of the inexpensive microphone incorporated into the sensor. Field performance
of these sensors should be evaluated to ensure that the measurements accurately measure real
The importance of pre-‐ and post-‐use calibration for noise measurement equipment is
well defined in CFR 1910.95(d)(2)(ii). The calibration offset, added on the sensor
microcontroller during development of the sensor, improved accuracy of the sensors and
allowed for consistent measurements of noise levels within 2 dBA of the reference SLM. A
calibration constant unique to each individual sensor may further improve accuracy. However,
current sensor performance is adequate for use in a stationary sensor network and to provide
insight as a surveying tool. Data collected by these sensors can be aggregated to produce hazard
maps and bring attention to areas of the facility where intervention is most needed.
The versatility of these noise sensors, coupled with their low price and small form
factor, offers potential for their use in projects beyond a stationary monitoring network. For
example, the device could be repackaged from the stationary monitor into a wearable personal
model. Additionally, access to the FFT output allows enables access to octave band information
that may be useful in identifying determinants of exposure, rather than just noise level. The
33
circuitry
can
be
optimized
using
high
or
low-‐pass
filters
to
capture
information
on
noise
ranges
of interest. The FFT sample period can also be modified according to the specifications for slow,
The performance of these noise sensors from 75 dBA to 94 dBA is promising for the
future of occupational noise assessment. Collecting more information on the noise hazards
present in the workplace allows us to make the best-‐informed decision to protect worker
health. Incorporating inexpensive technology, like the sensor developed in this study, into a
stationary network of monitors will allow for the collection of real-‐time data on hazard levels.
With this information, employers can feel well informed in their decision-‐making and cater their
exposure assessment and hazard abatement plans to what is most needed for their facility.
Throughout this project, I have learned a great deal about the importance of efficiency.
The small inefficiencies that are not readily noticeable during the evaluation of one sensor
become exceptionally clear when you scale up to 50 sensors. Designing data collection and
storage processes with simplicity in mind will make the acquisition of data from many low-‐cost
Another lesson learned from this project was bridging communication gaps between
departments. Working closely with the engineering department on the software and hardware
of the sensor allowed us to ensure the sensor was designed with scientific principles in mind.
However, it was often challenging to agree on what the next step should be, given the diverse
backgrounds of the project team. I learned to clearly state my opinions and back them with
evidence. This skill, along with the lessons learned in simplifying, will help me to become a more
effective and efficient communicator and researcher in my future career.
34
References
Alberti, P. W. (1992). "Noise induced hearing loss." British Medical Journal 304(6826): 522-‐522.
ANSI,
American
National
Standards
Institute.
(1983).
ANSI
S1.4:
Specifications
for
Sound
Level
Meters.
7
CFR
1755.22(s)(3)(v).
BLS,
United
States
Bureau
of
Labor
Statistics.
(2016).
Table
6.
Incidence
rates
and
numbers
of
nonfatal
occupational
illnesses
by
major
industry
sector,
category
of
illness,
and
ownership,
2014.
Carrol,
Y.I.,
Eichwald,
J.,
Scinicariello,
F.,
Hoffman,
H.J.,
Deitchman,
S.,
Radke,
M.S.,
Themann,
C.L.,
Breysse,
P.
(2017)
"Vital
Signs:
Noise-‐Induced
Hearing
Loss
Among
Adults
-‐
United
States
2011-‐2012."
MMWR
Morbidity
and
Mortality
Weekly
Report
2017;66:139-‐144.
CDC,
Centers
for
Disease
Control
and
Prevention.
(2015).
"Types
of
Hearing
Loss."
from
https://www.cdc.gov/ncbddd/hearingloss/types.html.
Directive,
C.
(2002/49/EC).
Declaration
by
the
Commission
in
the
Conciliation
Committee
on
the
Directive
relating
to
the
assessment
and
management
of
environmental
noise.
Dunn,
D.
E.,
R.
R.
Davis,
C.
J.
Merry
and
J.
R.
Franks
(1991).
"Hearing
loss
in
the
chinchilla
from
impact
and
continuous
noise
exposure."
The
Journal
of
the
Acoustical
Society
of
America
90(4):
1979-‐1985.
Filipponi,
L.,
S.
Santini
and
A.
Vitaletti
(2008).
Data
Collection
in
Wireless
Sensor
Networks
for
Noise
Pollution
Monitoring.
Distributed
Computing
in
Sensor
Systems:
4th
IEEE
International
Conference,
DCOSS
2008
Santorini
Island,
Greece,
June
11-‐14,
2008
Proceedings.
S.
E.
Nikoletseas,
B.
S.
Chlebus,
D.
B.
Johnson
and
B.
Krishnamachari.
Berlin,
Heidelberg,
Springer
Berlin
Heidelberg:
492-‐497.
Hakala,
I.,
I.
Kivela,
J.
Ihalainen,
J.
Luomala
and
C.
Gao
(2010).
Design
of
Low-‐Cost
Noise
Measurement
Sensor
Network:
Sensor
Function
Design.
2010
First
International
Conference
on
Sensor
Device
Technologies
and
Applications.
Hong,
O.,
M.
J.
Kerr,
G.
L.
Poling
and
S.
Dhar
(2013).
"Understanding
and
preventing
noise-‐
induced
hearing
loss."
Disease-‐a-‐Month
59(4):
110-‐118.
Kardous,
C.
A.
and
P.
B.
Shaw
(2014).
"Evaluation
of
smartphone
sound
measurement
applications."
J
Acoust
Soc
Am
135(4):
EL186-‐192.
Kardous,
C.
A.
and
P.
B.
Shaw
(2016).
"Evaluation
of
smartphone
sound
measurement
applications
(apps)
using
external
microphones
–
A
follow-‐up
Study."
The
Journal
of
the
Acoustical
Society
of
America
140(4):
EL327-‐EL327.
Koehler,
K.
A.
and
T.
M.
Peters
(2013).
"Influence
of
analysis
methods
on
interpretation
of
hazard
maps."
Ann
Occup
Hyg
57(5):
558-‐570.
35
Koehler,
K.
A.
and
J.
Volckens
(2011).
"Prospects
and
pitfalls
of
occupational
hazard
mapping:‘between
these
lines
there
be
dragons’."
Annals
of
occupational
hygiene
55(8):
829-‐
840.
Kurmis,
A.
and
S.
Apps
(2007).
Occupationally-‐Acquired
Noise-‐Induced
Hearing
Loss:
A
Senseless
Workplace
Hazard.
International
Journal
of
Occupational
Medicine
and
Environmental
Health.
20:
127.
Lake,
K.,
J.
Zhu,
H.
Wang,
J.
Volckens
and
K.
A.
Koehler
(2015).
"Effects
of
Data
Sparsity
and
Spatiotemporal
Variability
on
Hazard
Maps
of
Workplace
Noise."
Journal
of
Occupational
and
Environmental
Hygiene
12(4):
256-‐265.
Liu,
S.
and
S.
K.
Hammond
(2010).
"Mapping
particulate
matter
at
the
body
weld
department
in
an
automobile
assembly
plant."
J
Occup
Environ
Hyg
7(10):
593-‐604.
Maisonneuve,
N.,
M.
Stevens,
M.
E.
Niessen
and
L.
Steels
(2009).
NoiseTube:
Measuring
and
mapping
noise
pollution
with
mobile
phones.
Information
Technologies
in
Environmental
Engineering:
Proceedings
of
the
4th
International
ICSC
Symposium
Thessaloniki,
Greece,
May
28-‐29,
2009.
I.
N.
Athanasiadis,
A.
E.
Rizzoli,
P.
A.
Mitkas
and
J.
M.
Gómez.
Berlin,
Heidelberg,
Springer
Berlin
Heidelberg:
215-‐228.
Masterson,
E.A.,
Bushnell,
P.T.,
Themann,
C.L.,
Morata,
T.C.
(2016)
Hearing
Impairment
Among
Noise-‐Exposed
Workers
—
United
States,
2003–2012.
MMWR
Morb
Mortal
Wkly
Rep
2016;65:389–394.
DOI:
http://dx.doi.org/10.15585/mmwr.mm6515a2
Murphy,
E.
and
E.
A.
King
(2016).
"Testing
the
accuracy
of
smartphones
and
sound
level
meter
applications
for
measuring
environmental
noise."
Applied
Acoustics
106:
16-‐22.
Nast,
D.,
W.
Speer
and
C.
Le
Prell
(2014).
"Sound
level
measurements
using
smartphone
"apps":
Useful
or
inaccurate?"
Noise
and
Health
16(72):
251-‐256.
NIOSH,
National
Institute
for
Occupational
Safety
and
Health.
(1998).
Criteria
for
a
Recommended
Standard-‐
Occupational
Noise
Exposure.
U.
S.
D.
o.
H.
a.
H.
Services.
Cincinnati,
Ohio.
OSHA,
Occupational
Safety
and
Health
Administration.
(1974).
Occupational
Noise
Exposure.
29.
1910.95.
OSHA, Occupational Safety and Health Administration. (2013). OSHA Technical Manual.
Peters,
T.
M.,
T.
R.
Anthony,
C.
Taylor,
R.
Altmaier,
K.
Anderson
and
P.
T.
O'Shaughnessy
(2012).
"Distribution
of
Particle
and
Gas
Concentrations
in
Swine
Gestation
Confined
Animal
Feeding
Operations."
Annals
of
occupational
hygiene
56(9):
1080-‐1090.
Pew
Research
Center,
P.
(2017).
"Mobile
Fact
Sheet."
from
http://www.pewinternet.org/fact-‐
sheet/mobile/#.
Prasanna
Kumar,
G.,
K.
Dewangan,
A.
Sarkar,
A.
Kumari
and
B.
Kar
(2008).
"Occupational
noise
in
rice
mills."
Noise
and
Health
10(39):
55-‐67.
36
Roberts,
B.,
C.
Kardous
and
R.
Neitzel
(2016).
"Improving
the
accuracy
of
smart
devices
to
measure
noise
exposure."
Journal
of
Occupational
and
Environmental
Hygiene
13(11):
840-‐846.
Santini,
S.,
B.
Ostermaier
and
A.
Vitaletti
(2008).
First
experiences
using
wireless
sensor
networks
for
noise
pollution
monitoring.
Proceedings
of
the
workshop
on
Real-‐world
wireless
sensor
networks.
Glasgow,
Scotland,
ACM:
61-‐65.
Tak,
S.,
R.
R.
Davis
and
G.
M.
Calvert
(2009).
"Exposure
to
hazardous
workplace
noise
and
use
of
hearing
protection
devices
among
US
workers-‐-‐NHANES,
1999–2004."
Am
J
Ind
Med
52.
Van
Renterghem,
T.,
P.
Thomas,
F.
Dominguez,
S.
Dauwe,
A.
Touhafi,
B.
Dhoedt
and
D.
Botteldooren
(2011).
"On
the
ability
of
consumer
electronics
microphones
for
environmental
noise
monitoring."
Journal
of
Environmental
Monitoring
13(3):
544-‐552.
WHO,
World
Health
Organization.
(2001).
"Occupational
exposure
to
noise:
evaluation,
prevention
and
control."
37
Appendix
A
This
document
outlines
a
procedure
for
the
lab
checkout
of
the
inexpensive
noise
sensors
prior
to
deployment
in
the
low-‐cost
sensor
network
N/A
• NTI
XL2
Sound
Level
Meter
and
M4160
microphone
(NTi
Audio,
Tigard,
Oregon,
USA)
o AC
power
supply
cord
o mini-‐USB
cable
o NTi
Test
Signal
CD
• Sound
Level
Meter
Calibrator
o SCAL1356,
General
Tools
&
Instruments,
Secaucus,
New
Jersey,
USA
• Fender
amplifier
o 2311000000,
Fender
Musical
Instruments
Corporation,
Scottsdale,
AZ,
USA
• Laptop
computer
o Auxillary
cord
• NTi
Audio
Test
Signals
for
Audio
and
Acoustic
Analyzers
V1.0
• Cardboard
footprint
for
setup
• Wooden
block
to
elevate
SLM
4.
References
N/A
5. Definitions
38
Q:! 1?KV8J>?8!
9: 1?8C5?8!DB8!8_><C]8ID!MK?!D8HD<I=
OROR K6;@486=<!=><!$)G!
OROROR!-789!=><!$)G!E9!45!B<C8<??@9I!=><!+9q+DD!47==E9S!6?!@9B@H6=<B!@9!1@I78<!O!
!
1@I78<!OQ!*7==E9!D79H=@E9?!ED!$)G!
!
ORORMR!0?<!=><!8E=685!L><<;!=E!>@I>;@I>=!=><!A6@9!A<97!6=!=><!=EC!ED!=><!?H8<<9S!
69B!=><9!C8<??!#'-#,R!$H8E;;!=E!>@I>;@I>=!K6;@486=@E9S!=><9!C8<??!#'-#,!
6I6@9!=E!EC<9!=><!H6;@486=@E9!?H8<<9!!
!
1@I78<!MQ!+C<9@9I!=><!H6;@486=@E9!?H8<<9!
!
ORORbR!G6Y<!?78<!=>6=!=><!?H8<<9!?65?S!r0?<8!K6;@486=@E9!6=!OO^RN!B*Rs!0?<!=><!
8E=685!L><<;!=E!>@I>;@I>=!,0'S!47=!BE!9E=!C8<??!#'-#,!5<=R!
b` !
!
1@I78<!bQ!K><HY!H6;@486=@E9!?<==@9I?!69B!>@I>;@I>=!,0'!
!
OROR^R!-789!=><!$)G!K6;@486=E8!E9!=E!=><!OO^RN!B*!?<==@9I!69B!D@=!=><!EC<9@9I!ED!
=><!H6;@486=E8!E9=E!=><!$)G!A@H8EC>E9<R!."4)-`!%&-)\!*K!IKD!C>D!DB8!
V57<6?5DK?!>C!DK!LK>?!85?!
!
1@I78<!aQ!3EL<8!E9!$)G!K6;@486=E8!69B!C;6H<!E9=E!$)G!A@H8EC>E9<!
ORORaR!38<??!#9=<8!=E!4<I@9!8799@9I!H6;@486=@E9R!-><!?H8<<9!L@;;!@9B@H6=<!
rK6;@486=@E9!8799@9Is!69B!=><9!r$7HH<??D7;;5!D@9@?><B€s!L><9!HEAC;<=<R!
!
1@I78<!jQ!K6;@486=@E9!8799@9I!69B!HEAC;<=<!?H8<<9?!
^N !
!
ORORjR!0?<!=><!8E=685!L><<;!=E!>@I>;@I>=!=><!A6@9!A<97!6I6@9S!=><9!>@I>;@I>=!=><!
$)G<=<8q,-(!EC=@E9!69B!C8<??!#9=<8!=E!8<=789!=E!=><!>EA<!?H8<<9!
ORMR $<=7C!DE8!=><!)64!K><HYE7=!
ORMROR!)65!=><!H68B4E68B!4;7<C8@9=!E9!=><!=64;<!69B!D@;;!@9!=><!?CE=?!L@=>!=><!
HE88<?CE9B@9I!<F7@CA<9=!
!
1@I78<!PQ!$<=7C!ED!<F7@CA<9=!DE8!;64!H><HYE7=!C8EH<??!
!
ORMRMR!KE99<H=!=><!6AC;@D@<8!=E!=><!HEAC7=<8!L@=>!=><!67V@;@685!H64;<!
!
1@I78<!PQ!)EH6=@E9?!=E!HE99<H=!(02!H64;<!E9!6AC;@D@<8!69B!;6C=EC!
ORbR KE99<H=!=><!$)G!=E!=><!3,&!C8EI86A!!
^O !
!
ORbROR!+C<9!=><!AE?=!8<H<9=!J<8?@E9!ED!$<8@6;E4\<H=R<V<!E9!=><!;6C=EC!
ORbRMR!K>69I<!=><!HE9D@I786=@E9?!E9!=><!3,&!C8EI86A!
ORbRMROR #J<9=!'6A<Q!'E@?<K><HY[5555AABB!
ORbRMRMR ,<CE8=!(J<86I@9I!-@A<Q!M!?<H!
ORbRMRbR "6=6!36=>Q!KQÅ0?<8?Å)64[6BA@9ÅB<?Y=ECÅ)6786Å3,&!+7=C7=?!
ORbRMR^R $=6=@E9685!GEB<!
!
1@I78<!_Q!)65E7=!69B!HE9D@I786=@E9!ED!3,&!C8EI86A!DE8!;64!H><HYE7=!
!
ORbRbR!/<8@D5!=><!K+G!CE8=!6??@I9<B!=E!=><!$)GR!)<D=!H;@HY!69B!?<;<H=!r-<?=s!=E!
<9?78<!=><!B<J@H<!@?!HEAA79@H6=@9I!L@=>!=><!C8EI86AR!
!
1@I78<!`Q!K><HY!=><!K+G!CE8=!6??@I9<B!=E!=><!$)G!69B!=<?=!=>6=!@=!@?!HEAA79@H6=@9I!
ORbR^R!38<??!r$=68=!K6C=78<s!=E!4<I@9!8<HE8B@9I!A<6?78<A<9=?!
ORbR^ROR #9=<8!5E78!;EH6=@E9!6?!%,#:!
!
OR^R +C<9!=><!AE9@=E8!48EL?<8!DE8!=><!;EL[HE?=!AE9@=E8!=>6=!5E7!68<!<J6;76=@9I!!
OR^ROR !KE99<H=!=E!=><!r(@8F76;@=5s!L@8<;<??!9<=LE8Y!
^M !
!
OR^RMR!+C<9!K>8EA<!69B!<9=<8!=><!%3!6BB8<??!DE8!=><!AE9@=E8!6?!ONRO^MR^^ROVVS!
D@;;@9I!@9!=><!;6?=!=LE!B@I@=?!L@=>!=><!97A4<8!%"!DE8!=><!AE9@=E8!5E7!68<!
=<?=@9I!
OR^RbR!38<??!?=68=!H6C=78<!=E!4<I@9!8<HE8B@9I!MRa!?<HE9B!B6=6!B@8<H=;5!D8EA!=><!
AE9@=E8!
!
;: (56!#B8V^K>D!KM!%K<H8!.8IHK?!
MROR 37=!=><!'-@!-<?=!$@I96;!K"!@9=E!=><!HEAC7=<8R!$<;<H=!r3;65!67B@E!K"!7?@9I!
X@9BEL?!G<B@6!3;65<8s!D8EA!=><!67=EA6=@H!CEC[7C!L@9BELR!!
MROROR K>EE?<!-86HY!_!E9!=><!G<B@6!3;65<8[!3@9Y!'E@?<!cH8<?=!D6H=E8S!3(,‚k!bRa!
cONR`!B*dd!
!
1@I78<!ONQ!-<?=!K"!69B!CEC[7C!L@9BEL!L><9!5E7!C7=!K"!@9=E!HEAC7=<8!
!
MRMR -789!=><!6AC;@D@<8!E9!69B!A6Y<!?78<!=>6=!=><!=<?=!?@I96;!@?!C;65@9IS!=><9!C8<??!
C67?<!
MRbR ,<HE8B!=><!=@A<!69B!?=68=!5E78!D@8?=!bN[?<H!=<?=!C<8@EBR!->@?!@?!5E78!r(A4@<9=s!
=<?=!;<J<;R!$<<!1@I78<!OO!DE8!<V6AC;<!=64;<!DE8!8<HE8B@9I!?=68=!69B!?=EC!=@A<?R!
!
1@I78<!OOQ!#V6AC;<!8<HE8B!ED!?=68=!69B!?=EC!=@A<?!DE8!GE9@=E8!a!
!
MR^R +9H<!=><!D@8?=!=<?=!C<8@EB!>6?!HE9H;7B<BS!79[C67?<!=><!=<?=!?@I96;!
MRaR 0?<!=><!JE;7A<!HE9=8E;?!E9!=><!HEAC7=<8!=6?Y!468!=E!6B\7?=!=><!JE;7A<!79=@;!
=><!$)G!8<6B?!6CC8EV@A6=<;5!ja!B*(!
MRjR ,<HE8B!=><!=@A<!69B!?=68=!=><!bN[?<H!=<?=!C<8@EBR!->@?!@?!5E78!rja!B*(s!=<?=!;<J<;R!
^b !
!
MRPR ,<C<6=!$=<C?!MRa[!MRj!DE8!=<?=!;<J<;?!6=!Pa!B*(!69B!_a!B*(R!.E7!A65!9<<B!=E!
8<?=68=!=><!=<?=!?@I96;!=86HYS!6?!=><!B786=@E9!@?!bQNN!A@9!69B!5E78!=<?=@9I!A65!
=6Y<!;E9I<8R!!
MR_R -789!=><!$)G!K6;@486=E8!E9!=E!=><!`^!B*!?<==@9I!69B!@9?=6;;!=><!H6;@486=E8!E9=E!
=><!AE9@=E8!A@H8EC>E9<R!c1@I78<!OMd!
!
1@I78<!OMQ!$)G!K6;@486=E8!6DD@V<B!=E!=><!AE9@=E8!A@H8EC>E9<!
!
MR`R ,<HE8B!=><!=@A<!69B!?=68=!=><!bN[?<H!=<?=!C<8@EBR!->@?!@?!5E78!r`^!B*(s!=<?=!;<J<;R!
MRONR!,<AEJ<!=><!$)G!H6;@486=E8!D8EA!=><!AE9@=E8!A@H8EC>E9<!
MROOR!+C<9!=><!AE9@=E8!48EL?<8!69B!C8<??!?=EC!
MROOROR KEC5!6;;!E7=C7=!J6;7<?!69B!C6?=<!@9=E!6!9E=<C6B!D@;<!
MROORMR $6J<!D@;<!6?!r'E@?<K><HYE7=T....GG""TGE9@=E822s!
MROMR!$<;<H=!=><!9<V=!AE9@=E8!DE8!H><HYE7=!69B!8<C<6=!$=<C?!OR^[MROO!
MRObR!+9H<!5E7!>6J<!=<?=<B!6;;!=><!AE9@=E8?S!C8<??!r$=EC!K6C=78<s!E9!=><!3,&!
C8EI86AR!-><!D@;<!L@;;!4<!?6J<B!@9!=><!DE;B<8!@9B@H6=<B!@9!r"6=6!36=>s!
MRO^R!$6J<!6;;!5E78!@9B@J@B76;!AE9@=E8!D@;<?!69B!=><!$)G!D@;<!=E!6!D;6?>!B8@J<!DE8!;6=<8!
C8EH<??@9I!
!
! !
^^ !
!
Appendix
B
Appendix
B
includes
the
SAS
codes
and
output
for
the
t-‐tests,
mean
bias
by
test
level,
and
the
boxplot
of
differences
by
test
level.
Output
45
Mean
Bias
by
Test
Level
Code
Data noise;
set noise;
bias= (monitor-reference)/reference;
run;
46
Output
47
title 'Box Plot for Noise Levels';
proc boxplot data=noise3;
plot difference*level;
run;
Output
48